Persuasive Technology 11th International Conference PERSUASIVE 2016 Salzburg Austria April 5-7-2016 Proceedings

Alexander Meschtscherjakov
Boris De Ruyter
Verena Fuchsberger
Martin Murer
Manfred Tscheligi (Eds.)
LNCS 9638
Persuasive
Technology
11th International Conference, PERSUASIVE 2016
Salzburg, Austria, April 5–7, 2016
Proceedings
123
Lecture Notes in Computer Science 9638
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen
Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zurich, Zürich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology, Madras, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7409
Alexander Meschtscherjakov
Boris De Ruyter Verena Fuchsberger
•
Martin Murer Manfred Tscheligi (Eds.)

•
Persuasive
Technology
11th International Conference, PERSUASIVE 2016
Salzburg, Austria, April 5–7, 2016
Proceedings
123
Editors
Alexander Meschtscherjakov Martin Murer
University of Salzburg University of Salzburg
Salzburg Salzburg
Austria Austria
Boris De Ruyter Manfred Tscheligi
Philips Research Europe University of Salzburg
Eindhoven Salzburg
The Netherlands Austria
Verena Fuchsberger and
University of Salzburg
Salzburg Austrian Institute of Technology
Austria Vienna
Austria
ISSN 0302-9743 ISSN 1611-3349 (electronic)

Lecture Notes in Computer Science
ISBN 978-3-319-31509-6 ISBN 978-3-319-31510-2 (eBook)
DOI 10.1007/978-3-319-31510-2
Library of Congress Control Number: 2016933467
LNCS Sublibrary: SL3 – Information Systems and Applications, incl. Internet/Web, and HCI
© Springer International Publishing Switzerland 2016

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are
believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors
give a warranty, express or implied, with respect to the material contained herein or for any errors or
omissions that may have been made.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG Switzerland
Preface
Persuasive Technology is a vibrant and highly interdisciplinary research field that

focuses on the design, development, and evaluation of interactive technologies with the
aim of changing users’ attitudes and behaviors. Attitude and behavior change is
achieved by means of persuasive strategies, such as social influences embodied in the
design of interactive technologies, without any coercion or deception. Persuasive
technologies are used to change people’s behavior in various domains such as
healthcare, sustainability, education, or marketing.
PERSUASIVE, the International Conference on Persuasive Technology, is the
leading venue for ground-breaking research and novel designs of persuasive
technologies. It is the annual conference in which to discuss the latest persuasive
theories, strategies, applications, and artifacts with academics and practitioners from all
over the world. Over the past decade the conference was held at exciting places such as
Chicago, Padua, Sydney, Linköping, Columbus, Copenhagen, Claremont, Oulu, Palo
Alto, and Eindhoven.
PERSUASIVE 2016 was the 11th edition of the conference and took place in April
2016 in Salzburg, Austria. The conference theme was “Contextual Persuasion: Sup-
porting Life Situations and Challenges by Persuasive Design.” With this conference
theme, the ubiquity and situatedness of persuasive interactions was emphasized: How
are interactions with persuasive technologies influenced and facilitated by spatial,
temporal, social, or individual conditions and characteristics? How can we analyze,
design, and evaluate for specific contexts or conditions?
On April 4, a Doctoral Consortium, a tutorial on “Mobile Persuasion Design,” and a
Persuasive Game Jam were held as part of the pre-conference program. On April 5,
seven half-day workshops were held. On April 6 and 7, the main conference was held
in seven single-track sessions, including oral presentations of accepted short and long
papers, as well as a poster and demo session during which accepted work-in-progress
and demonstrations were presented. It also included an opening keynote by Mark
Aloia, Global Lead for Behavior Change at Philips HealthTech, and a closing keynote
by Cees Midden, Professor of Human–Technology Interaction at Eindhoven University
of Technology.
This volume contains the accepted short and long papers presented during the main
track. Overall, 73 papers were submitted (59 long papers with a maximum length of 12
pages and 14 short papers with a maximum length of 6 pages) with 197 authors from
more than 20 countries from Asia, Australia, Europe, North and South America.
Papers were selected for presentation at the conference after a thorough peer-review
process. The submitted papers were reviewed by experts in the field of persuasive
technologies in a double-blinded review process. Overall, 63 reviewers were randomly
(excluding any conflict of interest) assigned to the papers. They provided a detailed
textual review of the assigned paper and rated each paper, leading to a ranking of the
VI Preface
papers. The Program Committee chairs examined the papers and their reviews and
compiled the final list of papers to be presented at the conference.
From the 73 submitted papers, 30 were accepted, yielding an acceptance rate of 41.1 %.
From the 59 long papers submitted, 27 were accepted (i.e., acceptance rate of 45.8 %).
From the 14 submitted, three were accepted (i.e., acceptance rate of 21.4 %).
In addition to the papers presented in this volume, the conference also published
adjunct proceedings, which included the accepted work-in-progress submissions to the
posters track, the accepted demo submissions to the demos track, the accepted position
papers to the doctoral consortium, as well as a description of the seven workshops:
• User Experience Design for Persuasion and Behavior Change
• Empowering Cities for Sustainable Well-Being
• The Challenge of Device Overload: Using the Persuasive Framework to Effectively
Use Modern Technologies to Encourage Health-Promoting Behaviors
• Where Are We Bound for? Persuasion in Transport Applications
• Persuasive Designs for Learning – Learning in Persuasive Design
• Behavior Change Support Systems (BCSS 2016): Epic for Change, the Pillars for
Persuasive Technology for Smart Societies
• Personalization in Persuasive Technology Workshop
To make this conference a success, a great number of people supported in various
ways. We would like to thank the authors for their high-quality contributions, and the
reviewers for their valuable feedback. Furthermore, we would like to express our
appreciation to the organizational and scientific committees, who took care of the
workshops, tutorials, doctoral consortium, posters, demos and showcases, the game
jam, and the main conference.
April 2016 Alexander Meschtscherjakov

Boris De Ruyter
Verena Fuchsberger
Martin Murer
Manfred Tscheligi
Organization
General Chair
Manfred Tscheligi University of Salzburg and Austrian Institute
of Technology, Austria
Program Chairs
Boris De Ruyter Philips Research, The Netherlands
Alexander University of Salzburg, Austria
Meschtscherjakov
Organizational Chairs
Verena Fuchsberger University of Salzburg, Austria
Martin Murer University of Salzburg, Austria
Meschtscherjakov
Poster Chairs
Alexandra Millonig Austrian Institute of Technology, Austria
Rita Orji McGill University, Canada
Demo and Showcase Chairs

Marc Busch Austrian Institute of Technology, Austria
Margaret Morris Intel, USA
Doctoral Consortium Chairs

Jaap Ham Eindhoven University of Technology, The Netherlands
Cees Midden Eindhoven University of Technology, The Netherlands
Luciano Gamberini University of Padua, Italy
Workshop Chairs
Maurits Kaptein Tilburg University, The Netherlands
Peter Fröhlich Austrian Institute of Technology, Austria
VIII Organization
Persuasive [Game] Design Jam Chairs

Bernhard Maurer University of Salzburg, Austria
Agnis Stibe MIT Media Lab, USA
Tutorial Chair
Harri Oinas-Kukkonen University of Oulu, Finland
Social Media Chair

Local Arrangements
Carina Bachinger University of Salzburg, Austria
Kristina Karl University of Salzburg, Austria
Alexandra Leitner University of Salzburg, Austria
Scientific Committee
Magnus Bang Linkoping University, Sweden
Shlomo Berkovsky CSIRO, Australia
Robert Biddle Carleton University, Canada
Marc Busch Austrian Institute of Technology, Austria
Cheryl Campanella Bracken Cleveland State University, USA
Samir Chatterjee Claremont Graduate University, USA
Luca Chittaro HCI Lab, University of Udine, Italy
Janet Davis Whitman College, USA
Berardina De Carolis University of Bari, Italy
Boris De Ruyter Philips Research, The Netherlands
Peter De Vries University of Twente, The Netherlands
Sebastian Egger Austrian Institute of Technology, Austria
Alexander Felfernig Graz University of Technology, Austria
BJ Fogg Stanford University, USA
Jill Freyne CSIRO, Australia
Peter Fröhlich Austrian Institute of Technology, Austria
Verena Fuchsberger University of Salzburg, Austria
Luciano Gamberini University of Padua, Italy
Mark Gilzenrat CNN Digital, USA
Manuel Giuliani University of Salzburg, Austria
Thomas Grah University of Salzburg, Austria
Ulrike Gretzel University of Queensland, Australia
Marco Guerini FBK_Irst, Italy
Magdalena Gärtner University of Salzburg, Austria
Jaap Ham Eindhoven University of Technology, The Netherlands
Organization IX
Curtis Haugtvedt Ohio State University, USA

Stephen Intille Northeastern University, USA
Sriram Iyengar Texas A&M University Health Science Center, USA
Giulio Jacucci University of Helsinki, Finland
Anthony Jameson DFKI, Germany
Antti Jylhä University of Helsinki, Finland
Maurits Kaptein Tilburg University, The Netherlands
Sarvnaz Karimi CSIRO, Australia
Saskia Kelders University of Twente, The Netherlands
Alina Krischkowsky University of Salzburg, Austria
Sitwat Langrial SUC, Oman
Thomas MacTavish Illinois Institute of Technology, USA
Elke Mattheiss Austrian Institute of Technology, Austria
Bernhard Maurer University of Salzburg, Austria
Thomas Meneweger University of Salzburg, Austria
Meschtscherjakov
Alexandra Millonig Austrian Institute of Technology, Austria
Omar Mubin Western Sydney University, Australia
Sahiti Myneni University of Texas, USA
Katja Neureiter University of Salzburg, Austria
Rita Orji McGill University, Canada
Andreas Riener Ingolstadt University of Applied Sciences, Germany
Peter Ruijten Eindhoven University of Technology, The Netherlands
Privender Saini Philips, The Netherlands
Juan Salamanca Universidad Icesi, Colombia
Hanna Schraffenberger Leiden University, The Netherlands
Johann Schrammel Austrian Institute of Technology, Austria
Salvatore Sorce Università degli Studi di Palermo, Italy
Anna Spagnolli University of Padua, Italy
Kristian Tørning Danish School of Media and Journalism, Denmark
Julita Vassileva University of Saskatchewan, Canada
Charlotte Vinkers Philips Research, The Netherlands
Vance Wilson Worcester Polytechnic Institute, USA
Daniela Wurhofer University of Salzburg, Austria
Kyung-Hyan Yoo William Paterson University, USA
Johannes de Boer Saxion University of Applied Sciences,
The Netherlands
Arlette van Wissen Philips Research, The Netherlands
X Organization
Steering Committee
Magnus Bang Linkoping University, Sweden
Shlomo Berkovsky CSIRO, Australia
Samir Chatterjee Claremont Graduate University, USA
BJ Fogg Stanford University, USA
Peter Hasle Aalborg University, Denmark
Sponsors
Contents
Individual Differences
Tailoring Web Pages for Persuasion on Prevention Topics: Message

Framing, Color Priming, and Gender . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Luca Chittaro
Supporting Users in Setting Effective Goals in Activity Tracking . . . . . . . . . 15

Katja Herrmanny, Jürgen Ziegler, and Aysegül Dogangün
Persuasive and Culture-Aware Feedback Acquisition . . . . . . . . . . . . . . . . . . 27

Malik Almaliki and Raian Ali
Theoretical Reflections
Crowd-Designed Motivation: Combining Personality and the

Transtheoretical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Roelof A.J. de Vries, Khiet P. Truong, and Vanessa Evers
The EDIE Method – Towards an Approach to Collaboration-Based

Persuasive Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Sandra Burri Gram-Hansen
Persuasive Backfiring: When Behavior Change Interventions Trigger

Unintended Negative Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Agnis Stibe and Brian Cugelman
Captology and Technology Appropriation: Unintended Use as a Source for

Designing Persuasive Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Alina Krischkowsky, Bernhard Maurer, and Manfred Tscheligi
Prevention and Motivation
Self-Reflecting and Mindfulness: Cultivating Curiosity and Decentering

Situated in Everyday Life. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Ralph Vacca and Christopher Hoadley
Alcohol Behaviour Change: Lessons Learned from User Reviews

of iTunes Apps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Omar Mubin, Abdullah Al Mahmud, and Muhammad Ashad Kabir
XII Contents
Persuasive Strategies to Improve Driving Behaviour of Elderly Drivers by a

Feedback Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Perrine Ruer, Charles Gouin-Vallerand, and Evelyne F. Vallières
Creating Awareness of Sleep-Wake Hours by Gamification . . . . . . . . . . . . . 122

Ezgi Ilhan, Bahar Sener, and Hüseyin Hacihabiboğlu
Methods and Models
Cicero: Middleware for Developing Persuasive Mobile Applications . . . . . . . 137

Antonello D’Aloia, Matteo Lelli, Duckki Lee, Sumi Helal,
and Paolo Bellavista
Formalization of Computational Human Behavior Models for Contextual

Persuasive Technology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
Tylar Murray, Eric Hekler, Donna Spruijt-Metz, Daniel E. Rivera,
and Andrew Raij
The Persuasive Potential Questionnaire (PPQ): Challenges, Drawbacks,

and Lessons Learned . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Alexander Meschtscherjakov, Magdalena Gärtner, Alexander Mirnig,
Christina Rödel, and Manfred Tscheligi
Persuasive Practices: Learning from Home Security Advisory Services . . . . . 176

Mateusz Dolata, Tino Comes, Birgit Schenk, and Gerhard Schwabe
Persuasive Patterns in Q&A Social Networks . . . . . . . . . . . . . . . . . . . . . . . 189

Ifeoma Adaji and Julita Vassileva
Games and Gamification
Utilizing a Digital Game as a Mediatory Artifact for Social Persuasion to

Prevent Speeding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Bernhard Maurer, Magdalena Gärtner, Martin Wuchse,
Alexander Meschtscherjakov, and Manfred Tscheligi
Smile Catcher: Can Game Design Lead to Positive Social Interactions? . . . . . 211
Niaja Farve and Pattie Maes
More than Sex: The Role of Femininity and Masculinity in the Design
of Personalized Persuasive Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Marc Busch, Elke Mattheiss, Michaela Reisinger, Rita Orji,
Peter Fröhlich, and Manfred Tscheligi
A Gamified Solution to Brief Interventions for Nightlife Well-Being . . . . . . . 230

L. Gamberini, A. Spagnolli, M. Nucci, G. DeGiuli, C. Villa, V. Monarca,
A. Privitera, L. Zamboni, and S. Leclerq
Contents XIII
Long-Term Effects of Computerized Simulations in Protracted Conflicts:

The Case of Global Conflicts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
Ronit Kampf
Interventions for Behavior Change
Understanding Changes in the Motivation of Stroke Patients Undergoing

Rehabilitation in Hospital . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
Michelle Pickrell, Bert Bongers, and Elise van den Hoven
Developing a Virtual Coach for Chronic Patients: A User Study on the

Impact of Similarity, Familiarity and Realism . . . . . . . . . . . . . . . . . . . . . . . 263
Arlette van Wissen, Charlotte Vinkers, and Aart van Halteren
Improving Adherence in Automated e-Coaching: A Case from Insomnia

Therapy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
Robbert Jan Beun, Willem-Paul Brinkman, Siska Fitrianie,
Fiemke Griffioen-Both, Corine Horsch, Jaap Lancee, and Sandor Spruit
Online Peer Groups as a Persuasive Tool to Combat Digital Addiction . . . . . 288

Amen Alrobai, John McAlaney, Keith Phalp, and Raian Ali
Design Strategies and Techniques
Red Radiators Versus Red Tulips: The Influence of Context on the

Interpretation and Effectiveness of Color-Based Ambient Persuasive
Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Shengnan Lu, Jaap Ham, and Cees Midden
Investigating Politeness Strategies and Their Persuasiveness for a Robotic

Elderly Assistant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
Stephan Hammer, Birgit Lugrin, Sergey Bogomolov, Kathrin Janowski,
and Elisabeth André
RightOnTime: The Role of Timing and Unobtrusiveness in Behavior

Change Support Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
Piiastiina Tikka and Harri Oinas-Kukkonen
Persuasive Information Security: Techniques to Help Employees Protect

Organizational Information Security. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
Marc Busch, Sameer Patil, Georg Regal, Christina Hochleitner,
and Manfred Tscheligi
Lock Up the Lighter: Experience Prototyping of a Lively Reflective Design

for Smoking Habit Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
Kenny K.N. Chow
XIV Contents
Erratum to: Self-Reflecting and Mindfulness: Cultivating Curiosity

and Decentering Situated in Everyday Life. . . . . . . . . . . . . . . . . . . . . . . . . E1
Ralph Vacca and Christopher Hoadley
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365

Individual Differences
Tailoring Web Pages for Persuasion on Prevention
Topics: Message Framing, Color Priming, and Gender
Luca Chittaro ✉
( )
Human-Computer Interaction Lab, University of Udine, via delle Scienze 206, 33100 Udine, Italy
[email protected]
http://hcilab.uniud.it
Abstract. On the Web, as in more traditional influence contexts, the most effec‐
tive persuasive strategies often depend on the individual characteristics of the
message recipient. Unfortunately, most persuasive technology applications
currently employ a one-size-fits-all approach to interventions. The study we illus‐
trate investigates two different techniques (message framing and color priming)
that can be used in tailoring a persuasive Web page about a prevention topic. The
findings of our study highlight interactive effects between message framing and
color priming, and advance the results in the literature by showing that red
enhances the effects of framing in a gender-based fashion. The obtained results
also provide practical guidance for automatic tailoring of persuasive Web pages
about prevention topics, suggesting a strategy based on gender, an information
about the user that is typically readily available in social network profiles, and
other Web sites to which people register.
Keywords: Tailoring · Web pages · Color priming · Message framing · Gender

differences · Prevention · Health · Safety · Persuasive technology
1 Introduction
Individual characteristics of the message recipient often affect the effectiveness of

persuasive strategies on the Web as well as in other influence contexts. Unfortunately,
most persuasive technology applications currently employ a one-size-fits-all approach
to interventions, failing to deliver tailored persuasion that leverages user’s characteris‐
tics [1]. In particular, Web audiences are large and heterogeneous, and influencing them
with a one-size-fits-all intervention is difficult.
Software systems that are able to tailor content and presentation of Web pages based
on a user’s profile exist–see [2] for a review – and could be used for tailoring Web-based
persuasive interventions. However, these tailoring systems need to be programmed with
proper rules to deliver the most effective content and presentation based on user’s char‐
acteristics. In persuasive interventions, the definition of such rules should be guided by
theoretical frameworks of persuasion, and needs to be supported by studies of how
different people are affected by different versions of a Web page.

A. Meschtscherjakov et al. (Eds.): PERSUASIVE 2016, LNCS 9638, pp. 3–14, 2016.
DOI: 10.1007/978-3-319-31510-2_1
4 L. Chittaro
Two important features of Web pages that could affect their effectiveness are the
framing of the message contained in the page (for example, a message about health can
be framed in terms of gain or loss associated with following or not the recommended
behavior [28]), and the colors used by the page (for example, different background colors
of a Web page can prime different attributes related to its content, influencing viewers
[22]). The study in this paper focuses on both persuasion techniques (message framing
and color priming), exploring how they could be exploited in tailoring a persuasive Web
page. More specifically, the purpose of the paper is to (i) investigate message framing
and color priming in Web pages that deal with prevention topics, (ii) study possible
gender differences in users’ susceptibility to the considered techniques, which could be
very important for automatically tailoring Web pages based on gender as an individual
user’s characteristic, (iii) derive possible design guidance for tailoring Web interven‐
tions on prevention topics.
2 Related Work
A persuasive message can be framed in terms of the benefit (gain) or cost (loss) asso‐
ciated with adopting or not its recommendation. Several studies have shown that loss
and gain frames affect people’s attitudes and behavior in a different way even when they
describe objectively equivalent situations, see [19, 27, 28] for reviews. In particular, the
effects of framing can change with the addressed type of behavior, with gain framing
being more effective when the message is a prevention recommendation (e.g., sun screen
use, physical exercise, vaccinations,…), and loss framing more effective when it is a
detection recommendation (e.g., checkups, breast self-exam, HIV testing,…).
However, most studies did not consider gender as a possible individual difference
that may affect the influence of loss or gain framing on message recipients, see [19]. The
few studies which considered gender in health communication (e.g., about sunscreen
[27] and condoms [18] to prevent health risks) or in other domains (e.g., filling honestly
the tax return to prevent risks such as fines, penalties and jail sentences [16]) suggest
that gain framing could be more effective with women, while men could be more sensi‐
tive to loss framing. It is also worth noting that Fagley and Miller [10, 11], who studied
gender differences in risky decision problems, go as far as saying that framing experi‐
ments which do not address possible gender effects may simply reflect the gender that
predominates in the sample and be uninterpretable.
Another important feature of Web pages that can affect users is color. Research on
color priming has investigated effects of color on different aspects of cognition and
behavior, e.g. [7–9, 14, 23]. A study about Web persuasion was proposed by [22], who
considered simple Web pages containing a product description and showed how back‐
ground color can influence consumers. Participants who examined a car description on
a Web page with a red and orange flame-like background were later more likely to
mention safety as an important attribute for buying a car than participants who were
exposed to a different (green) background, showing that red can help in priming safety.
One of the color priming studies conducted by [23] focused specifically on prevention
aspects. Participants read descriptions of three pairs of brands on a computer screen and
Tailoring Web Pages for Persuasion on Prevention Topics 5
then reported their brand preferences. Within each pair, one brand highlighted a negative
outcome people try to avoid, whereas the other brand highlighted a positive outcome
people try to approach. For example, one of the pairs concerned toothpaste, with brand
A particularly good for cavity prevention and brand B particularly good for tooth
whitening. Across the three pairs, a red background increased preference for the brand
that emphasized prevention. After ruling out mood as an alternative explanation with a
post-hoc study, the study concluded that a red background contributes to activating
avoidance motivation.
Gerend and Sias [13] investigated color priming together with message framing in
persuasion about preventive behavior. In particular, the study concerned printed mate‐
rials (a red or grey leaflet recommending a vaccination), in which the message was
presented as either loss-framed or gain-framed. The results showed an interaction
between color priming and message framing: loss framing was more effective but only
when primed with red. Participants in this study were all male (a limitation that is
acknowledged by its authors), and this could explain why the sample was more sensitive
to the loss frame. The fact that red enhanced the effect of loss framing is consistent with
the increase in risk avoidance motivation pointed out by [23] and with the role of red as
a safety prime found by [22]. Moreover, Gerend and Sias [13] suggest that the enhancing
interactive effect of color red could originate from the fact that red primes threat via
associations with blood and danger, and acts as a peripheral threat cue that affects
processing of persuasive health messages. More generally, other authors point out that
red, as the single color most commonly associated with danger [34], has been shown to
make people more vigilant [23], and the amount of attention directed to the message is
an important aspect for the success of a persuasive intervention. Both perspectives would
be consistent with the activation of a recently postulated “human alarm system” [6], a
psychological system that people use to detect and handle threatening cues and, when
activated, prompts people to process more alertly what is going on (for a discussion of
the human alarm system, and some of its effects on judgment see [32]).
3 Method
The goal of our study was to advance the investigation of message framing and color
priming in Web pages for prevention topics, aiming at deriving possible guidance for
tailored persuasion based on gender.
We considered fires as a prevention domain of interest for their social relevance, e.g.
in 2014, in the US alone, 19’050 civilians were hurt (3,275 dead, 15,775 injured) as the
result of fires [17], with home fires accounting for 84 % of the death toll, and a fire
occurring every 86 s in the country. Adjusting for population, the fire death rate of some
former USSR countries such as Russia, Latvia, and Estonia, is about ten times higher
than the fire death rate of the US [5]. Fire risks are thus an important worldwide safety
issue, and also a public health issue because they lead to hospitalizations of survivors,
who can suffer long-term or even lifelong health consequences.
Current approaches to foster awareness of personal fire safety on the Web have
explored the creation of different pages based on a two-groups age segmentation,
6 L. Chittaro
producing a version of the Web site devoted to children, e.g. [24, 30], and one to adults,
e.g. [25, 31]. Web pages targeted at adults currently follow a one-size fits all approach
that uses the same messages for all visitors.
The study in this paper explores if tailoring message framing and color in this kind
of Web pages could increase their effectiveness. Among the topics dealt with by fire
prevention campaigns, the described experiment focuses on domestic smoke alarms, a
preventive measure that significantly increases occupants’ chances of surviving a deadly
home fire [25, 31].
3.1 Design, Participants, and Hypotheses

We followed a between-groups design in which participants read a Web page about fire
prevention that stressed the importance of having smoke alarms in houses. The textual
message of the Web page was either gain- or loss-framed. More specifically, the message
in the loss-framed (respectively, gain-framed) version of the Web page pertained how
many people die (could be saved) every year if fires were prevented, the negative effects
of noticing late a fire in the house (the positive effects of noticing early a fire in the
house), the negative effects of the lack of smoke alarms in the house (the positive effects
of the presence of smoke alarms in the house). Table 1 compares the two versions of the
message, highlighting the parts that differ.
The study was conducted on a sample of 126 (65 male, 61 female) non-colorblind
participants, recruited at the university library or through personal contact. They were
volunteer undergraduate, graduate and doctoral students enrolled in different programs
(Agricultural Science, Business Administration, Computer Science, Foreign Languages,
Engineering, Literature,…), and people from other occupations who received no
compensation. Age ranged from 19 to 38 (M = 25.3, SD = 4.0).
Participants were told that the purpose of the study was to evaluate a Web page. They
first filled a short demographic questionnaire. Then, they read the Web page assigned
to them on a computer screen. No interaction with the computer keyboard or mouse was
required to read the page. Color priming was manipulated by making the background
and the two uppercase section titles in the Web page either red or grey, equated on value
(the relative lightness versus darkness of a color). As a result, four versions of the Web
page were created (gain-framed with red or grey, loss-framed with red or grey), and each
participant read one of the four pages. In all versions of the Web page, the text paragraphs
were contained in a rectangular white area and were displayed in black lettering over
the white area to maximize readability. The rectangular white area was placed over the
red or grey background that filled the page, and its size and position did not change. The
only factors that changed were message frame (gain, loss) and color prime (red, grey)
as already described above. The Web page did not contain any other text (e.g., menus,
links, copyright,…). After examining it, participants were asked about their:
Table 1. Gain- and loss-framed message. For reader’s convenience, the section titles and the text
paragraphs in the message are an English translation of the original (non-English) ones, and we
have highlighted differences between the gain- and loss-framed versions in italic.
Gain-Framed Version Loss-Framed Version

HOME FIRES HOME FIRES
Fire has always been a leading cause of Fire has always been a leading cause of
accidental death. Just think that if there were accidental death. Just think that because of
no fires in buildings, more than 5000 people a fires in buildings, more than 5000 people a
year would be saved in Europe alone. year die in Europe alone.
Homes are particularly exposed to the Homes are particularly exposed to the
possibility of fire, due to a relevant possibility of fire, due to a relevant
concentration of electrical devices, sources of concentration of electrical devices, sources of
heat and flammable materials. heat and flammable materials.
Several studies show that in common home Several studies show that in common home
environments, detecting early the presence of environments, detecting late the presence of a
a fire can help prevent reaching conditions of fire can put at risk of reaching conditions of
non-sustainability (that is, the death of non-sustainability (that is, the death of
people). people).
Fortunately, many families saved their lives Unfortunately, many families lost their lives
because they did install an important safety because they did not install an important
feature in their homes: smoke alarms. safety feature in their homes: smoke alarms.
With smoke alarms, you are warned that Without smoke alarms, you are not warned
smoke is entering your bedroom while you that smoke is entering your bedroom while
are sleeping: in this way, you wake up in you are sleeping: in this way, you wake up too
time, when it is still possible to escape from late, when it is not anymore possible to escape
the building. from the building.
The causes of a fire can be very common and The causes of a fire can be very common and
trivial: a short circuit in an electrical trivial: a short circuit in an electrical
appliance, a cigarette left burning on an appliance, a cigarette left burning on an
ashtray, a cloth over a lit lamp,... The ashtray, a cloth over a lit lamp,... The absence
presence of a smoke alarm allows us to detect of a smoke alarm does not allow us to detect
these events early and frees us from a deadly these events early and imprisons us in a
trap. deadly trap.
THE SOLUTION IS SIMPLE THE SOLUTION IS SIMPLE

Fire alarms are cheap and easy to install. It is Fire alarms are cheap and easy to install. It is
advisable to install a smoke alarm outside advisable to install a smoke alarm outside
each bedroom or sleeping area of the house. If each bedroom or sleeping area of the house. If
you live in a multi-story home, you need to you live in a multi-story home, you need to
install a smoke alarm on each floor. install a smoke alarm on each floor.
• Perceived level of attention and alertness to the Web page. We were interested in
assessing this aspect because, in general, when a persuasive attempt employs textual
messages, it is fundamental that the user attentively reads the information contained
in them. Moreover, in this study, it is important to assess if and when the threatening
cue (color red) is actually able to increase the level of attention and alertness as the
color priming literature would suggest it is able to.
8 L. Chittaro
• Attitudes towards the specific prevention solution (smoke alarms) recommended by

the page. We particularly focused on beliefs concerning the effectiveness of the
recommended solution (response efficacy), because a broad array of theoretical
perspectives – see [12, 26, 28, 33] for thorough discussions - suggests that perceived
response efficacy is an important predictor of the likelihood of the recommended
behavior being carried out by the message recipient.
Based on the previously surveyed literature on color priming, which has shown the
general role of red in increasing vigilance and priming safety, and the specific study [13]
which has shown (on a male sample) that the enhancing effect of red on persuasion
shows itself with the loss frame (to which men are likely more sensitive), we hypothe‐
sized that we were going to obtain a similar result with men in our study. However, since
our study extends the investigation to women, who are likely more sensitive to the gain
frame, we hypothesized that we were going to obtain the best results in terms of attention
and persuasion with women through the gain frame, enhanced by the general role of red
mentioned above.
3.2 Measures
To measure the level of attention and alertness, we employed the items used in [32].
We asked participants whether the Web page led them to be alert (1 = very weakly,
7 = very strongly) and to be attentive (1 = very weakly, 7 = very strongly), and their
answers to the two items were averaged to form a reliable scale of participants’
alertness (Cronbach’s alpha = .78).
To measure participants’ attitudes towards the recommended prevention solution,
we used three questions that respectively asked how useful, important, and effective is
to have smoke alarms in houses. Answers were given on a 7-point Likert scale (1 = not
at all, 7 = a lot), and were averaged to form a reliable scale of perceived response
efficacy (Cronbach’s alpha = .92).
4 Results
We conducted a three-way ANOVA with framing (loss, gain), color priming (red, gray),
and gender (male, female) as factors. The results of the analysis yielded:
• for attention, a main effect of gender, F(1,118) = 10.17, p < .01, ηp2 = .079, and an
interaction among all three factors, F(1,118) = 10.53, p < .01, ηp2 = .082;
• for response efficacy, a main effect of gender, F(1,118) = 5.24, p = .024, ηp2 = .042,
and an interaction among all three factors, F(1,118) = 5.94, p = .016, ηp2 = .048.
Since a triple interaction among the independent variables characterized attention as well
as perceived response efficacy, we followed a standard unfolding procedure to determine
its meaning, by computing separately subsidiary two-way ANOVAs. Unfolding of the
triple interaction for attention (see Fig. 1) showed that, in men, color red generated more
attention with loss (M = 4.97, SD = 1.66) rather than gain framing (M = 3.35, SD = 1.14),
F(1,28) = 9.08, p < .01, ηp2 = .25, and the gain frame generated less attention when
coupled with red (M = 3.35, SD = 1.14) rather than grey (M = 4.22, SD = 1.15),
F(1,29) = 4.39, p = .045, ηp2 = .13. On the contrary, in women, it was the gain frame that
generated more attention when coupled with red (M = 5.33, SD = 1.05) rather than grey
(M = 4.37, SD = 1.33), F(1,28) = 4.90, p = .035, ηp2 = .15. Finally, the difference in
attention between women and men was statistically significant only for the gain frame
enhanced with red, with higher attention elicited in women (M = 5.33, SD = 1.05) rather
than men (M = 3.35, SD = 1.14), F(1,26) = 23.95, p < .001, ηp2 = .47.
Fig. 1. Triple interaction among gender, color priming, and message framing on attention elicited
by the Web page. Capped vertical bars denote ± 1 SE (Color figure online).
Unfolding of the triple interaction for response efficacy (see Fig. 2) revealed no
statistically significant differences in men, while in women the gain frame produced
10 L. Chittaro
Fig. 2. Triple interaction among gender, color priming, and message framing on perceived
response efficacy of smoke alarms elicited by the Web page. Capped vertical bars ± 1 SE (Color
figure online).
higher perceived response efficacy when coupled with red (M = 6.58, SD = 0.71) rather
than grey (M = 5.62, SD = 1.01), F(1,28) = 9.05, p < .01, ηp2 = .24.
The difference in perceived response efficacy between women and men was statis‐
tically significant only in the gain frame with red, which elicited stronger perception of
response efficacy in women (M = 6.58, SD = 0.71) rather than men (M = 5.10,
SD = 1.73), F(1,26) = 9.20, p < .01, ηp2 = .26.
5 Discussion
The results for attention and response efficacy confirmed our expectation that the effects
on women and men were going to differ, and better results were in general going to be
obtained by employing a gain frame strategy with women and a loss frame strategy with
men, enhancing it with red color priming in both cases.
The finding that, for male participants, loss framing was more effective in eliciting
attention but only when primed with red is consistent with and reinforces the results of
the previously mentioned study by Gerend and Sias [13] of a male-only sample that read
printed materials about a vaccination. In particular, the pattern suggested by the charts
concerning men in Fig. 1 (and also in Fig. 2) is remarkably similar to the one found in
that study when they measured the effectiveness of their persuasive message. However,
to the best of our knowledge, our study is the first to extend the investigation of the
interactive effects of message framing and color priming to women, showing that a
different pattern characterizes women, with gain framing (instead of loss framing)
primed with red as the possible strategy of choice.
Our results contribute evidence to what has been suggested (but not extensively
studied in different conditions and domains) by some authors in the literature (see
Sect. 2), i.e. that in communication about prevention topics a gain framing strategy
could be more effective with women, while men could be more sensitive to loss
framing. However, our study suggests that for this difference to become significant,
the persuasive intervention may have to enhance the framing of the textual message
with further stimuli. In our case, the additional stimulus was a peripheral threat cue
(red color in the page). An interesting consideration concerns how a subtle cue (such
as page color) peripheral to the information being processed can actually have
considerable effects. Previous research on Web pages had shown that different
choices of background color can change the perception of Web page loading time
[15], can influence final scores on general knowledge tests conducted through Web
pages [14] or influence users into considering some features of a product descrip‐
tion more important than others [22]. In our study, color priming was instead effec‐
tive in amplifying the effects of a persuasive framing strategy in terms of attention
obtained by the Web page and elicited perception of efficacy of the recommenda‐
tion made by the Web page. Taken together, these different studies highlight how the
design of persuasive Web pages should be very careful and grounded on persuasion
research, even for those peripheral page features such as background color that are
typically dealt with only from the point of view of usability (e.g., readability of
information) and aesthetics (e.g., pleasant and coherent color schemes).
It is worth remembering that color priming is context-dependent: while red activates
avoidance motivation [23] when associated with safety topics, it can have different
effects in other contexts, e.g. in interpersonal relationships it can enhance male (physical
and sexual) attraction towards a woman [9] instead of avoidance motivation. One should
thus be careful in trying to generalize the results of our experiment outside the context
of messages about risk prevention.
On a practical level, our findings indicate that gender-based tailoring of Web pages
in persuasive interventions concerning prevention topics can be effective. The versions
of the Web page we tested produced different results in participants, based on their
gender. For an automated system that has access to gender information, e.g. most appli‐
cations in social networks, it would be very easy to present the user with the version of
the Web page that maximizes the desired effects.
As a final consideration, a limitation of the study might involve cultural aspects: the
meaning attributed by people to colors can be culture-dependent (e.g., in some cultures
12 L. Chittaro
black is the color of mourning, while in others it is white) and our experiment involved
only a European sample. However, our finding on color concerned red, whose impact
is supposed to be universal [7].
6 Conclusions
This paper explored the effectiveness of a loss frame versus a gain frame strategy in
formulating a persuasive message for a Web page that included either a grey or red color
prime. We focused on possible gender differences that could provide design guidance
for rules that drive automated systems for Web page tailoring.
First, the results of the study support the idea of gender-based tailoring of Web pages
in persuasive attempts concerning prevention topics: the most effective conditions were
different between men and women.
Second, it confirms gender differences in the effectiveness of message framing
hypothesized in the literature [16, 18, 27]: gain framing was the best strategy for female
users, while loss framing could be more suited to male users.
Third, it confirms the interactive effects between message framing and color priming
obtained in the literature with men [13], but extends those results by showing that red
enhances the effects of framing in a gender-based fashion: it is better to associate red to
a gain frame for female users and to a loss frame for male users. This extended result
provides practical guidance for automatic tailoring of persuasive Web pages about
prevention topics.
A particular situation in which the suggested automatic tailoring could become less
effective concerns users who access social networks or Web sites with a fake profile in
which they misrepresent their gender. However, in the context of large-scale campaigns,
the benefits of maximizing effectiveness on the large number of people who build real
profiles on the social network will overshadow the suboptimal results that could be
obtained with the smaller number of those who fake their identity. Moreover, it must
also be noted that automatic technologies able to detect gender misrepresentation on the
Web are available, e.g. Twitter claims that its gender recognition technology is 90 %
accurate [29]. Gender recognition technologies based on multilingual automatic text
analysis [3] allow a social network to detect user’s gender also when the profile does
not include gender information (as in current Twitter profiles).
We now plan to carry out further studies to extend and refine rules for tailoring
persuasive Web pages based on simple information about the visitor that can be readily
available, such as gender. For example, recent research on health and safety campaigns,
e.g. [4, 20, 21], has suggested that men could be less sensitive than women to messages
that resort to high physical threats (such as death and injuries), while they might respond
better to humorous appeals, a difference that – if confirmed – could be important for
Web content tailoring.
References
1. Berkovsky, S., Freyne, N.J., Oinas-Kukkonen, H.: Influencing individually: fusing

personalization and persuasion. ACM Trans. Interact. Intell. Syst. 2, 2 (2012)
2. Bunt, A., Carenini, G., Conati, C.: Adaptive content presentation for the web. In: Brusilovsky,
P., Kobsa, A., Nejdl, W. (eds.) The Adaptive Web. LNCS, vol. 4321, pp. 409–432. Springer,
Heidelberg (2007)
3. Burger, J.D., Henderson, J., Kim, G., Zarrella, G.: Discriminating gender on twitter. In:
Proceedings of the Conference on Empirical Methods in Natural Language Processing
(EMNLP 2011), pp. 1301–1309. Stroudsburg, USA: Association for Computational
Linguistics (2011)
4. Conway, M., Dubé, L.: Humor in persuasion on threatening topics: Effectiveness is a function
of audience sex role orientation. Pers. Soc. Psychol. Bull. 28, 863–873 (2002)
5. CTIF - International Association of Fire and Rescue Services (2008). World Fire Statistics 13
6. Eisenberger, N.I., Lieberman, M.D.: Why rejection hurts: A common neural alarm system
for physical and social pain. Trends Cogn. Sci. 8, 294–300 (2004)
7. Elliot, A.J., Maier, M.A.: Color and psychological functioning. Curr. Dir. Psychol. Sci. 16,
250–254 (2007)
8. Elliot, A.J., Maier, M.A., Moller, A.C., Friedman, R., Meinhard, J.: Color and psychological
functioning: The effects of red on performance attainment. J. Exp. Psychol. Gen. 136, 154–
168 (2007)
9. Elliot, A.J., Niesta, D.: Romantic red: Red enhances men’s attraction to women. J. Personal.
Soc. Psychol. 95, 1150–1164 (2008)
10. Fagley, N., Miller, P.M.: The effects of framing on choice: Interactions with risk-taking
propensity, cognitive style, and sex. Pers. Soc. Psychol. Bull. 16, 496–510 (1990)
11. Fagley, N., Miller, P.M.: Framing effects and arenas of choice: Your money or your life?
Organ. Behav. Hum. Decis. Process. 71, 355–373 (1997)
12. Floyd, D.L., Prentice-Dunn, S., Rogers, R.W.: A meta-analysis of research on Protection
Motivation Theory. J. Appl. Soc. Psychol. 30, 407–429 (2000)
13. Gerend, M.A., Sias, T.: Message framing and color priming: how subtle threat cues affect
persuasion. J. Exp. Soc. Psychol. 45(4), 999–1002 (2009)
14. Gnambs, T., Appel, M., Batinic, B.: Color red in web-based knowledge testing. Comput.
Hum. Behav. 26(6), 1625–1631 (2010)
15. Gorn, G.J., Chattopadhyay, A., Sengupta, J., Tripathi, S.: Waiting for the Web: How screen
color affects time perception. J. Mark. Res. 41(2), 215–225 (2004)
16. Hasseldine, J., Hite, P.A.: Framing, gender and tax compliance. J. Econo. Psychol. 24, 517–
533 (2003)
17. Haynes, M.J.: Fire Loss in the United States during 2014. National Fire Protection
Association, Quincy, MA (2015)
18. Kiene, S.M., Barta, W.D., Zelenski, J.M., Cothran, D.L.: Why Are You Bringing Up
Condoms Now? Eff. Message Content Framing Eff. Condom Use Message. Health Psychol.
24, 321–326 (2005)
19. Levin, I.P., Schneider, S.L., Gaeth, G.J.: All frames are not created equal: A typology and
critical analysis of framing effects. Organ. Behav. Hum. Decis. Process. 76, 149–188 (1998)
20. Lewis, I., Watson, B., Tay, R.: Examining the effectiveness of physical threats in road safety
advertising: The role of the third-person effect, gender, and age. Transp. Res. Part F 10, 48–
60 (2007)
14 L. Chittaro
21. Lewis, I., Watson, B., White, K.M.: An examination of message-relevant affect in road safety
messages: Should road safety advertisements aim to make us feel good or bad? Transp. Res.
Part F 11, 403–417 (2008)
22. Mandel, N., Johnson, E.J.: When Web pages influence choice: Effects of visual primes on
experts and novices. J. Consum. Res. 29(2), 235–245 (2002)
23. Mehta, R., Zhu, R.J.: Blue or Red? Exploring the Effect of Color on Cognitive Task
Performances. Science 323, 1226–1229 (2009)
24. National Fire Protection Association (2013a). Safety Information for kids. http://
www.nfpa.org/itemDetail.asp?categoryID=1803&itemID=42601&URL=Safety
%20Information/For%20kids, Last Accessed 27 November 2015
25. National Fire Protection Association (2013b). Safety Information for Consumers. http://
www.nfpa.org/safety-information/for-consumers, Last Accessed 27 November 2015
26. Rogers, R.W.: Cognitive and physiological processes in fear appeals and attitude change: A
revised theory of Protection Motivation. In: Cacioppo, J.T., Petty, R.E. (eds.) Social
Psychophysiology: A sourcebook, pp. 153–176. Guilford Press, New York (1983)
27. Rothman, A.J., Salovey, P., Antone, C., Keough, K., Martin, C.D.: The influence of message
framing on intentions to perform health behaviors. J. Exp. Soc. Psychol. 29, 408–433 (1993)
28. Rothman, A.J., Salovey, P.: Shaping perceptions to motivate healthy behavior: The role of
message framing. Psychol. Bull. 121, 3–19 (1997)
29. Twitter (2012). http://advertising.twitter.com/2012/10/gender-targeting-for-promoted-
products.html, last accessed 27 November 2015
30. U.S. Fire Administration (2013a) Fire Safety for Kids. http://www.pct3vfd.com/news/
firesafetykids/html/index.html, Last Accessed 27 November 2015
31. U.S. Fire Administration (2013b) Fire Safety for Citizens. http://www.ready.gov/home-
fires, Last Accessed 27 November 2015
32. van den Bos, K., Ham, J., Lind, E.A., Simonis, M., van Essen, W.J., Rijpkema, M.: Justice
and the human alarm system: The impact of exclamation points and flashing lights on the
justice judgment process. J. Exp. Soc. Psychol. 44(2), 201–219 (2008)
33. Witte, K., Allen, M.: A meta-analysis of fear appeals: Implications for effective public health
campaigns. Health Educ. Behav. 27, 591–616 (2000)
34. Wogaiter, M.S., Conzola, V.C., Smith-Jackson, T.L.: Research based guidelines for warning
design and evaluation. Appl. Ergon. 33, 219–230 (2002)
Supporting Users in Setting Effective Goals
in Activity Tracking
Katja Herrmanny, Jürgen Ziegler, and Aysegül Dogangün ✉

( )
Personal Analytics, Interactive Systems Research Group,

University of Duisburg-Essen, Duisburg, Germany
{katja.herrmanny,juergen.ziegler,ayseguel.doganguen}@uni-due.de
Abstract. In this paper we present the development of the pedometer app Move
My Day which implements goal setting as its main persuasive design principle.
Manual goal input as well as two strategies to support users in setting realistic
goals, namely reference routes and personal goal recommendation, were imple‐
mented. The proposed algorithm for adaptive personal goal recommendation is
designed in a way that it recommends short-term goals considering motivational
aspects and gradually raises goals in the long term to meet physical activity
recommendations. In a 12 week field study, we investigated the potentials of the
two support strategies. Results indicate that about half of the users appreciate goal
setting support and that especially personal goal recommendation seems to have
potential to support users in setting effective physical activity goals.
Keywords: Persuasive technology · Behavior change support system · Activity

tracking · Goal setting · Personalization · Pedometer · Physical activity
1 Introduction
Physical activity plays an important role for a healthy lifestyle as it can prevent and
alleviate wide-spread diseases, such as diabetes, coronary heart disease, hypertension
or types of cancer. Increasing physical activity is thus a goal frequently addressed in
persuasive systems research. One of the core design principles used in these systems is
goal setting. According to the goal setting theory [1] pursuing a goal has a motivating
effect on performance. To be motivating and effective, goals should be precise, chal‐
lenging and achievable. Moreover, self-set goals should be preferred over externally
assigned goals as they have been shown to be more effective [1]. However, research [2]
has revealed that people have problems in setting realistic goals, especially for moderate-
intensity physical activity. Also our own research results of an online pre-study (n = 194,
age: M = 27, range = 14–74) showed that participants had problems in estimating real‐
istic goal distances. From these results and findings reported in the literature, it becomes
clear that better strategies for supporting users in setting realistic activity goals are
required when designing persuasive technologies to increase physical activity.
While from a motivational point of view goals should be challenging and realistic,
these goals should also conform to common physical activity recommendations
provided from a medical perspective. A key problem is that users start from very different

DOI: 10.1007/978-3-319-31510-2_2
16 K. Herrmanny et al.
initial activity levels. For many users, standard activity recommendations would be too
high compared to their current activity level to constitute a realistic next goal. For others,
these goals are achievable but may not be challenging. Therefore, we propose an adap‐
tive algorithm that is designed to gradually raise goals in the long term to meet standard
activity recommendations, but recommends realistic and motivating short-term goals.
The algorithm calculates suggestions of individual step goals for a week.
As an alternative support strategy we suggest reference routes, i.e. usually well-
known or user-specific routes or distances which are transformed to the corresponding
number of steps aiming at making these goals more precise.
We implemented a pedometer app integrating these strategies as well as the possi‐
bility to manually enter goals and investigated the supportive potential of these strategies
in an empirical field study.
2 Related Work
Adaptivity of persuasive systems for behavior change is often addressed in research [3–
5]. Based on a literature review, op den Akker et al. [6] identified adaptive goal setting,
among others, as an important research topic. The authors recommend to define a user-
specific, challenging and achievable goal based on the user’s prior activity data.
However, most activity tracking devices and applications that allow goal setting just
assign fixed, non-adaptive goals or provide the opportunity to set a goal without any
support. Example applications for assigned goals are the Houston system [7],
Fish’n‘Steps [8], and UbiFit [9]. An example for user-generated activity goals is an
application targeting persons 50 + on a PDA [4] which enables users to set daily and
weekly individual goals, but does not provide any goal setting support. Until now, only
few approaches provide user-specific, adaptive goals or goal recommendations.
An early approach used a fixed threshold to select a goal level [10]. Users who
achieved more than 5000 steps a day as baseline were given 10000 steps as their new
goal. Those below 5000 steps were given a goal of 5000 steps a day. Burns et al. [11]
also used participants’ baseline activity in their application ActivMon, measured in the
first study week, to define a goal level of 5 % more than baseline activity.
A smartphone application developed by King et al. [12] allows users to type in their
weekly activity goal or choose among three suggested goals based on health recom‐
mendations. The suggested goals increase week by week, provided the previous week
goal has been reached. Otherwise, a lower goal is recommended.
The persuasive system Move2Play offers a personalized recommendation of a
training plan [13]. It bases on a user model, a domain model and the common recom‐
mendation of 10000 steps a day. The user model contains information e.g. on the user’s
fitness and active time slots. The domain model contains data e.g. about the current day,
the user’s age and sex. The author does not mention how this data is considered in the
recommendation nor describe the concrete algorithm.
Also the virtual training system MOPED [14] implements user-specific and adaptive
recommendations in terms of specific exercises. Therefore, a user model was initialized
with manually entered information about sex, age, weight etc. During usage, further data
Supporting Users in Setting Effective Goals in Activity Tracking 17
are added such as fitness test results and historical training data such as exercise level
or frequency as well as the user’s heart rate. Based on these data, recommendations for
the exercise count and speed of the concrete exercises are calculated.
3 Move My Day
We developed the pedometer App Move My Day to investigate our research question.
Three versions with different functions were developed as described in Sect. 4.1. (Three
additional versions were developed for another investigation that is not topic of this
paper.) Goal setting is the predominant persuasive design principle. Permanent graphical
and textual feedback indicates the progress towards the goal (see Fig. 1). In another view
a list of selectable goal components is presented, i.e. distances used as goals or sub-
goals. The app allows for goals consisting of more than one component. Moreover, the
app contains views for recording reference routes (see Sect. 3.1), for typing in goals,
and for viewing one’s personal history. History includes records, average steps per day
and week as well as statistics of the previous days and weeks. In another view, the app
offers the opportunity to manually correct the step number counted by the system. This
function was integrated to obtain valid data also in case of special situations e.g. when
the smartphone’s battery was exhausted or the user forgot it at home that day.
Fig. 1. Screenshots of the home view and goal selection of the pedometer app Move My Day
Goals are set for one week starting on Monday. They are not assigned automatically,
but are set by the users themselves, however, supported by the app. The app provides
three different ways to set a goal– two of them implement support strategies, one is
unsupported. First, goals can be chosen from a list of pre-defined reference routes which
are described in the following section. These reference routes are provided to make goals
more concrete and conceivable than just numerical step counts. Second, as an alternative
to reference routes, Move My Day provides a personal goal recommendation that is
updated each Monday. Its calculation is based on historical user data, general health
recommendations, as well as on motivational aspects and is described in detail in
Sect. 3.2. Third, users can type in their goals manually, without any support of the app.
It is possible to combine different types of goal components into one goal. However, in
some study conditions not all of the three methods are available (see Sect. 4.1).
3.1 Reference Routes

Reference routes represent well-known or self-recorded distances. For instance, one
pre-defined reference route is the distance of a marathon converted to an approximate
number of steps. Further reference routes refer to the distance between two well-known
cities, formula one race tracks, or routes popular with tourists (e.g. Avenue de Champs
Élysées in Paris). Varying the area of interest ensures that each user can associate at
least some reference routes with a distance. Varying length of the routes ensures that
different activity levels can be considered. Moreover, it enables users to either select
one single week goal or a number of smaller sub-goals. Besides pre-defined reference
routes, users can record own routes, e.g. when making a walking tour. They can be
labelled individually and do appear in the reference route list to be chosen as a goal.
3.2 Calculating Goal Recommendations
We implemented an algorithm to calculate weekly goal recommendations considering

motivational aspects and standard activity recommendations. As – following Locke and
Latham – goals should be challenging and achievable to be motivating, both aspects
have been considered in the algorithm. It bases on historical user data, namely the daily
mean performance which is a realistic (day) goal and the daily record, which has been
proven to still be achievable but challenging. The average of both will be calculated and
multiplied with seven to create a challenging but achievable week goal. Using the daily
instead of weekly values allows the system to calculate recommendations from the
second day of usage and normally leads to more challenging goals.
However, this calculation can also cause too high goals if the record is an outlier.
Therefore, we defined a maximum increase of 2000 steps a day based on meta analyses
[15, 16] of physical activity intervention studies using pedometers. Analyses showed a
mean increase of approximately 2000 steps a day compared to the baseline values. As
most studies covered more than one week between baseline and intervention outcome,
an increase of 2000 steps can be seen as challenging. However, considering that in the
described case the record is even higher, the new goal can still be seen as achievable.
Especially in the beginning of app usage, identical or very similar average and record
step counts could hinder increase or lead to an increase that is considered to be too low
against the background of the goal setting theory [1]. To avoid this, the algorithm
includes a check whether the goal is at least 10 % higher than the average step number.
If it is not, the initially calculated goal recommendation will be replaced by the average
step count per day plus 10 % and again multiplied with seven to get a week goal. This
is a higher increase than Burns et al. [11] suggested to keep the goal challenging.
Fig. 2. Flow chart of the goal recommendation algorithm
The algorithm should also consider specific situations. These situations are:
• Historical data are not yet available (cold-start situations).
• The previous step count is unrealistically low (e.g. because usage started the day
before, but not in the morning, or the person is far too little active).
• The user has already achieved a high average level that should not raise to unreach‐
able values.
To meet these demands, thresholds are used to ensure reasonable overall minimum and
maximum values (see Fig. 2). Also, a default value is used to alleviate the cold start
problem which exists in the first week of usage. Modelled after the physical activity
recommendations of the WHO [17], the default, minimum and maximum values are
age-dependent, dividing age groups below 18 years and above. If the user has not entered
his or her age, calculation for adults is used. The default goal recommendation for adults
in case that no historical values are available is 50000 steps a week. This is above mean
step levels former research has identified, but below health recommendations that prop‐
agate 10000 steps a day respectively 70000 steps a week [18]. Thus, it can be seen as a
challenging but achievable goal for an average person. For children and adolescents,
who are recommended much higher levels of physical activity, the default goal is 100000
steps. Unrealistically low goal recommendations, e.g. caused by reasons described
above, are replaced by a minimum value of 21000 steps for adults and 63000 steps for
children and adolescents. Referring to Tudor-Locke et al. [19] the minimum chosen for
adults can be classified as “Limited Activity”, which is in the lower range of physical
activity but above “Basal Activity”. As WHO activity recommendations for persons
under 18 years are three times higher than those for adults, the minimum value for
children and adolescent has been defined as the one for adults multiplied with three. As
maximum values that are implemented to avoid unachievable goals we chose 140000
steps for adults (which is two times the recommendation of 10000 steps a day) and
300000 steps for children and adolescents. In this case, we decided not to multiply the
140000 steps with the factor three as it seemed to be an unrealistic high goal and is
consequently not expected to be motivating.
The goals retrieved from the described algorithm are adaptive to the user’s current
medium-term activity level. In phases with high performance, goals increase whereas
they decrease in phases with low performance due to the decreasing mean step number.
However, by not using the previous week as other approaches do (see Sect. 2) but the
mean and record performance, too high variations from one week to the other are
avoided. Goal adaption is therefore very smooth. In the long-term view, the algorithm
tends to recommend increasing goals by integrating the record which cannot decrease.
4 Field Study Investigating Goal Setting Support
4.1 Method
We performed a comparative field study to investigate users’ preferences regarding goal

setting strategies and the supportive potentials of the personal goal recommendation and
reference routes. We published Move My Day at Google Play Store and promoted the
app via press to generate a representative sample. Before downloading, prospective users
were informed about taking part in a scientific study by using the app.
Users’ app usage behavior was logged and stored on a server for data analysis.
Logged data include test condition, created goal components, selected goal components,
steps performed, app usage (i.e. opening the app), relevant technical data as well as
gender, year of birth and former experience with activity tracking if provided by the
user. Besides this information retrieved directly from the app, there were no other types
of data collection. Also, there was no direct contact between the investigator and the
users. Moreover, no end date of the study was communicated. Users can use the app
without time limit. This study design was chosen to get real usage data without any bias
evoked by the study situation. When starting the app for the first time, users were
randomly assigned to one of three study conditions. The conditions contain either the
adaptively calculated goal recommendation, reference routes or both. Additionally, we
integrated the opportunity of setting goals manually in all conditions, in order to inves‐
tigate, if users want to get supported by the app or not. This design decision results in
the following conditions:
• Condition Rec + M: This condition included adaptively calculated goal recommen‐
dations as well as the possibility to manually type in a goal.
• Condition Ref + M: This condition included the opportunity to choose from reference
routes or goals typed in manually.
• Condition Rec + Ref + M: This condition included all three possibilities to set a goal.
4.2 Data Processing and Analysis
The observation period covered 12 weeks. Logged data were filtered to meet our inclu‐
sion criteria for a high data quality:
• First usage week was eliminated as we supposed that users would test the functions
and select different goal components just to try them.
• If users terminated using the app during the investigation period, also the last incom‐
plete week was eliminated.
• Only weeks of active usage were included in the analysis. Our defined criterion for
active usage was opening the app at least seven times a week.
• Further criteria were used to check the trustworthiness of data, especially regarding
manual corrections. Criteria include negative step counts resulting from manual
corrections, unrealistic high step counts (more than 70000 steps a day), and manual
corrections on Sunday with nearly the amount of missing steps for reaching the goal.
Unfortunately, due to technical problems some data was not stored for some data sets
resulting in further exclusions. Furthermore, we removed combined goals (i.e. goals
consisting of more than one type of goal component) from statistical comparison of the
support strategies as no clear allocation is possible. As there is a naturally high variation
in step performance and goals, statistical methods considering such variance had to be
chosen. Therefore, we focus on descriptive data, effect sizes and report confidence
intervals for better interpretation of the results. As there is an ongoing debate about the
informative value of significance testing, we also provide results of significance tests
for the sake of completeness.
4.3 Results
Sample. After the described exclusions, a sample size of 79 participants (27 male, 34
female) and 206 weeks remained. Mean age was 47 years (n = 62, min = 12, max = 72).
18 participants didn’t indicate their age or sex. Asked about previous experience with
activity tracking (at least one month during the last year), 12 indicated to have experi‐
ence, 50 were unexperienced and 17 did not indicate any experience information.
Investigation of Used Goal Components. We first investigated the number of sub-

goals, a goal consisted of. Mostly, a goal included one component (142 times), followed
by two (34 times), three (14 times), four (9 times), seven (6 times), and five (1 time). In
condition Ref + Rec + M users could choose between all three types of goal components.
48 % of set goals were entered manually whereas for 44 % a support strategy was used
(ref. route: 24 %, personal rec.: 20 %). 8 % were combined goals. Also in condition
Rec + M most goals were manual inputs (64 %), followed by recommendation (31 %)
and combined goals (5 %). In condition Ref + M most goals were supported by reference
routes (65 %), 39 % were unsupported inputs and 6 % combined goals. Notably, all
reference routes in all conditions were predefined ones. Even though users recorded own
routes, they were not set as goals.
Investigation of Steps Per Week. We compared the steps users took per week
between the three conditions as well as between the types of goal. Results are
presented in Tables 1 and 2. Participants in condition Rec + Ref + M achieved
higher performance than those in condition Rec + M and Ref + M, however with
high standard deviation. Calculation of effect size showed a small effect
. Difference is statistically not significant (F(2, 75) = 0.699, p = .500).

Comparing the number of steps taken between the different types of set goals,
highest performance was found for goals based on recommendation followed by
reference routes and manual input. Analysis shows a small effect and no statistical
significance .
Table 1. Descriptive results of achieved steps per week, grouped by condition

Condition n min max M SD 95 % CI
Rec + M 50 2730 98424 36954.08 27346.47 29182.30 - 44725.86
Ref + M 39 913 70288 34378.62 18216.04 28473.66 - 40283.57
Rec + Ref + M 36 591 95428 44361.81 30877.15 33914.48 - 54809.13
Table 2. Descriptive results of achieved steps per week, grouped by used type of goal
Type of goal n min max M SD 95 % CI
Recommend. 9 4145 80133 42195.67 26702.71 21670.15 - 62721.19
Ref. Route 28 591 70288 35462.86 22439.32 26761.80 - 44163.92
Man. Input 41 913 92783 31774.44 25923.79 23591.88 – 39957.00
Investigation of Set Step Goals. Set goals were also compared between the conditions
and the types of chosen goals. Mean goals were very similar for all study conditions (see
Table 3). For all conditions they comply on average with common physical activity
recommendations. Regarding the type of goal the targeted steps also comply with phys‐
ical activity recommendations for personal recommendation and reference routes. For
manual input they are slightly below (see Table 4).
Table 3. Descriptive results of set goals, grouped by condition

Condition n min max M SD 95 % CI
Rec + M 55 3500 200000 56159.27 38062.04 45869.66 - 66448.88
Ref + M 85 8 220789 55659.46 42905.84 46404.88 - 64914.03
Rec + Ref + M 66 1770 345456 53745.38 55145.24 40188.99 - 67301.77
Table 4. Descriptive results of set goals, grouped by used type of goal

Type of goal n min max M SD 95 % CI
Recommend. 30 21000 97398 52412.87 21658.91 44325.30 - 60500.44
Ref. Route 71 3031 220789 58959.63 46988.02 47837.75 - 70081.52
Man. Input 92 8 200000 45589.48 33199.60 38714.04 - 52464.92
As mean goals are of limited informative value when investigating if realistic goal
setting can be supported, further analysis focused on the difference between actually set
goals and a reasonable goal. Based on goal setting theory [1], a reasonable goal should
be challenging and achievable. However, there are large inter-personal performance
differences. Goals that seem high or even not achievable for one person might not even
be challenging for another. Thus, we investigated deviation of set goals from an indi‐
vidually realistic level. This level was estimated by the steps actually reached.
Comparing the mean deviation values between the different types of goals shows
that performance was overestimated in all groups (see Table 5). Highest deviations are
revealed for combined goals, lowest for goals based on the personal recommendation.
This is a small effect . A Kruskal-Wallis test shows that differences are
statistically not significant (χ2(2) = 2.27, p = .322).
Table 5. Descriptive results of deviations (pos. and neg.) between estimated realistic perform‐
ance and set goals, grouped by used type of goal
Type of goal n min max M SD 95 % CI Mean rank
Rec. 9 -32089 30083 -5396.56 20201.27 -20924.63 - 10131.51 47.89
Ref. Route 28 -85527 41399 -12180.25 28826.70 -23358.08 - -1002.42 47.64
Man. Input 41 -172566 27783 -17361.44 32306.63 -27558.67 - -7164.21 39.88
Comparing the step count deviations ensures a uniform level of deviation. However,
additionally we compared the percentage deviations from the realistic goal level in order
to consider highly varying goals and performances. Therefore, we calculated the abso‐
lute value of mean percentage differences between the estimated realistic performance
and set goals (see Table 6) and compared them between the types of goal. Again, we
found a small effect and no significance (F(2, 75) = 1.807, p = .171).
Table 6. Descriptive results of the absolute value of percentage deviations between estimated
realistic performance and set goals, grouped by used type of goal
Type of goal n min max M SD 95 % CI Mean rank
Rec. 9 10.21 406.63 75.17 125.80 22.77 – 176.94 40.22
Ref. Route 28 0.06 10105.92 764.54 2267.26 52.33 – 1752.71 36.39
Man. Input 41 0.91 1093,32 162.79 228.81 80.48 – 254.19 41.46
Realistic goal setting might not only be supported when participants make use of a
type of goal component. Also just offering these types of goal component could have a
supportive effect on manually entered goals. Therefore, we compared the deviations for
the manually entered goals of all conditions. As Table 7 shows, differences between
goals and estimated realistic level as well as standard deviations are obviously lower in
condition Rec + Ref + M than in the other conditions. For condition Ref + M descriptive
analysis shows a slightly lower negative mean value and lower standard deviation than
for Rec + M. For this comparison between the study conditions a Kruskal Wallis test
revealed a significant effect (χ2(2) = 9.40, p = .009). Effect size is medium-ranged
. Post hoc testing indicates a significant difference between Rec + M and
Rec + Ref + M (p = .041) as well as Ref + M and Rec + Ref + M (p = .012).
Table 7. Descriptive results of deviations (pos. and neg.) between estimated realistic perform‐
ance and manually entered goals, grouped by condition
Condition n min max M SD 95 % CI Mean rank
Rec + M 20 -172566 27783 -21018.15 39976.46 -39727.71 - -2308.59 19.00
Ref + M 10 -60739 10274 -28474.50 25241.67 -46531.30 - -10417.70 15.00
Rec + Ref + M 11 -19803 23069 -610.09 10301.07 -7530.44 - 6310.26 30.09
4.4 Discussion
When having all three opportunities of goal setting, participants used supported and unsup‐
ported goal setting equally. When just one support strategy was provided, users preferred
manual goal input or reference routes, depending on the study condition. Personal goal
recommendation was used less than the other opportunities, but still in 20-31 % of set
goals. Obviously it is a matter of individual preference whether users want goal setting
support or not and which strategy they like. Regarding reference routes only predefined
ones were selected although participants had used the recording function. Taking a deeper
look at the recorded routes showed that step numbers were very low and thus not suitable
for week or day goals. Labeling of these routes, e.g. “Home to work”, indicates, that people
used the recording function to evaluate the step number of their everyday routes. Consid‐
ering that the observation period did not cover holiday time, it seems plausible to assume
that there was not yet much opportunity to record suitable routes (e.g. a longer walk, or a
hiking trip). Thus, investigating the use of personal reference routes remains an interesting
research aspect for long-term evaluation.
The absolute goal levels did not differ substantially between the study conditions,
whereas deviations between personally realistic and actually chosen manually entered
goals did. The significant medium-sized effect indicates that adding both supportive
strategies, reference routes as well as personal recommendation, to the manual goal
provides larger support in goal setting than just one of these opportunities even when
these components are not selected.
When comparing absolute goal levels between the types of set goals, those based on
reference routes were apparently the highest, those on manual input the lowest.
However, evaluating how realistic these goals were, shows a unified picture: Goals based
on personal recommendation obviously had lowest mean deviation between goals and
estimated realistic goal level as well as lowest standard deviation for all calculations
(absolute difference and percentage difference), indicating that the personal recommen‐
dation has most potential in supporting users in setting realistic activity goals. Goals
based on reference routes do not seem superior to manually entered goals regarding user
support. Although there is no statistically significant effect, at least descriptive differ‐
ences between the achieved steps are in accordance with this tendency as highest
performance was achieved when using the personal recommendation. In general, natu‐
rally high inter-personal variation regarding goals as well as step performance makes
valuation of statistical significance difficult. However, descriptive data, found small
effects and a unified picture regarding the different calculations indicate, that the found
differences might not be random effects.
5 Conclusion and Limitations
We presented the app Move My Day which implements goal setting as its core persuasive
design principle. Two strategies to support users in setting realistic goals, namely refer‐
ence routes and personal recommendation calculation, as well as manual goal input were
implemented. In a 12 week field study, we investigated the potentials of these supportive
strategies. Results show that users set goals with and without support equally, if all types
of goal setting are provided. Reference routes were the preferred support strategy, but
also the personal recommendation was used. However, on average the chosen goals that
based on personal recommendation tended to be more realistic than goals from other
sources and might have led to descriptively higher performance.
Even though the found effects are not strong, results indicate that many users appre‐
ciate being supported in goal setting and that personal recommendation based on the
described algorithm might have a supportive influence in realistic goal setting. When
defining goals by manual input, providing reference routes and personal recommenda‐
tion as additional supportive information seems to be most effective. However, in order
to investigate users’ acceptance and preferences regarding the type of goal setting and
the influence of just offering support on manual goal input, in our study design all
conditions additionally contained the possibility of unsupported goal setting. For further
investigation it would be interesting to only provide one opportunity of goal setting per
condition and have a larger sample size, to alleviate comparison of the strategies.
However, for the design of activity tracking technologies we recommend integrating
manual input as well as goal setting support, ideally user-adaptive.
Regarding self-recorded reference routes, long-term data will reveal if people use
them for their goals and if they are effective in supporting realistic goal setting.
Although the developed algorithm for personal goal recommendation shows prom‐
ising results, there might be limitations for long-term use. In long-term use, the algorithm
will get more and more insensitive to new records or variations in the mean performance.
Possibly it will be necessary to neglect older information units in order to remain the
flexibility of the algorithm, which is topic of our further evaluation. We also plan to
differentiate days of work, holiday and illness to improve the recommendation. We will
investigate these aspects in a long-term analysis of the ongoing study.
References
1. Locke, E.A., Latham, G.P.: Building a practically useful theory of goal setting and task
motivation: A 35-year odyssey. Am. Psychol. 57(9), 705–717 (2002)
2. Saini, P., Lacroix, J.: Self-setting of physical activity goals and effects on perceived difficulty,
importance and competence. In: 4th International Conference on Persuasive Technology
(Persuasive), pp. 33–39. ACM Press, New York (2009)
3. Hawkins, R.P., Kreuter, M., Resnicow, K., Fishbein, M., Dijkstra, A.: Understanding tailoring
in communicating about health. Health Educ. Res. 23(3), 454–466 (2008)
4. King, A.C., Ahn, D.K., Oliveira, B.M., Atienza, A.A., Castro, C.M., Gardner, C.D.:
Promoting physical activity through hand-held computer technology. Am. J. Prev. Med.
34(2), 138–142 (2008)
5. op den Akker, H., Jones, V.M., Hermens, H.J.: Tailoring real-time physical activity coaching
systems: a literature survey and model. User Model. User-Adap. Inter. 24(5), 351–392 (2014)
6. op den Akker, H., Klaassen, R., op den Akker, R., Jones, V.M.; Hermens, H.J.: Opportunities
for smart & tailored activity coaching. In: 26th IEEE International Symposium on Computer-
Based Medical Systems (CBMS), pp. 546–547. IEEE Press (2013)
7. Consolvo, S., Everitt, K., Smith, I., Landay, J.A.: Design requirements for technologies that
encourage physical activity. In: SIGCHI Conference on Human Factors in Computing
Systems (CHI), pp. 457–466. ACM Press, New York (2006)
8. Lin, J.J., Mamykina, L., Lindtner, S., Delajoux, G., Strub, H.B.: Fish’n’Steps: Encouraging
physical activity with an interactive computer game. In: Dourish, P., Friday, A. (eds.)
UbiComp 2006. LNCS, vol. 4206, pp. 261–278. Springer, Heidelberg (2006)
9. Consolvo, S., Klasnja, P., McDonald, D.W., Avrahami, D., Froehlich, J., LeGrand, L., Libby,
R., Mosher, K., Landay, J.A.: Flowers or a robot army? In: 10th International Conference on
Ubiquitous Computing (UbiComp), pp. 54–63. ACM Press, New York (2008)
10. Bickmore, T.W., Caruso, L., Clough-Gorr, K.: Acceptance and usability of a relational agent
interface by urban older adults. In: CHI 2005 Extended Abstracts, pp. 1212–1215. ACM
Press, New York (2005)
11. Burns, P., Lueg, C., Berkovsky, S.: Activmon: Encouraging physical activity through ambient
social awareness. In: CHI 2012 Extended Abstracts, pp. 2363–2368. ACM Press, New York
(2012)
12. King, A.C., Hekler, E.B., Grieco, L.A., Winter, S.J., Sheats, J.L., Buman, M.P., et al.:
Harnessing different motivational frames via mobile phones to promote daily physical activity
and reduce sedentary behavior in aging adults. PLoS ONE 8(4), e62613 (2013)
13. Bielik, P.: Personalized training plan recommendation and activity tracking for a healthier
lifestyle. Bull. ACM Slovakia 3(4), 39–40 (2011)
14. Buttussi, F., Chittaro, L.: MOPET: A context-aware and user-adaptive wearable system for
fitness training. Wearable Comput. Artif. Intell. Healthc. Appl. 42(2), 153–163 (2008)
15. Bravata, D.M., Smith-Spangler, C., Sundaram, V., Gienger, A.L., Lin, N., Lewis, R., Stave,
C.D., Olkin, I., Sirard, J.R.: Using pedometers to increase physical activity and improve
health: a systematic review. JAMA 298(19), 2296–2304 (2007)
16. Kang, M., Marshall, S.J., Barreira, T.V., Lee, J.O.: Effect of pedometer-based physical
activity interventions: A meta-analysis. Res. Q. Exerc. Sport 80(3), 648–655 (2009)
17. World Health Organization: Global recommendations on physical activity for health (2010).
http://whqlibdoc.who.int/publications/2010/9789241599979_eng.pdf?ua=1
18. Choi, B.C.K., Pak, A.W.P., Choi, J.C.L.: Daily step goal of 10,000 steps: A literature review.
Clin. Invest. Med. 30(3), 6 (2007)
19. Tudor-Locke, C., Johnson, W.D., Katzmarzyk, P.T.: Accelerometer-determined steps per day
in US adults. Med. Sci. Sports Exerc. 41(7), 1384–1391 (2009)
Persuasive and Culture-Aware Feedback Acquisition
Malik Almaliki1 ✉ and Raian Ali2

( )
1
Taibah University, Medina, Kingdom of Saudi Arabia
[email protected]
2
Bournemouth University, Bournemouth, UK
[email protected]
Abstract. User feedback is an important factor to improve software quality. For

example, it can provide information on missing features and clarify user trends
and preferences for future improvement. However, gathering user’s feedback is
not an easy process since the majority of users lack motivation and interest in
providing feedback, especially in a constant and frequent style. In addition,
studies have noted that the cultural differences among users also play a role in
affecting their motivations to feedback acquisition. In this paper, we empirically
investigate the role of culture in affecting users’ perception and motivations to
give feedback. Our study identifies some key differences between Western and
Middle Eastern users on what motivate them to provide feedback and what could
have an influence on the quality of the feedback they give. This also makes the
case for the need to design a persuasive and culture-aware feedback acquisition.
Keywords: User’s feedback · Persuasive technology · Software engineering
1 Introduction
In the context of software system, the software’s ability to adapt to different cultures in
users’ space is important for improving its success in wider contexts and constructs a
key requirement for professional and ethical reasons [1]. Generally speaking, most soft‐
ware designs follow a western cultural cues. This has caused a design gap when users
coming from different cultures (i.e. eastern cultures) use the software within their
cultural context. An example is clearly seen in the different ways people from all over
the world use social networks. This is perhaps due to the fact that software industry is
largely led by western management and developers [2]. Therefore, software systems
that are marketed worldwide need to be tailored to fit the different cultures [3] as designs
that are successful in one culture may fail dramatically in others [4, 5].
User’s feedback constructs an important source for information needed for planning
software evolution and adaptation [6]. This means that users need to be persuaded to
provide feedback frequently. However, persuading users is not an easy task as the
majority of users have little motivation and interests in providing such feedback in an on-
going style [7]. Moreover, motivating users would highly depend on their culture and
values. This sheds the lights on the potential use of persuasive technology in order to
empower users’ willingness to give feedback thus their experience and software’s success.

DOI: 10.1007/978-3-319-31510-2_3
28 M. Almaliki and R. Ali
Persuasive technology refers to the technology that is designed to change the attitudes or
behaviors of users through persuasion and social influence, but not through coercion [8].
It has been used as an effective approach to increase users’ engagement in many areas
including the area of human-computer interaction. Recent research indicates that persua‐
sive technology is more effective when it is tailored to the culture of its intended target
audience [9]. However, little research has explicitly investigated the relationship between
persuasive technology and culture [10].
In this paper, the authors qualitatively (using a focus group approach) study the effect
and impact of cultural backgrounds on users’ social motives to give feedback. The study’s
focus is on two different cultural backgrounds: Middle Eastern users and European users.
The paper also advocates the need to design the acquisition process taking into account the
persuasion goal of feedback acquisition in different cultures. The results of this study are
meant to help devising a persuasive and systematic method for conducting a socially aware
feedback acquisition that can adapt to different types of users in terms of their cultural
backgrounds (Middle Eastern and European users). This will ultimately maximize feed‐
back quality, users’ satisfactions and motivations to give feedback.
The paper is structured as follows. In Sect. 2 we briefly discuss feedback acquisition
and its relation to cultural backgrounds. In Sect. 3 we describe the research method
adopted. In Sect. 4 we present and discuss the results of our study and in Sect. 5 we
present our conclusion.
2 Feedback Acquisition and Cultural Differences
In [7] the authors conducted an empirical study to investigate users’ behaviour to feed‐
back requests and discover what motivates them to provide feedback. The study
consisted of two phases: a qualitative and a quantitative phase. In the qualitative phase
7 interviews were conducted whereas in the quantitative phase, a survey was used and
100 responses were collected. The results of their study showed a preliminary indication
that the variant cultural backgrounds of users has a noticeable impact on their behaviour
and how they are socially motivated to give feedback. The preliminary results and survey
questions related this aspect are provided in [22].
While a number of researchers have already investigated cultural differences in rela‐
tion to software such as how graphics, language, object formatting, colours, and layout
of web sites and other user preferences is perceived in different cultures [11–14], to the
our best knowledge no studies have yet investigated how users with different back‐
grounds behave in response to feedback acquisition and how their culture frame affect
their motivations to feedback requests. This paper aims at qualitatively investigating
this aspect to help improving the design of a persuasive and culture-aware feedback
acquisition.
3 Research Method
In empirical research, researchers might need to qualitatively follow up or build upon

quantitative results for the purpose of explaining or further investigating the quantitative
Persuasive and Culture-Aware Feedback Acquisition 29
results [15]. This design is suitable for researchers who need qualitative data to explain
significant, non-significant or surprising quantitative results [16, 17]. The results
discussed in [7] indicated that the variant cultural backgrounds of users have a noticeable
impact on users’ behaviour and how they are socially motivated to give feedback. This
was only an indicator which would need confirmation and clarification. To achieve that,
this study follows up the quantitative results in [7] with a qualitative phase through focus
groups which is a powerful tool to get insights and stimulate discussions in a small group
of participants.
A focus group is a qualitative research method in which a group of people are gath‐
ered to be asked about their opinions, beliefs, or attitudes regarding an issue, phenomena,
service, etc. The questions are asked in an interactive setting which allows participants
to talk freely about their thoughts to other group members. Due to the nature of the study
where social/group interaction is needed and the type of the asked questions which
mainly relate to the culture impact (Middle Eastern and European cultures) on social
perception of feedback acquisition of both the feedback provider and those who watch
it, this paper adopted Focus Group as a data collection method in this investigatory study.
3.1 Focus Group Design

Four Semi-structured focus groups were conducted in two countries (Saudi Arabia and
the UK) with 27 participants to further explore how Middle Eastern and European users
behave in response to feedback acquisition and how their culture frame affect their
motivations to feedback requests. Participants were carefully selected in order to guar‐
antee a high level of diversity and to avoid bias (e.g. various age groups, backgrounds
and gender). The first two focus groups took place on June-2014 in the UK and were
conducted with European participants to investigate how their culture impacts their
perception to feedback acquisition. Whereas the other two focus groups were held on
July-2014 and conducted with Saudi participants to investigate the Middle Eastern
culture’s impact to feedback acquisition. Each focus group session lasted for about an
hour which makes an amount of four hours in total.
The focus groups protocol was developed in the light of the results discussed in [7]
in which an indication to cultural differences between European and Middle Eastern
users on what motivate them to provide feedback was discovered. In particular, the four
social factors (Feedback acquisition as a social activity, Social recognition, Volume of
already given feedback and Visibility and similarity of others feedback) that influence
how Middle Eastern and European users are socially motivated to give feedback served
as a foundation to develop the protocol of the focus group. The focus group protocol is
available in [22].
The protocol was iteratively reviewed and revised by 3 researchers to ensure clarity
and understandability. Participants were briefed to the session and the discussed topic
through a 10 min presentation in which some example of feedback acquisition in soft‐
ware application were also given to more familiarise the participants with the discussed
topic. Each participant received £15 amazon vouchers as an appreciation for taking part
in the study.
3.2 Sampling
Purposeful sampling is a common technique in qualitative research [18]. In this study,
purposeful sampling was used to recruit the participants. The inclusion criteria of this
study allowed for participants who are European or were born and raised in Europe to
take part in the first two focus groups dedicated to study European people (7 participants
for the first focus group and 6 for the second). On the other hand, participants who are
Saudis or were born and raised in Saudi Arabia were recruited to take part in the other
two focus groups dedicated to study Middle Eastern people in which 7 participants took
part in each focus group.
In addition, the inclusion criteria allowed for participants within an age range of 18
to 71 and average computer users who use typical and diverse set of popular software
applications rather than domain specific software for everyday life activities. This
sampling criterion were developed to allow for more variety in selecting participants
and reflecting users’ experience with popularly used software applications. This can
maximize the generalizability of the results. For more details about the participants’
characteristics (e.g. age, gender, home country, etc.) please refer to [22].
The authors assumed that Saudi users could be a good fit to represent the Middle
Eastern culture. It is due to the fact that Saudi Arabia is one of the largest countries in
the Middle East and could fairly represent users’ cultures in the region when it comes
to the use of software. Statistics indicate that Saudi Arabia has over 6 million active
Facebook users which is the highest Facebook user rate in the region. With more than
3 million active Twitter users, Saudi Arabia takes the lead not only in the Middle East,
but in the world in its Twitter users’ growth rate. In addition, More than 90,000,000
videos are watched daily on YouTube in Saudi Arabian which is more than any daily
YouTube video views number in the world [19]. Based on this statistics, the available
resources to the authors and time, the authors assumes that Saudi Arabia is a reasonable
fit to be adopted in this study to represent Middle Eastern culture to feedback acquisition.
3.3 Analysis
Focus groups were audio taped and transcribed verbatim analysis was performed in
several steps which included: (1) initial exploration of the gathered data by reading the
transcripts; (2) coding data by labelling and segmenting the text; (3) using codes to
generate themes by gathering similar codes together; (4) connecting, comparing and
interrelating themes. Credibility of the findings was maximized by using an inter-coder
agreement check and academic advisor’s auditing [18, 20].
4 Findings
As previously stated, the focus group design covered how a number of social factors
influence Middle Eastern and European users with regard to feedback acquisition. In
this section, these social factors are used to structure, represent and discuss the main
themes of the findings. The themes and codes that highlight the encountered behavioural
differences between the two studied groups (Middle Eastern and European users) to
feedback acquisition are shown in Table 1.
Table 1. A breakdown of the themes and codes of the analysis.
4.1 Visibility and Similarity of Others Feedback
Generally speaking, the responses from Middle Eastern participants and European
participants are noticeably different when they were asked whether the visibility of
others feedback (the ability to others feedback before giving feedback) and the similarity
of their feedback to others feedback would have effect on their willingness to give feed‐
back. Although feedback visibility plays a role in motivating both Middle Eastern and
European participants to give feedback, Middle Eastern participants seemed to be more
concerned and socially motivated by this factor than European participants. This trend
became more obvious when they were asked whether knowing the similarity of their
feedback to others would affect their willingness to give feedback. One the participants
mentioned in this regard “I’m normally interested in replying to reviews that I do not
agree with”. Figure 1 gives a general view of how the different cultures’ impact on this
social factor could affect the quality of given feedback. In addition, the following
dimensions of this theme were extracted from the participants’ responses:
Anonymity of Feedback Providers. Users ability to give feedback anonymously (i.e.

using nicknames instead of the real names) is shown to play a role in motivating both parties
(Middle Eastern and European users) to give feedback especially when the given feedback
is publicly open and seen by other users. This is perhaps due to several factors (i.e. disso‐
ciative anonymity, invisibility, solipsistic introjection, dissociative imagination, and mini‐
mization of authority) as discussed in [7] which allow users to enjoy more freedom in
expressing their opinion about a product or a software service. One of the participants said
“I would be more interested in truly engaging when no one knows me. It feels like you are
free to say whatever you like and you will still be unknown”. However, anonymity does not
seem to be a motivating factor when a feedback provider is part of an online closed
community such as a Facebook group. In general, users prefer to know the identity of the
feedback provider in their closed social network/community since it makes them more
comfortable participating and discussing a software service or a product.
Fig. 1. The effect of feedback visibility and similarity in Middle Eastern and European culture.
On the downside, anonymity seems to affect the objectivity and relevancy of

provided feedbacks by both parties but it is more prominent with the Middle Eastern
than European users. Their overall responses indicated that anonymity could results in
the so-called online disinhibition effect [21] which gives users space to escape from their
social constrains. However, this feeling of freedom can result in an overexpression and
less objectivity and relevancy of their opinion about a service or a product. One of the
participants commented “The problem is that some people think they can say anything
or be unfair or even harm others because nobody knows them. That’s not the point of
being anonymous”. Figure 2 gives a clearer view of how the different cultures’ impact
on this social factor could affect the quality of given feedback.
Fig. 2. The effect of anonymity on feedback in Middle Eastern and European culture.
Cross Conversation, Feedback Objictivety and Relevancy and Language Used. In

the context of cross conversation (irrelevant feedback) and how it affects the provided
feedback, users’ responses indicated that feedback given by European users might enjoy
a slightly higher degree of relevancy and objectivity than the feedback given by Middle
Eastern users. This is perhaps due to the fact that Middle Eastern users are more socially
engaged which could put some constrains on the relevancy and objectivity of their given
feedback However, in all cases cross conversation (irrelevant feedback), subjectivity
and the harshness degree in the language used among already given feedback can results
in a low response rate by both parties (Middle Eastern and European users). It can also
result in a harm to the software product or the provided service.
Fig. 3. The effect of cross conversation and language used on feedback in Middle Eastern and
European culture.
An example of this was given by one of the participants “if the cross conversations
or the harshness of the used language leverage among users feedback such as two large
groups of users fighting around irrelevant specific religious or political party, this can
highly result in users from either group degrading and disliking the service or the
product just to cause a harm to the service provider who they think he might be from
the opposite party”. Figure 3 gives a clearer view of how the different cultures’ impact
Gender, Social Position or a Personal Relationship with a Feedback Provider.

Being able to know/see the gender, social position or your personal relationship with
a feedback provider gives a clear view about how users from the different two
cultures (Middle Eastern and European cultures) are socially motivated to give feed‐
back. In the European culture, users do not seem to be influenced by these factors to
give feedback at all. This is perhaps due to the fact that they feel more socially- free‐
wheeling than Middle Eastern users as described by some of the participants.
A participant commented “I believe living in this community (European community)
makes you less socially-dependent and do not easily accept to accomplish things with
the help of your parents or friends for example. You always want to do things by
yourself. This is how we grew up”.
On the other hand, users coming from Middle Eastern culture are significantly
different to the European users in their perception of these factors. They feel highly
motivated by these factors to give feedback especially with people/users they know (i.e.
having a personal relationship with the person asking for feedback or the users giving
feedback). Interestingly, the gender of a feedback provider is considered to be source
of curiosity that motivates Middle Eastern users to give feedback which is not the case
with the European users. Males/females would find it interesting to see how females/
males think of a particular software service or product. One of the participants said “I
would really love to see how males’ feedback would be on a certain aspect of a software
such as the interface colours. You know we love girly colours and this will always make
my feedback clashes with males which is fun”.
On the down side, motivating Middle Eastern users by these factors can result in a
questionable quality of their given feedback in terms of its objectivity and relevancy to the
discussed software service or product. Some of the Middle Eastern participants mentioned
Fig. 4. The effect of gender, position or relationship on feedback in Middle Eastern and European
culture.
that they always tend to be supportive and on the side of their friends (i.e. software service
provider who is asking for their feedback) regardless of the discussed product and their real
opinion about it. Although this can be harmful to the quality of the feedback, it gives a nice
example of a caring relationship among users. Figure 4 gives a clearer view of how the
different cultures’ impact on this social factor could affect the quality of given feedback.
4.2 Volume of Already Given Feedback
When participants were asked whether the number of feedback already provided on a
software service or a product would affect their willingness to give feedback, Middle
Eastern users showed a consensus on that they would like to provide feedback if there
were only few feedback given rather than a large number of already given feedback.
One of the participants commented “it makes me feel sorry when only few reviews are
given especially when the app or service is good. I would certainly find a time to write
my own review and help”. In comparison, European users had a similar attitude but they
perceive this as a less important factor when compared to Middle Eastern users.
Feedback Objectivity and Relevancy. The degree of feedback relevancy and objec‐
tivity given by both parties (Middle Eastern and European users) does not seem to be
influenced by the volume of already given feedback. However, the low number of
already given feedback can sometimes impose potential risk to the software service or
product. Several users from both parties indicated that when a low number of feedback/
reviews is already given it makes them lose interest in the provided service since it does
not seem to be popular among users otherwise it would be highly reviewed by a large
number of users. One of the participants commented “The first thing I do before down‐
loading an app is to look at the number of reviews. A low number means to me less
popular and useful”. Figure 5 gives a clearer view of how the different cultures’ impact
4.3 Social Recognition
Participants were asked whether being recognized by the community as feedback

providers would affect their willingness to give feedback. The responses from the two
groups were noticeably different. Users from Middle Eastern backgrounds indicated that
being socially recognized as a feedback provider is an influential factor that could posi‐
tively maximize their willingness to give feedback. There could be still some constraints
on this, e.g., some participants commented that “it is nice to be visible only when others
can see their feedback which led to some changes on the system”.
Fig. 5. The effect of feedback volume on feedback in Middle Eastern and European culture.
On the other hand, European users seemed to be far less motivated by the same aspect
as few participants would have thought in a similar way. In fact, some of the European
participants mentioned that some of the socially recognized feedback providers might
have been selected and sponsored by the software/service provider for marketing reasons
(e.g. being a celebrity). One of the participants commented “You know nowadays in
advertisements they use popular names and faces to attract people. It can be the same
case for feedback too”. Although Middle Eastern users are also aware of this threat but
they would still be more interested in the social aspect (social recognition) regardless
of the potential threats it might cause. Generally, these potential threats could result in
users having negative attitudes towards the provided software service or the product.
Figure 6 gives a general view of how the different cultures’ impact on this social factor
could affect the quality of given feedback. In addition, the following dimensions of this
theme were extracted from the participants’ responses:
Feedback Objectivity and Relevancy. Social recognition as a motivating factor to

feedback acquisition does not seem to play a role in affecting the objectivity and rele‐
vancy of feedback given by European users. In contrast to, the degree of relevancy and
objectivity of feedback given by Middle Eastern users to the provided service might be
slightly harmed. This is due to the fact that Middle Eastern users are more socially
involved with their community and this can push them sometimes to act differently when
they are being socially recognized as feedback providers. One of the participants
commented “You know lots of people could be watching me. I will always try my best
to be ideal in their eyes”. This indeed could have negative impact on their given feedback
such as imposing favouritism in their opinions regarding the provided software service
which made them socially recognized at the first place. However, this social aspect could
have a positive impact on the social behaviour within the online society of Middle
Eastern users since socially recognized users would feel more socially constrained and
Fig. 6. The effect of social recognition on feedback in Middle Eastern and European culture.
their behaviour is always under the spot light of the community. This could result in an
overall improved social behaviour and an online user community.
Suitable and Unsuitable uses of Social Recognition. As indicated by users’ responses

from both parties (Middle Eastern and European users) social recognition could be more
beneficial and motivating if it is used in a small and closed community of users where
all of the users somehow know each other (i.e. Facebook groups). This makes users
more interested in being socially recognized to people who they know and more moti‐
vating for users to follow up a socially recognized person who they know and trust. One
of the participants commented “It would be nice if we are all friends and know each
other. I will trust and follow who I know and I will be happy to be recognized in front
of people who care about me, right?”
On the other hand, users indicated that social recognition can lead to users following
socially recognized users only and ignoring the feedback of unrecognized users even if
their feedback is far better in quality. This of course could harm the provided software
service and its reputation since vital information and knowledge about the provided
service could be overlooked. In addition, some users mentioned that social recognition
could lead to users’ addiction to the used software especially young users who are eager
to be socially recognized and more socially active. It is a trade-off between using this
factor to benefit users or harm them and the provided service.
4.4 Feedback Acquisition as a Social Activity
Similar to the above dimension, Middle Eastern users showed a much higher interest in
conducting feedback acquisition as a social activity (i.e. social games) and emphasized
that it would increase their willingness to give feedback. This was true especially for
young users. Example of such an activity could be the users’ ability to visualize how
their direct and indirect social contacts are rating a certain service and how their feedback
influenced the trend in their community. Compared to this, the majority of interviewed
European showed a negative trend towards this factor. In fact, they believe feedback
requests should be straightforward and simple by default and conducting it as a social
activity could make the process of feedback acquisition more discouraging, complex
and distracting from the main purpose which is evaluating a provided software service.
One of the participants said in this regard “I think it should not look more than what it
is supposed to do. It is to get your feedback and not a game to play with”.
Feedback Objectivity and Relevancy. Conducting the feedback acquisition as a

social activity could have a harmful effect on the objectivity and relevancy of the given
feedback by Middle Eastern users. It is perhaps due to users’ engagement with the
activity more than the provided service as well as the burden of social constraints. One
of the participants commented “I imagine I would be nicer about my opinion to people
I know more than others”. Although this might result in given feedback that does not
reflect users’ true opinion about the provided software service, it could also result in a
more socially active users which could be a positive sign to the software and the users’
community. Users’ satisfaction with the social activity should not affect their view about
the software service. This highlights the need to carefully design and apply this factor.
Figure 7 gives a clearer view of how the different cultures’ impact on this social factor
could affect the quality of given feedback.
Fig. 7. Eeffect of feedback acquisition as a social activity in Middle Eastern and European culture.
In general, this paper advocated that culture variations has a noticeable impact on
how Middle Eastern and European users’ behaviour in response to feedback acquisition
and how they are socially motivated to give feedback. The results suggest that having
carfully tailopred persuasive method for the design of an adaptive and cultural-aware
feedback acquisition could highly improve the quality of the collected feedback as well
as users’ satisfactions and response rate to feedback acquisition.
5 Conclusion
In this paper reported on an empirical study that investigated qualitatively the culture
impact on users’ behaviour and motivations to feedback acquisition. The result high‐
lighted key differences between Middle Eastern and Western users with regard to their
social motives to give feedback. Generally, European users found to be more socially-
independent and showed less interest in being socially motivated to give feedback than
Middle Eastern users. The paper also advocates the need to have a persuasive and
culture-aware feedback acquisition which opens the gate for further research in this area.
Acknowledgment. We would like to acknowledge that this research was supported by an FP7
Marie Curie CIG grant (the SOCIAD Project). We would also like to thank the participants who
took part in our study for their valuable input.
References
1. Yunker, J.: Beyond borders: Web globalization strategies. New Riders, Boston (2002)
2. Reinecke, K., Bernstein, A.: Culturally adaptive software: moving beyond
internationalization. In: Aykin, N. (ed.) HCII 2007. LNCS, vol. 4560, pp. 201–210. Springer,
Heidelberg (2007)
3. Nielsen, J., Pfeil, U., Zaphiris, P., Ang, C.S.: International user interfaces Cultural differences
in collaborative authoring of Wikipedia. J. Comput. Mediated Commun. 12(1), 88–113 (1996)
4. Honold, P.: Culture and context: An empirical study for the development of a framework for
the elicitation of cultural influence in product usage. Int. J. Hum.-Comput. Interact. 12(3–4),
327–345 (2000)
5. Abdelnour-Nocera, J., Clemmensen, T., Kurosu, M.: Reframing HCI through local and
indigenous perspectives. Int. J. Hum. Comput. Interact. 29, 201–204 (2013)
6. Ali, R., Solís, C., Omoronyia, I., Salehie, M., Nuseibeh, B.: Social adaptation - when software
gives users a voice. In: Filipe, J., Maciaszek, L.A. (eds.), ENASE, pp. 75–84. SciTePress,
(2012)
7. Almaliki, M., Ncube, C., Ali, R.: The design of adaptive acquisition of users feedback: An
empirical study, RCIS (2014)
8. Fogg, B.J.: Persuasive Technology: Using Computers to Change What We Think and Do.
Morgan Kaufmann, San Francisco (2002)
9. Fogg, B.J., Iizawa, D.: Online persuasion in facebook and mixi: A cross-cultural comparison.
In: Oinas-Kukkonen, H., Hasle, P., Harjumaa, M., Segerståhl, K., Øhrstrøm, P. (eds.)
PERSUASIVE 2008. LNCS, vol. 5033, pp. 35–46. Springer, Heidelberg (2008)
10. Khaled, R.: Culturally-relevant persuasive technology (2008)
11. Marcus, A.: Global/intercultural user interface design. Hum. Comput. Interact. Des. Issues,
Solutions, Appl. 107, 355–381 (2003)
12. Callahan, E.: Interface design and culture. Ann. Rev. Inf. Sci. Technol. 39(1), 255–310 (2005)
13. Aykin, N., Quaet‐Faslem, P.H., Milewski, A.E.: Cultural ergonomics. Handbook of Human
Factors and Ergonomics, Third Edition, 177–190 (2006)
14. Frandsen-Thorlacius, O., Hornbæk, K., Hertzum, M., Clemmensen, T.: Non-universal
usability?: a survey of how usability is understood by Chinese and Danish users. In: SIGCHI
Conference on Human Factors in Computing Systems, ACM (2009)
15. Creswell, J.W., Plano Clark, V.L., Gutmann, M.L., Hanson, W.E.: Advanced mixed methods
research designs. Sage, Thousand Oaks, CA (2003)
16. Morse, J.M.: Approaches to qualitative-quantitative methodological triangulation. Nurs. Res.
40(2), 120–123 (1991)
17. Tashakkori, A., Teddlie, C.: Mixed methodology: Combining qualitative and quantitative
approaches. Sage Publications, Thousand Oaks (1998)
18. Creswell, J.W.: Qualitative Inquiry And Research Design: Choosing Among Five
Approaches, 3rd edn. Sage Publications Inc., Thousand Oaks (2013)
19. Socialclinic: The State of Social Media in Saudi Arabia. The Social Clinic (2013). http://
www.thesocialclinic.com/the-state-of-social-media-in-saudi-arabia-2012-2/
20. Miles, M.B., Huberman, A.M.: Qualitative data analysis: An expanded sourcebook, 2nd edn.
Sage, Thousand Oaks (1994)
21. Suler, J.: The online disinhibition effect. Cyber Psychol. Behav. 7(3), 321–326 (2004)
22. Almaliki, M.: Engineering an Adaptive and Socially-aware Feedback Acquisition. Ph.D.
thesis, Bournemouth University, 88–102 (2015)
Theoretical Reflections
Crowd-Designed Motivation:
Combining Personality
and the Transtheoretical Model
Roelof A.J. de Vries(B) , Khiet P. Truong, and Vanessa Evers
University of Twente, Enschede, The Netherlands

{r.a.j.devries,k.p.truong,v.evers}@utwente.nl
Abstract. Current approaches to design motivational technology for

behavior change focus on either tailoring motivational strategies to indi-
vidual preferences or on implementing strategies from behavior change
theory. Our goal is to combine these two approaches and translate behav-
ior change theory to text messages, tailored to personality. To this end,
we conducted an online survey with 481 participants exploring the rela-
tionship between behavior change theory (the Transtheoretical Model)
and personality in the context of physical activity. Our results show that
(1) people’s personalities correlate with their stage of change and (2)
people’s personalities and their stages of change correlate to preferences
for certain processes of change. We discuss the implications of the results
for designing motivational technology.
Keywords: Behavior change · Persuasion strategies · Personality ·

Processes of change · Stages of change · Transtheoretical model
1 Introduction
Over the past few years, human-computer interaction (HCI) research, in partic-
ular persuasive technology research, has focused on designing motivational tech-
nology, assisting or encouraging people to change their behavior [1]. To increase
the effectiveness of the strategies used in these technologies, researchers aim for
personalization [2]. This can be done by, for example, tailoring the strategies
to certain user characteristics, like personality [3]. Additionally, authors have
been advocating the use of behavior change theory [4] aiming to increase the
effectiveness of the strategies used in these technologies. However, using theory
or models when designing motivational technology for behavior change comes
with a challenge: there is no well-established method to translate theoretical
constructs and insights to persuasive or motivational interaction designs to be
used in practice.
We aim to combine these two approaches to increase the effectiveness of
persuasive and motivational strategies. For our long-term goal, the objective is
to design a smartphone application that motivates users through text messages

c Springer International Publishing Switzerland 2016
DOI: 10.1007/978-3-319-31510-2 4
42 R.A.J. de Vries et al.
to become regular runners by implementing strategies from behavior change

theory or models and tailoring these text messages to personality.
The strategies that we aim to translate and personalize come from the Trans-
theoretical Model (TTM) of behavior change [5]. According to the TTM, behav-
ioral change consists of five stages of change. When moving through these stages,
people encounter processes of change, i.e., experiences and actions that influence
the progression through the stages [6]. Different processes are associated with
different stages of change [7]. For example, rewarding desirable behavior (i.e.,
Reinforcement Management) is expected to be most useful in stages where desir-
able behavior is performed (i.e., in the Action or Maintenance stage), whereas
making someone aware of the risks of undesirable behavior (i.e., Consciousness
Raising) is expected to be most useful in stages where the undesirable behavior
is still performed (i.e., in the Precontemplation, Contemplation or Preparation
stage). We expect that when designing messages that capture a process of change
that fits with the stage of change a person is in, the effectiveness of the messages
will increase. To increase the effectiveness even further, we could account for the
individual preferences people have for persuasive strategies [8,9], for example
based on their personality [10], which is similar to a persuasion profile [11].
In this paper we report on the first step in a novel approach to translate
and personalize strategies from behavior change theory to inform the design
of motivational technology. Using a crowdsourcing platform, we measured the
participants’ stage of change, personalities and preference for a certain process
of change. We show that (1) personality is correlated to the stage of change a
person is in, and (2) personality and stage of change are related to the preference
for a certain process of change.
In the following sections we report on theoretical background, related work,
our expectations, the design and results of the crowdsourcing study with ques-
tionnaires, and we end with a discussion and conclusion.
2 Theoretical Background: TTM and Personality
The Transtheoretical Model (TTM) from Prochaska et al. [5] is a dynamic, inte-
grative behavior change model focused on the individual. The stages of change
associated with the TTM can be practically applied [12] and classify people into
(not necessarily linearly) progressing stages for changing behaviors, i.e., Precon-
templation, Contemplation, Preparation, Action, and Maintenance. While the
stages of change are useful in explaining when changes in cognition, emotion, and
behavior take place, the processes of change help to explain how and why the
progression through these stages occur. Ten covert and overt processes will usu-
ally be experienced when successfully progressing through the stages of change
and attaining the desired behavioral change. The ten processes can be divided
into two groups: Experiential processes and Behavioral processes. Experiential
processes are focused on changing people’s ideas and Behavioral processes are
focused on changing people’s actions, see Table 1 for an overview. The effective-
ness of the processes of change depends on their associated stages of change.
Crowd-Designed Motivation: Personality and the TTM 43
Table 1. The processes of change divided in experiential and behavioral processes with
a short description.
Experiential processes
Consciousness raising (CR): The individual seeks increased knowledge about the
causes, consequences and cures for their problem behavior
Dramatic relief (DR): The individual’s emotions about the problem behavior and
possible solutions are evoked
Environmental reeval. (ER): The impact that the individual’s problem behavior has
on their environment is reevaluated
Social liberation (SOL): Attempts are made to increase alternatives for the
individual’s former problem behavior
Self-reevaluation (SR): Cognitions and emotions regarding the individual with
respect to their problem behavior are reevaluated
Behavioral processes
Self-liberation (SEL): The individual has the belief that he can change and commits
to it by choosing a course of action
Helping relationships (HR): The individual seeks trust and open discussion about
the problem behavior as well as support for the healthy behavior change
Counterconditioning (CC): The individual substitutes positive behaviors for the
individual’s problem behavior
Reinforcement manag. (RM): Steps or changes made by the individual are rewarded
when in a positive direction or punished when in a negative direction
Stimulus control (SC): Stimuli that may cue a lapse back to the problem behavior
are avoided and prompts for more healthier alternatives are inserted
If we could also account for individual preferences, for example those caused by
differences in personality [10], we could increase the effectiveness even further.
Personality is a way to describe long-lasting individual characteristics (simi-
larities and differences) between people. In the psychology literature, a lot of dif-
ferent personality classifications can be found. The most important one is the Big
Five [13] model, also known by its acronym for the personality traits; OCEAN.
This framework classifies people in five dimensions: Openness to experience,
(O), Conscientiousness (C), Extraversion (E), Agreeableness (A) and Neuroti-
cism (N). A well-known measure for this five-factor model is the NEO-PI-R [14].
Although there are other personality theories, we chose to work with this model:
it is practical in use, it is temporally stable [15], and measurement tools (i.e.,
questionnaires) for the model are widely available and well validated [16].
3 Related Work
The use of theory or models, such as the TTM, has been advocated in design-
ing strategies to change behavior (e.g., [17,18]) because this will help evaluate
this theory or model and the use of this theory or model then offers an expla-
nation when the designed intervention strategy does or does not work. Hence,
a theoretical foundation will help in understanding and targeting determinants
of behavior, like the stages of change or personality. However, there is little
guidance on how to apply theory to the design of intervention strategies [18].
In the context of changing or determining physical activity behavior, some
direct relations have been found between physical activity and certain person-
ality traits. Overall, Extraversion and Conscientiousness seem to be positively
correlated with physical activity and exercise behavior [19–21] while Neuroticism
appears to be negatively correlated to physical activity [21].
Research on personality as a factor for tailoring messages to encourage phys-
ical activity has been rather limited. The most closely related works can be
found in [22,23]. In a study by Courneya et al. [22], exercise behavior, motives,
barriers, and preferences were correlated to the Big Five. Among other things,
it was found that: the personality trait Openness was related to the motive of
fun and enjoyment; Conscientiousness was related to the motive of fitness and
health; Extraversion was related to the motive of socializing and meeting peo-
ple, Agreeableness was negatively related to preference for competitive exercises;
and Neuroticism was related to the barriers of lack of energy, lack of motiva-
tion and embarrassment. From these correlations, one could derive guidelines
on how (not) to tailor motivational messages to each personality type. Another
study [24] also addressed the relation between personality and motives in exer-
cise participation. It was found that different personalities have different motives
for change in exercise participation, indicating that people with different per-
sonalities should be motivated in different ways to participate in exercise. Halko
et al. [23] explores the relationship between personality (Big Five) and persua-
sion in the context of health-promotion with mobile applications. Their results
showed that all personality types had different preferences for (mobile) persua-
sive messages (for healthy living). Finally, there have been some studies that
suggest that the stages of change can benefit from personalization [25], that
future research should tailor messages that promote physical activity to peo-
ple’s personalities [26], and that “Individuals with certain personality traits are
more likely to be perceptive toward the idea of physical activities” [3, p. 8]. All
these studies indicate that personality is a decisive factor in explaining the indi-
vidual nature of people and their motivations and barriers for physical activity
participation.
There has been some recent research into the role of personality when
designing tailored persuasive strategies [9,27,28]. For example, Kaptein et al. [9]
developed six persuasive strategies and a questionnaire to measure the user’s
susceptibility to those six persuasive strategies. They tested a setup where they
tailored to the user’s susceptibility versus a contra-tailored setup in the con-
text of snacking and found a greater decrease in the tailored version. In a
study about tailored persuasive messages for advertisement (advertising mobile
phones), Hirsh et al. [27] let participants rate the level of persuasiveness of the
messages that were tailored to each personality type. For example, people with
the Extraversion personality type would receive messages like “With XPhone,
you’ll always be where the excitement is” [27, p. 579] because extraverts are espe-
cially sensitive to rewards and social attention. The results show a clear benefit
in tailoring messages to personality type features. Similar results were obtained
in a study [28] where an application was developed to persuade users to study
more using persuasion strategies that were tailored to users’ personalities: dif-
ferent personalities indeed preferred different (persuasive) study behaviors. All
these studies show promising results and are good examples of how persuasive
strategies and personality targeted design can influence participants and increase
their participation in a HCI context.
4 Expectations
Our long-term goal is to develop technology that motivates people to exercise
and adhere to exercising for a longer period of time such that long-term behavior
change can be accomplished. We argue that motivational text messages tailored
to personality, as a stable, distinctive factor, and framed in behavior change
theory, will contribute to longer-term exercise adherence.
In the current work, we explored the relation between personality and the
stages, and between personality and the stages, and the processes of change. For
the first relation, we expected to find that personality correlates to the stages of
change. More specifically, given that Extraversion and Conscientiousness have
been found to correlate positively to fitness and health [19–22], and Neuroticism
was found to correlate negatively to physical activity [21], we expected to find
a positive correlation for Conscientiousness and Extraversion and a negative
correlation for Neuroticism in relation to stage of change. For the second relation,
we expected to find that different personality traits together with the stages
of change related to different processes of change, but we had no expectation
specifically, about how traits, stages and processes related.
5 Study
Our study was framed as an online crowdsourcing (language-elicitation) task
with questionnaires. The same study was also described in [29] but reports on
different results. In the current paper, we focused on the results of the ques-
tionnaires. The participants were gathered through Amazon Mechanical Turk
(AMT), with a link to SurveyMonkey where the study was hosted. In the ques-
tionnaires, we measured the participants’ personalities, their self-assessed stage
of change and the processes of change.
5.1 Sample
The sample size consisted of 500 people. The data of 19 respondents was excluded
because their questionnaires were incomplete. The final sample included 481
respondents (250 male and 231 female). The minimum age was 18 and the max-
imum was 68. The average age was 31.09 (SD = 9.22) and the median 29.
With respect to education, 201 respondents received some college education, 183
obtained a college degree, 46 obtained a masters degree, 42 completed their high
school, 5 obtained a PhD and 4 received other types of education.
The AMT requirements for the respondents were that they had already com-
pleted >1000 tasks on AMT, >98 % of them were approved successfully and the
respondents were located in the United States. These requirements ensured that
respondents were already familiar with surveys, that they were serious about fill-
ing in the survey (only 19 were not, which is low for online anonymous surveys)
and that they had some proficiency in English.
5.2 Questionnaire Measures
To measure participants’ personality we used the 50-item IPIP representation

of the revised version of Costa and McCrae’s [14] NEO Personality Inventory1
which posed 50 statements (for example, “Make plans and stick to them.”).
Participants were asked to answer how self-descriptive they found these state-
ments (on a 5-point Likert scale, 1 being “very inaccurate” and 5 being “very
accurate”). We used a 1-item stage of change measure for exercise [30] where
participants were given a description of regular exercise and of the five stages
and rated their stage based on that description. Additionally, we used a 30-item
processes of change measure2 for exercise [31] which asked how often certain
experiences or habits (for example, “I feel more confident when I exercise regu-
larly.”) occurred in the last month (each measured by three items, ratings from
1 (never) to 5 (always)).
The reliability of the measures was overall very good (see Table 2 for person-
ality and Table 3 for processes). The only disputable measure was that of Social
liberation, with a Cronbach’s alpha of .63 which we found still acceptable (and
also comparable to other relevant work [7]). Otherwise the reliability scores were
between .75 and .90.
5.3 Procedure
Participants were recruited through AMT. They were informed of their compen-
sation, the goal of the survey experiment and the estimated completion time. On
SurveyMonkey, the goal of this survey was summarized and participants were
asked to complete a consent form. Our study was framed as an online crowd-
sourcing (language-elicitation) task with questionnaires. First, the participants
were given a crowdsourcing language-elicitation task: the description and analy-
sis of this task fall outside the scope of this paper. Second, participants were
asked to fill out questionnaires for personality, stage of change and processes
of change. We address the results of these questionnaires in the current paper.
1
adopted from http://ipip.ori.org/.
2
adopted from http://www.uri.edu/research/cprc/measures.htm.
Table 2. Averages (M), standard deviations (SD), and Cronbach’s alpha’s (α) for
all the construed scales. Scales are added, instead of averaged to keep origin clear.
Personality scales are 10 items with scoring from 1 to 5 added up (possible scores
from 10 to 50). Ordinal regression with stage of change as dependent variable and the
personality traits (OCEAN) as independent variables. (N = 481)
Trait M SD α ratio sig CI

Openness to exp. 38.94 6.65 .77 1.015 .252 [0.989 − 1.042]
Conscientiousness 37.75 7.43 .89 1.025 .074 [0.998 − 1.055]
Extraversion 31.02 8.91 .90 1.050 .000 [1.028 − 1.075]
Agreeableness 38.24 6.14 .80 0.969 .057 [0.939 − 1.001]
Neuroticism 24.61 8.91 .90 0.971 .024 [0.946 − 0.996]
The participants were debriefed about the detailed goals of this survey and
given a completion code to fill in on AMT to receive payment. The survey took
about 45 min to complete. Participants were compensated 3 US dollars for their
participation.
6 Results
Data from 481 participants was analyzed. Important to note is that the self-
assessed stages of change measure was not equally distributed: 175 participants
rated themselves to be in the Maintenance stage (M), 114 in the Preparation
stage (P), 91 in the Action stage (A), 68 in the Contemplation stage (C), and
33 participants rated themselves in the Precontemplation stage (PC).
6.1 Relation Between Personality and Stages of Change

Based on literature we expected that certain personality traits scores signif-
icantly relate to self-assessed stage of change. An ordinal logistic regression
was run to determine the effect of the traits (OCEAN) on the self-assessed
(ordinal) stages of change. The general model (OCEAN) statistically signifi-
cantly predicted the stages of change over and above the intercept-only model,
χ2 (5) = 66.526, p < .001. Concerning the contributing factors, an increase in
Extraversion was associated with an increase in stage of change, with an odds
ratio of 1.051 (95 % CI, 1.028 to 1.075), χ2 (1) = 18.578, p < .001. However,
a decrease in Neuroticism was associated with an increase in stage of change,
with an odds ratio of 0.971 (95 % CI, 0.946 to 0.996), χ2 (1) = 5.091, p < .024.
The other personality traits were not significantly (p < 0.05) related to stage of
change (see Table 2). Overall, the influence of the relations (as expressed in the
odds ratio) is small.
Table 3. Averages (M), standard deviations (SD), and Cronbach’s alpha’s (α) for
all the construed scales. Scales are added, instead of averaged to keep origin clear.
processes of change are 3 items with scoring from 1 to 5 added up (possible scores from
3 to 15). Standardized regression coefficients of personality traits, stages of change and
the processes of change are reported. (N = 481) 1 p < 0.05, 2 p < 0.01, 3 p < 0.001.
PoC M SD α R2 SoC β Oβ Cβ Eβ Aβ Nβ
3 1 2
CR 8.81 3.38 .89 .246 0.398 0.102 0.014 0.151 0.005 0.018
DR 9.10 3.10 .75 .112 0.2783 0.080 0.1271 0.072 -0.068 0.115
ER 10.65 2.88 .75 .071 0.087 0.1602 0.1141 0.072 0.011 0.047
SOL 10.64 2.57 .63 .121 0.087 0.064 0.1472 0.1712 0.1712 0.1431
SR 12.34 2.75 .86 .260 0.3893 0.2253 0.014 -0.010 0.1191 0.007
3 1 2
SEL 10.98 3.07 .82 .476 0.619 0.083 0.112 0.045 0.006 -0.001
3 3
HR 7.81 3.75 .90 .194 0.297 -0.077 0.035 0.203 0.027 -0.046
CC 8.08 3.29 .85 .462 0.5783 -0.005 0.1462 0.1112 0.008 -0.011
RM 11.21 3.11 .84 .310 0.4413 0.1643 0.062 0.058 0.074 0.001
SC 8.30 3.45 .77 .335 0.4713 0.053 0.1151 0.1262 -0.033 -0.015
6.2 Relation Between Personality and Stages and Processes

We expected that the stages of change and different personalities would relate
to different self-assessed processes of change in relation to exercise, but we had
no expectation specifically, about how traits, stages and processes related. We
were interested in understanding the relations between the continuous person-
ality trait variables (O, C, E, A and N) of the compound variable personality
and the continuous variables of the processes of change, which we can assess
with regression coefficients. Hence we carried out multiple regression analyses.
In Table 3 the standardized regression coefficient (β) scores are reported for the
predictor variables stages of change and personality traits (OCEAN) and the
outcome variables of the processes of change (10 times). The regression coeffi-
cient results suggest that different personality traits scores relate differently to
processes of change. All processes are significantly correlated to at least one per-
sonality trait, and all personality traits are significantly related to at least one
process. It should be noted that although there are significant personality-trait-
to-process relations, the stages of change are usually a much larger predictor
(this can be seen from the standardized β reported in Table 3). The personal-
ity trait results could be considered ‘nuances’ to the already existing relation
between stages and processes.
7 Discussion
Using theory in practice is not always easy and effective. Although there is a
general consensus on the value of most behavior change theories and more specif-
ically the Transtheoretical Model, there is also still plenty of room to increase
the effectiveness and salience of such theories by identifying more determinants

(e.g., personality) for specific situations (e.g., the exercise domain) and by reveal-
ing new dependencies between them. As a first step towards combining person-
ality and behavior change theory to motivate people to exercise, we assessed
the possibility of personality-based tailoring of the processes of change through
crowdsourcing and self-assessment measures. We conclude that (1) personality
traits (E and N) relate to the stages of change and (2) personality traits and
the stages of change relate to preferences for certain processes of change. In this
section we discuss the implications of the results separately for each expectation.
7.1 Relation Between Personality and Stages of Change

The results of the study show that people’s personalities are related to their
progression through the different stages of change for exercise behavior, specifi-
cally that the Extraversion trait was positively correlated with people progress-
ing through the stages, while the Neuroticism trait was negatively correlated
with progressing through the stages. In other words, people scoring higher on
Extraversion are more likely to be in higher stages of change, while people scor-
ing higher on Neuroticism are more likely to be in the lower stages of change.
No significant relation was found between Conscientiousness and the stages of
change. Although the correlations of personality traits to the stages of change are
relatively small, this is similar to other research on personality and physical activ-
ity [21] and considered still important for the health context. For Extraversion
and Neuroticism, which are significantly correlated to the stages, one tentative
explanation could be that people change their personality when changing their
behavior. But, because it is believed that personality is temporally stable [15],
a more likely explanation is that people with low Extraversion and high Neuroti-
cism scores need different motivations and see different barriers when trying to
change their behavior (compared to high Extraversion, low Neuroticism scoring
people) then those addressed in current motivational technology and programs
and therefore these people have more difficulties in changing their behavior.
7.2 Relation Between Personality and Stages and Processes

Our study also shows that there are relations between different personality traits
and different processes of change they find important in relation to exercise. Con-
scientiousness is related to six processes, Openness to Experience and Extraver-
sion are related to five, Agreeableness to two, and Neuroticism to one. Interesting
to see is that Neuroticism, which correlated negatively to the stages of change,
does not (significantly) relate to many processes. This could support our previous
interpretation that the processes believed to help people through the stages are
not very appealing to people scoring high on Neuroticism and therefore they also
do not progress through the stages. Similar results with health-promoting strate-
gies for people scoring high on Neuroticism were found in previous work [23].
Conscientiousness, which we expected to relate to the stages of change did not,
but in turn correlated to the most processes. Previous work also suggested
a relation between Conscientiousness and the stages of change, but found that
this was fully mediated by the relation between Conscientiousness and certain
processes [32]. In any case, the results show a relation between different processes
of change and personality traits, which serves as an indication that the tailoring
of processes to personality trait preferences could be very helpful in making the
messages more salient for behavior change.
7.3 Limitations of the Current Study
There were some limitations to the present study. Firstly, we used a cross-
sectional design which does not provide strong evidence for causation, only cor-
relation. A second limitation is that we ran our study on AMT. This could
misrepresent the ‘general’ population, although some studies have reported that
AMT can give very good representation of general society, especially for online
survey standards [33]. A third limitation is the self-assessment nature of the
measures, which asks people to think about what strategies influence them. This
does not necessarily mean these strategies will influence them, or that they will
be influenced by textual representations of these strategies.
8 Conclusion
As part of a larger study, we sought to leverage certain HCI practices, like crowd-
sourcing, to explore theory on behavior change from psychological research to
come to useful and practical insights about how to further adapt the processes
of change to (robust) user characteristics (e.g., personality traits). We identified
new dependencies between the different processes, stages and personality traits
in the context of the exercise domain. And we made a first step in translating
theoretical constructs and principles of behavior change theories to information
structures and interaction designs. These findings can help inform developers of
motivational and persuasive technology who want to use the TTM as a founda-
tion for long-term behavior change and who want to use personality to tailor to
individuals. Concretely, we argue that: (1) when designing for behavior change,
one should take into account the relation between someone’s personality and
the stages of behavior change; (2) when designing for more than one-size-fits-all,
one should take into account the relation between someone’s personality and the
preference for different processes of change.
In future research, we will address the limitations of the current study (see
Sect. 7.3), and carry out a long-term in-the-wild study to look into the effective-
ness of personality-tailored behavior change messages.
Acknowledgments. This research was funded by COMMIT/ and is part of the P3

project SenseI: Sensor-Based Engagement for Improved Health. We would like to thank
Maartje de Graaf for her input.
References
1. Hekler, E.B., Klasnja, P., Froehlich, J.E., Buman, M.P.: Mind the theoretical gap:
interpreting, using, and developing behavioral theory in HCI research. In: Pro-
ceedings of the SIGCHI Conference on Human Factors in Computing Systems,
PP. 3307–3316. ACM (2013)
2. Noar, S.M., Benac, C.N., Harris, M.S.: Does tailoring matter? meta-analytic review
of tailored print health behavior change interventions. Psychol. Bull. 133(4),
673–693 (2007)
3. Arteaga, S.M., Kudeki, M., Woodworth, A., Kurniawan, S.: Mobile system to moti-
vate teenagers’ physical activity. In: Proceedings of the 9th International Confer-
ence on Interaction Design and Children, pp. 1–10. ACM (2010)
4. Chatterjee, S., Price, A.: Healthy living with persuasive technologies: framework,
issues, and challenges. J. Am. Med. Inform. Assoc. 16(2), 171–178 (2009)
5. Prochaska, J.O., DiClemente, C.C.: Stages and processes of self-change of smoking:
toward an integrative model of change. J. Consult. Clin. Psychol. 51(3), 390 (1983)
6. Spencer, L., Adams, T.B., Malone, S., Roy, L., Yost, E.: Applying the transtheo-
retical model to exercise: a systematic and comprehensive review of the literature.
Health Promot. Pract. 7(4), 428–443 (2006)
7. Marcus, B.H., Rossi, J.S., Selby, V.C., Niaura, R.S., Abrams, D.B.: The stages
and processes of exercise adoption and maintenance in a worksite sample. Health
Psychol. 11(6), 386 (1992)
8. Lacroix, J., Saini, P., Goris, A.: Understanding user cognitions to guide the tailoring
of persuasive technology-based physical activity interventions. In: Proceedings of
the 4th International Conference on Persuasive Technology, p. 9. ACM (2009)
9. Kaptein, M., De Ruyter, B., Markopoulos, P., Aarts, E.: Adaptive persuasive sys-
tems: a study of tailored persuasive text messages to reduce snacking. ACM Trans.
Interact. Intell. Syst. (TIIS) 2(2), 10 (2012)
10. Alkış, N., Temizel, T.T.: The impact of individual differences on influence strate-
gies. Personality Individ. Differ. 87, 147–152 (2015)
11. Kaptein, M., Markopoulos, P., de Ruyter, B., Aarts, E.: Personalizing persuasive
technologies: Explicit and implicit personalization using persuasion profiles. Int. J.
Hum. Comput. Stud. 77, 38–51 (2015)
12. Nigg, C.R., Geller, K.S., Motl, R.W., Horwath, C.C., Wertin, K.K., Dishman, R.K.:
A research agenda to examine the efficacy and relevance of the transtheoretical
model for physical activity behavior. Psychol. Sport Exercise 12(1), 7–12 (2011)
13. Goldberg, L.R.: The development of markers for the big-five factor structure. Psy-
chol. Assess. 4(1), 26 (1992)
14. Costa, P.T., McCrae, R.R.: Normal personality assessment in clinical practice: The
NEO personality inventory. Psychol. Assess. 4(1), 5 (1992)
15. Rhodes, R.E., Courneya, K.S., Jones, L.W.: Personality and social cognitive influ-
ences on exercise behavior: adding the activity trait to the theory of planned behav-
ior. Psychol. Sport Exerc. 5(3), 243–254 (2004)
16. McCrae, R.R., Costa Jr., P.T.: A five-factor theory of personality. Handbook of
personality: Theory and research 2, 139–153 (1999)
17. Cole-Lewis, H., Kershaw, T.: Text messaging as a tool for behavior change in
disease prevention and management. Epidemiol. Rev. 32(1), 56–69 (2010)
18. Michie, S., Johnston, M., Francis, J., Hardeman, W., Eccles, M.: From theory to
intervention: mapping theoretically derived behavioural determinants to behaviour
change techniques. Appl. psychol. 57(4), 660–680 (2008)
19. Gallagher, P., Yancy, W.S., Denissen, J.J.A., Kühnel, A., Voils, C.I.: Correlates
of daily leisure-time physical activity in a community sample: Narrow personality
traits and practical barriers. Health Psychol. Official J. Div. Health Psychol. Am.
Psychol. Assoc. 32(12), 1227–1235 (2013)
20. Hoyt, A.L., Rhodes, R.E., Hausenblas, H.A., Giacobbi, P.R.: Integrating five-factor
model facet-level traits with the theory of planned behavior and exercise. Psychol.
Sport Exerc. 10(5), 565–572 (2009)
21. Rhodes, R.E., Smith, N.E.I.: Personality correlates of physical activity: a review
and meta-analysis. Br. J. Sports Med. 40(12), 958–965 (2006)
22. Courneya, K.S., Hellsten, L.A.M.: Personality correlates of exercise behavior,
motives, barriers and preferences: An application of the five-factor model. Per-
sonality Indiv. Differ. 24(5), 625–633 (1998)
23. Halko, S., Kientz, J.: Personality and persuasive technology: an exploratory study
on health-promoting mobile applications. In: Persuasive technology, pp. 150–161
(2010)
24. Ingledew, D.K., Markland, D.: The role of motives in exercise participation. Psy-
chol. Health 23(7), 807–828 (2008)
25. Ferron, M., Massa, P.: Transtheoretical model for designing technologies support-
ing an active lifestyle. In: Proceedings of the Biannual Conference of the Italian
Chapter of SIGCHI, p. 7. ACM (2013)
26. Latimer, A.E., Brawley, L.R., Bassett, R.L.: A systematic review of three
approaches for constructing physical activity messages: what messages work and
what improvements are needed? Int. J. Behav. Nutr. Phys. Act. 7, 36–53 (2010)
27. Hirsh, J.B., Kang, S.K., Bodenhausen, G.V.: Personalized persuasion: tailoring
persuasive appeals to recipients’ personality traits. Psychol. Sci. 23(6), 578–581
(2012)
28. Adnan, M., Mukhtar, H., Naveed, M.: Persuading students for behavior change by
determining their personality type. In: 2012 15th International Multitopic Confer-
ence (INMIC), pp. 439–449 (2012)
29. De Vries, R.A.J., Truong, K.P., Kwint, S., Drossaert, C.H.C., Evers, V.: Crowd-
designed motivation: Motivational messages for exercise adherence based on behav-
ior change theory. In: Proceedings of the SIGCHI Conference on Human Factors
in Computing Systems. ACM (in press, 2016)
30. Norman, G., Benisovich, S., Nigg, C., Rossi, J.: Examining three exercise staging
algorithms in two samples. In: 19th Annual Meeting of the Society of Behavioral
Medicine (1998)
31. Nigg, C., Norman, G., Rossi, J., Benisovich, S.: Processes of exercise behavior
change: Redeveloping the scale. Ann. Behav. Med. 21, S79 (1999)
32. Bogg, T.: Conscientiousness, the transtheoretical model of change, and exercise: a
neo-socioanalytic integration of trait and social-cognitive frameworks in the pre-
diction of behavior. J. Pers. 76(4), 775–802 (2008)
33. Mason, W., Suri, S.: Conducting behavioral research on amazon’s mechanical turk.
Behav. Res. Methods 44(1), 1–23 (2012)
The EDIE Method – Towards an Approach
to Collaboration-Based Persuasive Design
Sandra Burri Gram-Hansen ✉

( )
Department of Communication and Psychology, Aalborg University, Aalborg, Denmark

[email protected]
Abstract. This paper presents the initial steps towards a collaboration-based

method for persuasive design – the EDIE method (Explore, Design, Implement,
Evaluate). The method is inspired by Design-Based Research, but developed to
combine different design approaches that have dominated the persuasive tech‐
nology field over the past decade. The rhetorical notion of Kairos is considered
a key element in the EDIE method, resulting in a distinct focus on participatory
design and constructive ethics. The method is explained through a practical
example of developing persuasive learning designs in collaboration with the
Danish Ministry of Defence and the Installation Management Command.
Keywords: Persuasive design · Energy and environmental behaviour · Sustainability ·

Design-Based research · Participatory design · Constructive ethics · Persuasive design
framework
1 Introduction
In this paper, the initial steps towards a collaboration-based method for persuasive
design is presented and discussed. The framework is based upon the Design-Based
Research Methodology (DBR) [1], but distinguishes itself by combining different
context- and system-oriented approaches from the persuasive technology field, and by
enabling ethics to be applied constructively throughout the design process. The method
constitutes some of the results found by applying DBR when exploring the potential of
persuasive learning designs in the Danish army.
Over the past decade, different methods to persuasive design have been discussed
and exemplified, as researchers have taken either a primarily system-oriented [2] or
primarily user-centred approach to the design process [3, 4]. Simultaneously, the theo‐
retical underpinning of the notion of persuasive technologies has been explored and
developed through both practical exemplifications and thorough literary investigations
[5, 6]. Across these different approaches the rhetorical notion of Kairos [7] continuously
establishes itself as a key concept to persuasion, both from a narrow system-oriented
angle and a wider more contextual perspective. From the foundation of this classical
rhetorical understanding of timing and appropriateness, it has been argued that persua‐
sive initiatives are only successful when applied within the intended use context,

DOI: 10.1007/978-3-319-31510-2_5
54 S.B. Gram-Hansen
that the notion of persuasion and subsequently persuasive technologies hold a strong
ethical demand, and that a user-centred approach to design is a requisite to successful
persuasive designs [6].
Within the persuasive technology research field, several researchers have argued that
Value Sensitive Design and Participatory design are particular relevant to the develop‐
ment of persuasive designs [10–12], partly due to the ethical implications related to
designing for behaviour change, and partly in respect to the rhetorical notion of Kairos.
Kairos sums up the concept of the opportune moment, in consideration of time, place
and manner in which a persuasive initiative is taken. Kairos constitutes both a narrow
understanding of the appropriate time for something to take place, and a wider more
philosophical understanding of appropriateness within an intended use context. In order
for designers to grasp this opportune moment, they must strive to create an appropriate
balance between the persuasive initiative and the intended use context. [7]. Whilst both
the temporal and location-based perspectives of Kairos may possibly be defined and
formalised without user involvement, detecting the appropriate manner is dependent on
user participation [6]. As designers we may share our ideas regarding new approaches
to digital learning, but only those who understand the practice of the intended use context
can determine whether or not our ideas may be both applicable and persuasive.
This paper presents the argument that the system-oriented and user-centred approaches,
which have been developed and/or discussed in relation to persuasive technology, may be
constructively combined in the development process. Considering Kairos throughout the
design process facilitates this bridge between design approaches, leading to the initial
version of the EDIE method - a collaboration-based persuasive design approach.
2 The EDIE Method
As previously stated the Edie method is inspired by and developed in accordance with
the DBR methodology - an iterative process for developing, testing, evaluating, and
refining design solutions. As a research methodology it is most often applied in the
context of learning and education, but the applicability of the approach is not restricted
to this domain. By design based, it is understood that research is conducted as designs
are developed and implemented in practice and that the research is oriented towards
exploring best practices or generating directions for future design processes [1]. Partic‐
ular attention is directed towards DBR’s consistent focus on stakeholder collaboration
and the mentioned focus on the intended use context. By maintaining yet adjusting this
constituent, both the wider notions of Kairos and the ethical demand related to persua‐
sion, may be taken into consideration.
The EDIE method is a collaboration-based persuasive design methodology, consti‐
tuted by the four phases Explore, Design, Implement, Evaluate (EDIE) visualised in
Fig. 1. The method aims to facilitate a bridge between the human-centred and the system-
centred approaches, which have dominated the persuasive technology field. Rather than
focusing solely on either the user-centred or the system-oriented aspects of persuasive
design, the method provides directions for designers to include both perspectives in the
design process, and to let the different perspectives constructively influence each other.
The EDIE Method – Towards an Approach 55
Fig. 1. The EDIE method - a collaboration-based persuasive design method
The different steps in the process, enable designers to consider and apply aspects of
e.g. persuasive system design [13] and ambient persuasion [14], while consistently
considering the intended use context and applying ethics constructively throughout the
design process. Thereby, both the wider and the narrow perspectives of Kairos are
acknowledged and considered.
2.1 Basic Distinctions
Whilst DBR and similar approaches are iterative both in the understanding that all four
steps can be repeated several times, and internally within the test and refinement phase,
it is predominantly a research methodology, and iterations will expectedly come to an
end once the research questions have been answered. Persuasion on the other hand can
be understood as a continuous process of shaping, reinforcing or changing the behaviour
of users [2, 15], as users are constantly influenced not only by the persuasive system
with which they are interacting, but also by changes to their use context and their
community of practice. From a persuasive design perspective, it is important to acknowl‐
edge that when a technology is added to a context, whether this be within the physical
56 S.B. Gram-Hansen
or digital realm, the context and the user’s perception of the context is altered [6]. In
order for this to be taken into consideration when applying the the EDIE method, iter‐
ations must continuously take place both on a wider scale and within the different phases.
Furthermore, the EDIE method distinguishes itself from DBR by including ethical
reflections throughout the design process. One of the distinct claims of persuasive
design, may be seen as a strong ethical demand, calling for ethical reflections not only
regarding the overall persuasive intention, but also regarding the appropriateness of the
different persuasive initiatives [12]. The DBR approach already emphasises the impor‐
tance of stakeholder collaboration, but does not have a distinct focus on applied ethics.
The EDIE method on the other hand, considers ethics both within the individual phases
of the process, and more importantly, transfers ethical considerations from one phase to
another. Furthermore, participatory design is applied throughout the four phases of the
EDIE method, in order to ensure acknowledgment of domain experts, and to enable the
ethical considerations to be constructive for design process.
As mentioned above, persuasion can be defined as the intent to shape, reinforce or
change attitudes and behaviours. The persuasive intent will most likely spring from an
already established need for behavioural influence, and as a result, the first step of the
EDIE method distinguishes itself from DBR by a more exploratory approach to gaining
an understanding of the domain and potentially determine why the intended behaviour
does not already take place.
The design and implementation phases are distinguished by the inclusion of theories
and methods with specific relevance to persuasive design. Evaluation in the fourth phase
however, is directed towards establishing the effect of the persuasive initiatives, rather
than determine best practice or design directions.
In consideration of the aforementioned influence that technologies can have on the
context, new iterations must start by again exploring the intended use context in order
to identify significant changes. This particular distinction is visualised by the on-going
spiral in Fig. 1.
3 Applying the EDIE Method in Practice
The Danish Military is a highly complex organisation with a combination of armed

forces, technical staff and civilian employees. As such, employee tasks and work facili‐
ties differ greatly as 33% of the military employees are office workers, whilst others
rarely work indoors.
In order to optimise and standardise learning material across Danish Military Estab‐
lishments, an increasing amount of learning material is digitalised and distributed
centrally. One of the topics, which have been developed as a digital resource, is material
regarding environmental legislation and appropriate environment behaviour in the
Danish military. To ensure that the information is fully disseminated, all employees are
required to complete the e-learning course “The Environmental Driver’s License” (EDL)
with a minimum of 80% correct answers to the included questions.
Whilst EDL does ensure standardised dissemination of environment material, this
traditional e-learning course is nonetheless faced with several fundamental challenges:
• Far from all Danish military establishments are equipped with sufficient digital
resources to enable the majority of their employees to attend an e-learning course
• The learning material presented in EDL does not differentiate between the different
segments of employees in the Danish Military
Whilst the practical issue regarding insufficient digital resources may easily be
solved, the problems related to lack of differentiation in the learning material constitutes
a greater challenge. When all employees are required to relate to all the learning material,
the topic may be perceived as less relevant to the individual employee, thus they may
be lead to believe that the EDL is generally meant for someone else in the organisation.
The lack of relevance to individual employees may potentially have negative conse‐
quences both in relation to transferability of the knowledge acquired (it may be difficult
to transfer the content of EDL to practical tasks at the military establishments), and also
to the long-term effect of the e-learning course.
The aforementioned problems related to EDL formed the basis of a new approach
to environment education, initiated by the Installation Management Command (IMC)
in 2014. In order to not just inform employees about environmental requirements in the
Danish Military, but to also motivate a more environment friendly attitude and behaviour
in the organization, IMC initiated the development of a prototype for environmental
education which distinctively draws upon motivational elements, and which reforms the
notion of digital learning within the organization.
The persuasive learning design discussed in this paper aims to educate the drafted
recruits in appropriate waste management and action in case of accidents. The central
element in this persuasive learning design is the prototype of a location- and situation-
based game Acttention, which has been designed and developed specifically to meet the
requirements of the Danish Military. As such, the content is targeted directly towards
the drafted recruits, whilst the game is designed to be flexible and easily adjusted to fit
the needs of other employee groups. In the following sections, the individual steps of
the EDIE method are elaborated and exemplified in practice.
3.1 Phase 1 - Explore

In order to gain a better understanding of the intended use context, and to define any
particular requirements, that needed to be taken into consideration, Phase 1 constituted
a thorough analysis of intended use context and the existing solutions.
The intended use context was explored through a combination of field visits and
observation studies at the different army bases and supported by wider segmentation
and target group analysis.
The segmentation analysis was based on a quantitative evaluation of the army
employees’ attitude towards environmental education in the organization [16]. From
the analysis it was clear that whilst a majority of the respondents find energy and
environmental considerations important, they do not consider the subject a primary
concern to their function in the army. Moreover, the analysis concluded that 20% of
the respondents were hesitant or even reluctant towards the subject of energy and
environmental education within the organization. As such, the analysis underpinned
58 S.B. Gram-Hansen
the relevance of persuasive design in the case due to the understanding learning takes
place in the learner, and is dependent on his or her motivation towards the subject.
Consequently, traditional learning designs would likely prove to be insufficient.
In order to accommodate the requirement that the learning design should be appli‐
cable across different army bases, two very different army establishments were selected
as test locations for the pilot project, and field visits and in-situ interviews were
conducted at both locations. Observation studies and similar ethnographic methods are
often applied in relation to participatory design, and also exemplified by Davis when
she argued towards the potential of participatory design in relation to persuasive tech‐
nologies [4]. However, ethnography and similar approaches to understanding the
intended use context may hold even more potential, if considering the rhetorical notion
of Kairos a key concept to persuasion. Ethnography may facilitate the understanding of
the contextual perspective, which also sums up the three dimensions of Kairos. Addi‐
tionally, as the field research was done in collaboration with all primary project partic‐
ipants, the experiences shared formed the initial basis of a constructive ethical approach
throughout the design process [12].
The field research was followed up by a series of creative workshops attended by
representatives from the army, the navy and the air force, as well as from the IMC and
the Danish Ministry of Defence. Apart from providing valuable insights regarding the
basis of the mentioned reluctance towards the energy and environmental learning
designs, the workshop also aimed to establish a mutual understanding between the
participants. Thereby, the workshop also served as an initial step towards facilitating
the collaboration between stakeholders throughout the design process – an outcome that
was considered particularly important when working in a context with strong and evident
hierarchical structures. Moreover, the workshops were planned and executed in consid‐
eration of ontological ethics, thus strengthening the constructive influence of ethics [12].
The workshop participants applied LEGO and Play-Dough as they created proto‐
types of future green army barracks, leading to a number of specific requirements for
the architects to take into consideration. As the workshop progressed, it furthermore
constituted a significant contribution to the context exploration, as they provided a
number of rich explanations to practice within the intended use context. Amongst the
specific requirements uncovered during the workshop was the need for environmental
education to be included in other activities. Part of the reluctance towards the subject is
related to a well-known challenge in educational systems – that there is much to teach
and very little time to do it.
In order to also include evaluations of the existing technical solution, the PSD model
[13] was applied to analyse and evaluate the EDL. The PSD framework facilitated a
deeper understanding of the existing learning design, and helped initiate the discussion
of the potential of persuasive principles in learning technologies. The system analysis
led to the conclusion that the EDL failed to reduce and tailor the complex learning
material. The analysis clarified that one of the primary challenges of the EDL was the
intent to educate and motivate all employees through the same system. As mentioned,
this not only constituted a practical problem, it also imposed an educational challenge
as people are rarely motivated by material that they do not find relevant. For instance,
smokers who do not wish to stop smoking are seldom influenced by the many campaigns
regarding the subject, whereas smokers who have an endogenous wish to quit smoking
may be able to find help and support through the different initiatives. As a result, it was
decided that the prototype should focus on a specific group of employees (drafted recruits
in the Danish Army), but that the design should be scalable in way that would enable it
to be transferred to other groups of employees a later time.
3.2 Phase 2 - Design
Based on the information gathered from the analysis of practical problems, steps were
taken towards the development of a learning design that would both provide information
regarding appropriate environmental behaviour and motivate the learners to change both
their attitude and their behaviour. As was the case in the first phase, the design phase
was also conducted as a participatory design process. At this stage, the design group
was narrowed in and consisted of representatives from the IMC, researchers, and tech‐
nical developers with a background in experience design. All participants had been
involved in the first phase of the process, ensuring that not only practical but also ethical
considerations were brought along into this second phase.
In the acknowledgement that the employees do not see energy and environmental
issues as core focus areas, the persuasive learning design was developed with an intent
to influence the community of practice in the Danish Military. By doing so, the design
distinguished itself from traditional approaches to environmental education within the
army, by recognising that the learners do not consider the learning content important.
With a specific focus on army recruits, it was determined that traditional e-
Learning solutions such as the EDL would not be suitable, partly due to the age group
(18–20 year olds), and partly due to the daily practice of this particular segment.
Recruits spend most of their working hours outdoors, and the segment consists of a
highly diverse group of learners, where some may be on their way to university,
whilst others will have struggled all through school. Physical training is a high
priority for the recruits, as many of them are in poor physical condition when they
enlist. Competition is high within the segment, and training often includes an element
of behaviourism, where good results are rewarded. As a result, the design moved
towards a location-based game, which could be implemented as part of the physical
training.
The initial design proposal presented by the technical developers consisted solely of
the Acttention game. It was designed as a location-based game where the recruits would
interact with touch screens, be presented with waste management related questions, and
then sent to a relevant location on the Military base in order to link the content of the
question to a specific location. However, very early in the design process, this solution
was found inadequate both from a learning perspective and from a persuasive perspec‐
tive. As a learning technology, the game lacked information regarding the appropriate
action in different situations; information which was a necessity if the game was to
facilitate learning, rather than merely be a quiz game with a 33% chance of giving a
correct answer. From a persuasive design perspective, the game lacked transparent
persuasive intentions both with regards to the in-game situation and with regards to the
long-term effect. There was no apparent link between the individual components of the
game, and an overall persuasive intention.
60 S.B. Gram-Hansen
In order to address this issue, an instructional film based on the existing content from
EDL, was created to ensure that all recruits were presented with knowledge regarding
appropriate waste management. Consequently, the game was transferred from being a
learning game to becoming a tool for motivating reflections regarding appropriate
behaviour. A series of different scenarios with different solutions were implemented,
designed to simulate situations the recruits might find themselves in when needing to
recall their knowledge about appropriate waste management.
Persuasive design was considered both in the wider and more contextual aspect of
implementing a new technology within a context, and in the specific design of the game.
From a contextual perspective, the combination of specifically tailored learning material
and the distinct respect towards the general attitude that environmental education is
regarded as low priority, both became motivational elements in the learning design.
Also, the instruction film included examples from both the army bases and civilian life,
in order to highlight that the information provided by the film was relevant not only in
relation to personnel working on Danish army bases. By doing so, the instruction film
aimed to address one of the long-term goals of the Danish drafting procedure – educating
valuable members of society.
System specifically, persuasive design was taken into consideration with regards to
the user interaction with the touch screens and more specifically the feedback offered
by the system. In order to ensure that the Acttention game facilitated learning without
requiring too much attention, user feedback was provided as a combination of brief test
messages (e.g. right answer, wrong code), and supported by brightly coloured icons.
Furthermore, the correct answer was highlighted in green while wrong answers were
highlighted in red. Research conducted by Ham and Midden has shown that whilst users
have no relation to other colour codes such as blue/yellow, the combination of red/green
is subconsciously understood as positive and negative [14].
3.3 Phase 3 - Implement

Prior to testing the design with recruits, a pre-test was conducted at one of the designated
army bases, by a group of university students from a Masters program in Information
Architecture. None of the students had any knowledge about practice in the army, but
their feedback was considered significant as their education enables them to provide
very specific system-oriented evaluations. As such, this test enabled us to test the func‐
tionality of the game within the intended use context, and at the same time incorporate
some final adjustments to the design. For instance, the students found that the red/green
colour-based feedback should be extended, and that the possibility to skip the feedback
should be removed if the learner had solved a scenario incorrectly.
The first iteration of tests with recruits took place in November 2014 at Almegaard
army base on the small Danish island Bornholm, and involved 62 recruits representing
three platoons. This particular army base distinguishes itself by being one of the bases,
in which the physical facilities are preserved as listed buildings although the buildings
are still at use.
Almegaard is also one of the smaller army bases in Denmark, which means that most
locations, such as recruit quarters, garage area and environment area are less than 1 km
apart and can be reached on foot. Other army bases are much larger and the different
game related locations might be located so far apart that they cannot all be included. As
such, Almegaard served well as a location for the first test iteration, as it enabled a test
of the full range of the game.
The test took place over three days, thus enabling the design team to iteratively
evaluate and adjust from day to day. Adjustments to the setup were done in careful
consideration of data comparability, but whilst the overall learning design remained the
same, some changes were made in order to ensure higher stability in the game. Time
wise, it would have been possible to conduct all three tests during one day. However to
also exemplify the design’s ability to adapt to the practice within the intended use
context, the trial was included as an element within a three day course which besides
from waste management also included first aid and fire and rescue.
For each individual test of the learning design, the participating recruits met in the
army base auditorium, where they were informed about the overall plan for the event as
well as the evaluation activities that would take place during their participation in the
Acttention learning design.
The recruits then watched the short 15-minute instruction film before heading out to
the Acttention game base. Here they were given further information about how to play
the Acttention game, and divided in to smaller teams of 5–6 members. The Acttention
game is run via three touch screens and one administration screen, and these were set
up in a sheltered location with easy access to the surrounding perimeters. The recruits
were provided with individual chip bracelets, which enabled them to register their indi‐
vidual interaction with the system, thereby ensuring that all recruits were presented with
all the learning scenarios in the game.
Whilst a member of the design team gave the introduction to the course and evalu‐
ation, the commissioned officers were present and engaged throughout the process. Their
involvement both served as a way of introducing them to the learning design in practice,
thereby ensuring that they can handle future use of the system themselves, but it also
served as a way to ensure credibility of the learning design in the eyes of the participating
recruits and as an empowerment of the commissioned officers.
As the terms of the game and the team competition was explained, the recruits’
seriousness was replaced with a more playful and relaxed attitude – still without losing
focus on what was being required of them. The team spirit and the more relaxed and
collaborative attitude was facilitated further by elements such as selecting a team name
and enabling the system to tailor the touch screens to the different teams.
3.4 Phase 4 - Evaluate

The evaluation phase was specifically designed to evaluate both the learning outcome
and the persuasive potential of the design. From a design perspective there was a high
interest in evaluating if the persuasive learning design had motivated a more positive
attitude towards environmental education amongst the participants, however, as the
design had been developed within a learning context, it was also vital to ensure that the
recruits had not only enjoyed themselves but also acquired the necessary knowledge.
The evaluation methods combined qualitative and quantitative methods, in order
62 S.B. Gram-Hansen
to ensure that both the learning potential and the persuasiveness of the design were
assessed through the same evaluation process. In practice, this included questionnaires,
observation studies and in situ interviews
Preliminary findings indicated that the combination of instructional film and loca‐
tion-based learning facilitated a transition from individual learning to collaborate
learning. This blended learning approach and in particular the collaborative activities in
the team-based learning game, ensured that all members of the platoon were able to
engage in the learning experience and reflect upon the learning material.
Results from the first test showed that more than 71% of the recruits completed with
results that meet or exceed the expectations of the Defence Academy. From a persuasive
perspective, the learning design was well received by the recruits, who found it partic‐
ularly positive to be physically active while learning, and who found the competitive
element to be highly motivating. After completing the game, 78% of the recruits indi‐
cated that they found the learning material relevant and useful, and 81% indicated that
learning about appropriate waste management had been fun. In light of the previously
mentioned negativity towards the subject, the feedback regarding the learning experi‐
ence was perceived as highly positive.
Moreover, the design enabled persuasion to take place at different user levels. Whilst
the recruits may have been persuaded by engaging in the Acttention learning design, the
commissioned officers and instructors were motivated and persuaded by a combination
of experiencing Acttention in practice during the test iterations and more importantly
having some sense of ownership of the learning design. To a great extent, this can be
credited to the participatory design activities that have facilitated the entire development
process. Feedback from the involved instructors and officers stated that it was a
welcomed change to traditional communication within the organisation to see that their
domain knowledge and input was distinctively taken into consideration in this new
approach to digital learning. Both identified in the overall design decision to implement
environmental education into a physical training session, and also specifically within
the different scenarios.
4 Final Reflections - Future Research
Through the practical application of the EDIE method, it has been possible to combine
context- and system-oriented approaches to persuasive design, whilst including ethical
considerations throughout the design process. The consistent focus on participatory
design facilitates both a deeper understanding of the intended use context, and the inclu‐
sion of ethical reflections in relation to persuasive designs.
By ensuring a continuous engagement by both designers and instructors, the meth‐
odology distinguishes itself from other similar design approaches, where participatory
design most often takes place with a group of domain experts and only few representa‐
tives of the design team. In the acknowledgement that persuasive design is highly context
dependent, the methodology presented in this paper suggests that all members of the
design team must share experiences within the intended use context in order to partake
in the design process. Not only to ensure that all members of the team acquire sufficient
knowledge about the intended use context, but also to ensure that crucial information is
transferred from one phase of the methodology to another. In consideration of Kairos,
it may be argued that only users, who are already a part of the intended use context, fully
understand what may be appropriate within that context. Consequently, the individual
members of the design team’s different interpretations of the intended context, all
become equally important pieces of a puzzle, which they must participate in putting
together.
Furthermore, the consistent attendance of all designers ensured that a mutual under‐
standing was established between them, based on the interactions that took place and
the experiences that were shared. Several researchers have argued that ethics is a distinc‐
tive aspect of persuasive design. When taking ontological ethics into consideration, this
persuasive design methodology facilitates that ethics becomes constructive for the entire
design process.
Future research will involve a thorough clarification of the different phases of the
EDIE method, so that they more precisely incorporate the different aspects of persuasive
design. In 2016 the framework is to be applied in two individual cases equally complex
to the one presented in this paper. From this it is expected that the individual phases of
the approach will be more distinctly described, as will the specific considerations related
to applying ethics constructively in the design process.
Acknowledgements. The practical exemplification presented in this paper was arranged and
conducted in collaboration with Thilde Møller Larsen from The Defence Installation Management
Command, Denmark
References
1. Reeves, T.C.: Design research from a technology perspective. Educ. Des. Res. 1(3), 52–66
(2006)
2. Oinas-Kukkonen, H., Harjumaa, M.: A systematic framework for designing and evaluating
persuasive systems. In: Oinas-Kukkonen, H., Hasle, P., Harjumaa, M., Segerståhl, K.,
Øhrstrøm, P. (eds.) PERSUASIVE 2008. LNCS, vol. 5033, pp. 164–176. Springer,
Heidelberg (2008)
3. Lockton, D., Harrison, D., Stanton, N.A.: The design with intent method: a design tool for
influencing user behaviour. Appl. Ergonomics 41(3), 382–392 (2010)
4. Davis, J.: Generating directions for persuasive technology design with the inspiration card
workshop. In: Ploug, T., Hasle, P., Oinas-Kukkonen, H. (eds.) PERSUASIVE 2010. LNCS,
vol. 6137, pp. 262–273. Springer, Heidelberg (2010)
5. Torning, K., Oinas-Kukkonen, H.: Persuasive system design: state of the art and future
directions. In: Proceedings of the 4th International Conference on Persuasive Technology.
ACM, Claremont (2009)
6. Gram-Hansen, S.B., Ryberg, T.: Persuasion, learning and context adaptation. Special Issue
Int. J. Conceptual Struct. Smart Appl. 1(2), 28–37 (2013)
7. Kinneavy, J.L.: Kairos in classical and modern rhetorical theory. In: Sipiora, P., Baumlin,
J.S. (eds.) Rhetoric and Kairos, Essays in History, Theory and Practice. State University of
New York Press, Albany (2002)
64 S.B. Gram-Hansen
8. Amiel, T., Reeves, T.C.: Design based research and educational technology: rethinking
technology and the research agenda. Educ. Technol. Soc. 11, 29–40 (2008)
9. Gram-Hansen, S.B., Ryberg, T.: Acttention – influencing communities of practice with
persuasive learning designs. In: MacTavish, T., Basapur, S. (eds.) PERSUASIVE 2015.
LNCS, vol. 9072, pp. 184–195. Springer, Heidelberg (2015)
10. Davis, J.: Design methods for ethical persuasive computing. In: Proceedings of the 4th
International Conference on Persuasive Technology. ACM, Claremont (2009)
11. Davis, J.: Towards participatory design of ambient persuasive technology. In: Proceedings
of Pervasive 2008 Workshop, Australia, Sydney (2008)
12. Gram-Hansen, S.B., Ryberg, T.: From participatory design and ontological ethics, towards
an approach to constructive ethics. In: Ethicomp 2015, ACM SIGCAS, Leicester (2015)
13. Oinas-Kukkoen, H., Harjumaa, M.: Persuasive systems design: key issues, process model,
and system features. Commun. Assoc. Inf. Syst. 24, 28 (2009)
14. Ham, J., Midden, C.: Ambient persuasive technology needs little cognitive effort: the differential
effects of cognitive load on lighting feedback versus factual feedback. In: Ploug, T., Hasle, P.,
Oinas-Kukkonen, H. (eds.) PERSUASIVE 2010. LNCS, vol. 6137, pp. 132–142. Springer,
Heidelberg (2010)
15. Miller, G.R.: On being persuaded, some basic distinctions. In: Dillard, J.P., Pfau, M. (eds.)
The Persuasion Handbook, Developments in Theory and Practice. Saga Publications, London
(2002)
16. Operate, Målgruppeanalyse og Segmentering (2012)
Persuasive Backfiring: When Behavior Change
Interventions Trigger Unintended
Negative Outcomes
Agnis Stibe1(&) and Brian Cugelman2,3

1
MIT Media Lab, Cambridge, MA, USA
[email protected]
2
Statistical Cybermetrics Research Group,
University of Wolverhampton, Wolverhampton, UK
[email protected]
3
AlterSpark, Toronto, ON, Canada
Abstract. Numerous scholars study how to design evidence-based interven-

tions that can improve the lives of individuals, in a way that also brings social
benefits. However, within the behavioral sciences in general, and the persuasive
technology field specifically, scholars rarely focus-on, or report the negative
outcomes of behavior change interventions, and possibly fewer report a special
type of negative outcome, a backfire. This paper has been authored to start a
wider discussion within the scientific community on intervention backfiring.
Within this paper, we provide tools to aid academics in the study of persuasive
backfiring, present a taxonomy of backfiring causes, and provide an analytical
framework containing the intention-outcome and likelihood-severity matrices.
To increase knowledge on how to mitigate the negative impact of intervention
backfiring, we discuss research and practitioner implications.
Keywords: Backfire Taxonomy Behavior change Intention-outcome

matrix Likelihood-severity matrix Persuasive technology Intervention
design
1 Introduction
Scholars have focused on the ways in which technology can produce positive out-
comes, such as increasing users’ physical activity [29], reducing binge drinking [10],
quitting smoking [26], or managing mood and anxiety disorders [14]. There is con-
siderable research on this topic, with several systematic reviews and meta-analyses that
focus on a wide variety of positive outcomes [9, 38].
However, few papers report negative outcomes, and possibly fewer report a special
class of negative outcomes, called a backfire, which we define as an intervention that
triggers audiences to adopt the opposite target behavior, rendering the intervention
partially responsible for causing the behavior it was designed to reduce or eliminate.
Examples of backfiring interventions include drug use reduction programs that
trigger drug use by accidently creating a social norm that triggers some youth to feel
like everyone else is trying drugs except for them; traffic safety campaigns that use

DOI: 10.1007/978-3-319-31510-2_6
66 A. Stibe and B. Cugelman
shame which accidently triggers denial and resentment which leads to increased dan-
gerous driving; binge drinking screeners that trigger some youth who drink less than
average to feel pressured to catch up to their peers; or a tobacco industry sponsored
anti-smoking campaign that encouraged parents to lecture their children on not
smoking which triggered more youth to smoke [20].
In this paper, we aim to start a wider scientific discussion on intervention back-
firing, to provide analytical frameworks to help structure this discussion, to support
academic research on the topic, and to raising practitioners’ awareness of the potential
risks of intervention backfires so they can better manage the potential risks.
2 Background
Numerous fields provide recommendations on how to design behavior change inter-

ventions, including social marketing [1, 25], evidence-based behavioral medicine
[12, 13], several health behavior change approaches [2, 7, 33], socially influencing
systems [36, 37], persuasive technology [15], and classic persuasion literature [32].
However, the application of evidence-based intervention design frameworks does
not guarantee that intervention designers will achieve positive outcomes. In practice,
scientific models are more likely to inspire interventions, rather than to dictate exactly
how they are implemented [31].
Quite often, interventions that start-out with a solid theoretical underpinning, lose
their theoretical roots after adapting to real-world necessities, implementation com-
plexities, budget limits, stakeholder feedback, market testing, political tampering, etc.…
Then after moving through the process of translating abstract behavioral science prin-
ciples into concrete intervention materials, such as translating the health belief model
into a health app, it can be difficult to make clear links between the foundational theory
and applied interventions.
No matter how promising an intervention may have appeared on paper, in practice,
they always have the potential to exert unforeseen, and possibly negative outcomes.
This is one of the reasons why new behavior change interventions typically undergo
stringent monitoring and evaluation, before they are widely disseminated.
Scholars and practitioners, who report that they have disseminated a backfiring
technology, can easily feel embarrassed, or worse, find themselves not just stigmatized,
but potentially unfunded. Without doubt, there are many practical incentives and dis-
incentives that may motivate people to not publish any information that details how
their digital interventions backfired, causing even a small degree of unintended harm,
even in marginal population.
We believe that this stigma has created a climate where the existing body of
scientific literature may possess a significant degree of publication bias, resulting from
too many published studies that only report positive outcomes, and too few studies
reporting negative outcomes. This stigma has the potential to create a climate where
scientists and practitioners are at greater risk of disseminating harmful interventions.
Persuasive Backfiring: When Behavior Change Interventions Trigger 67
3 Framework
To aid discussions on intervention backfiring, this section presents two matrices that
define categories of backfires while clarifying their potential to undermine the efficacy
of behavior change interventions. Figure 1 presents the intention-outcome matrix,
which describes different types of outcomes.
Fig. 1. Intention-Outcome matrix
3.1 Axes of the Intention-Outcome Matrix

The intention-outcome matrix has two axes with four quadrants, which are described
below, and illustrated through a fictional binge-drinking reduction intervention.
• Intended Outcome. An outcome that was intended by the intervention designer.
For example, getting university students to reduce their binge-drinking.
• Unintended Outcome. Any outcome that the intervention designer did not intend,
whether it is a positive or negative outcome. For example, when a binge-drinking
screener accidently motivates some students to adopt other health outcomes, or
triggers some students to drink more alcohol after they use it in a competition to see
who can achieve the highest binge-drinking score.
• Positive Outcome. An outcome that serves the interest of both the intervention
designer and the target audience. This is a win-win situation, where both parties
benefit. For example, a binge-drinking screener where the target audience achieves
reduced alcohol consumption.
• Negative Outcome. An outcome that does not serve the interest of the target
audience. For example, a binge-drinking screener that causes small segments to
drink more alcohol, or exposes them to greater risk.
3.2 Quadrants of the Intention-Outcome Matrix

The intention-outcome matrix has four quadrants, that describes the four types of
outcomes, as described below.
• Target Behavior. The primary intended positive behavioral outcome being sought,
and typically reported.
• Unexpected Benefits. A positive behavioral outcome that was not intentionally
sought, but which was positive nonetheless, and may be reported as a compli-
mentary benefit of the intervention.
• Backfiring. This category includes a number of negative outcomes, when an
intervention causes the opposite of the outcome (e.g. more binge drinking instead of
less). It also includes “side effects”, when the primary behavior is achieved, but it
also triggers unintended negative outcomes (e.g. using peer pressure to influence
behavior which lowers audiences’ self-esteem). This quadrant is further subdivided
into a risk management matrix that contrasts the likelihood of backfiring (low to
high) with the potential severity (minor to major).
• Dark Patterns. When an intervention is used for the benefit of the developer, at the
expense of the target audience. This is in the realm of unethical applications, such as
coercion, deception, and fraud. For instance, scholars draw attention to dark game
design patterns that developers use to provide negative experiences to players, not
in their best interests, and without their consent [24, 27].
4 Method
The taxonomy of persuasive backfires presented in this paper was derived through a
grounded theory methodology [6, 17], based on a corpus of academic, applied, and
personal experiences with backfiring behavior change interventions [3].
We began the process by defining and limiting our selection criteria to interventions
that backfire, that cause the opposite behavior, or unanticipated negative consequences
that were contrary to the intentions of the program. We excluded flawed interventions
that did not contribute towards any outcomes, positive or negative, as these programs
constituted poor implementations, not backfires. The first type of flawed intervention
includes programs that audiences did not find motivating, as these programs lacked the
capacity to achieve significant outcomes. Similarly, we excluded interventions that
faced implementation barriers, or created barriers among its target audiences, as it was
not possible to assess the outcomes of these interventions.
To gather qualifying sources, we ran a call for references and examples across several
academic, professional, and personal networks. The types of references we collected
included journal papers, articles, program evaluations, and personal experiences. In total,
we collected 47 responses.
We systematically reviewed all sources, and only included submissions that
qualified as having demonstrated an unintended negative outcome, or which were
presented as having contributed an unintended negative outcome. We also included
behavior change interventions that were not implemented within technology per se, but
were reasonably implemented in a context that could be applied to online behavior
change campaigns or digital products. We also received submissions of backfiring
legislation, which is often used in conjunction with communication campaigns, to elicit
social change. In total 30 responses were included in our qualitative analysis.
We carried out a qualitative assessment of the corpus, with a view to developing
taxonomy of triggers for backfiring interventions.
5 Findings
The findings of our study are presented in Fig. 2, a risk management framework called
the likelihood-severity matrix. The four quadrants of backfires include (A) low like-
lihood and minor severity, (B) high likelihood and minor severity, (C) low likelihood
and major severity, and (D) high likelihood and major severity. Through our qualitative
assessment of backfires and their potential risks, we have synthesized twelve types of
backfires, within six categories: inexperience, fineprint fallacy, personality responses,
credibility damage, poor judgment, and social psychology.
Fig. 2. Likelihood-Severity matrix

The taxonomy of persuasive backfires, presented in Table 1, provides twelve types

of backfires within six broad categories.
Table 1. Taxonomy of persuasive backfires

Description Examples
A: LOW likelihood and MINOR severity
Inexperience
Superficializing
The superficial application of theory, such as Research on social media, particularly on
copying surface tactics without Foursquare, looks at driving user
understanding the underlying strategies, motivation through gamification
principles, or reasoning. mechanics that compete with intended
target behaviors of information sharing
[23]. More generally, research that focuses
on how gamification elements can drive
extrinsic motivation, may cease to be
persuasive over the long term, and may
potentially deplete intrinsic motivation.
B: HIGH likelihood and MINOR severity
Fineprint Fallacy
Overemphasizing
Motivate people to take action for one Stressing “low fat”, while still including
strongly emphasized benefit, while several unhealthy factors, such as
omitting (or hiding) harmful factors that high-sugar, or trans fats.
are in the fine print. Asking people to take one pill a day to lower
their risk of contracting HIV, which causes
some groups to engage in riskier sex (and
stop using condoms), contributing to an
increase in other sexually transmitted
diseases [22].
Personality Responses
Defiance Arousing
Resistance to messages that are incompatible Guilt and shame messages in anti-drinking
with a person’s self-identity, that can ads for drunk drivers are ignored, and in
induce unpleasant cognitive dissonance, some cases, backfire, when the message is
leading the audience to reject the message, incompatible with how individuals view
or oppose it. themselves.
Persuasive messaging involving an authority
might provoke opposition from groups and
individuals who are sensitive or resistant to
authorities.
One anti-smoking campaign failed due to
advocating the message that “teens
shouldn’t smoke… because they’re teens.”
Youths in the 10th-12th-grade were 12
percent more likely to smoke for each
parent-targeted ad they had seen in the last
(Continued)
Table 1. (Continued)
30 days. According to developmental
psychologists, teens 15 to 17 years old tend
to reject authoritative messages because
they believe they are independent, which
renders Philip Morris’ ad campaign largely
useless [16].
The presence of peer information decreases
the savings of nonparticipants who are
ineligible for automatic enrollment in a
saving plan, while higher observed peer
savings rates also decreased savings.
Discouragement from upward social
comparison seems to drive this reaction [4].
Self-Licensing
When someone does something good in one After donating to charity, people may feel
area, they sometimes feel like they have a licensed to behave less morally in
license to misbehave in other. subsequent decisions [30]. For instance,
donating to charity may have a dark side to
it, as it negatively affects subsequent,
seemingly unrelated moral behavior, such
as the intention to be environmentally
friendly.
C: LOW likelihood and MAJOR severity
Credibility Damage
Self-Discrediting
When the source disseminates discrediting People report less favorable thoughts and
information, causing a misalignment of attitudes towards a source, after reading
source and message credibility. weak arguments presented by a high vs.
low expertise source.
Too much fear mongering may discredit a
campaign to the point of disbelief or humor
[21].
Message Hijacking
Third party actors re-contextualize the Creative works designed to cause fear,
source’s message, bringing a new become a trendy meme with a different
meaning, which in many cases undermines meaning, such as humorous cigarette ads
the intervention by turning it into a public of smoking children, reefer madness,
joke. fashionable heroin chic [19], or the ad
campaign for TV PSA “This is your brain
and this is our brain on drugs” which
triggered numerous parodies [7].
England’s Beat Bullying Campaign triggered
bullying and violence. The campaign was so
popular at its launch that supplies of the
“Beat Bullying” wristbands quickly sold
(Continued)
out. Because of the scarcity of the bracelets,
and theme of the campaign, some kids were
bullied for wearing the bracelet [11].
D: HIGH likelihood and MAJOR severity
Poor Judgment
Mistailoring
When a tailored messaging system provides A drinking screener showed both low and
information that produces negative high drinking students how much they
outcomes in some users, that could have consume in comparison to an average
been avoided with an appropriate message. consumption. Those that were above the
norm felt encouraged to drink less, while
those below received an implied message
to drink more.
And a boomerang effect [35], when a
descriptive social norm was not
accompanied by an injunctive social norm
in a similar way as described above.
Mistargeting
When a message that was intended for one A one-size-approach to persuasive
audience segment is misinterpreted by messaging can deter healthy eating
another group of people. behavior change, leading to a negative
change in attitudes towards healthy eating
over time.
The Playpump program was meant for
children to pump water while playing, but
resulted in adults using the playground
pump, leading to back injuries and other
health problems [5].
Misdiagnosing
When a behavior change intervention does Gaze tracking software, designed to provide
not properly diagnose user behavior or proactive help as patients read medical
psychological processes. documents, used fixation time as a cue to
identify when users were struggling with
the material, and might need help. During
the trial, the participants with low health
literacy had a slower reading rate, causing
the system to inappropriately offer help
continually, which just annoyed the users,
leading to lower comprehension compared
to the control condition.
Misanticipating
Changes in policies or directives that lead to In the Netherlands, the drinking age was
unanticipated shifts in beliefs, attitudes or changed from 16 to 18, which news
behaviors. stations linked to an increase in drug use
among citizens in this age group.
(Continued)
Nebraska’s “Give Us Your Troubled Child”
law backfired, as the Nebraska Safe Haven
law enabled parents to drop off kids of any
age and the state for the state to take in,
resulting in some parents who dropped off
grownups [Gra2008].
Social Psychology
Anti-Modeling
Demonstrating negative behavior, which Being exposed to others experiencing the
exposes people to memory triggers of bad stress of quitting smoking, for example,
behaviors or temptations, potentially in triggers people to want to smoke.
moments of the greatest susceptibility. Anti-bad-behavior interventions can
remind people of the bad behavior, thus
potentially spark their motivation.
A cookie company that introduced a
100-calorie packs of snacks, triggered
people to eat far more of the small snack
packs [28].
Reverse Norming
Interventions that use examples of popular An anti-littering program used campaign
bad behaviors can establish the bad posters to stress how widespread littering
behavior as a social norm. was, accidently contributed to a social
norm for littering, by demonstrating how
many people were littering.
Two separate studies indicated that D.A.R.E.
program was ineffective and in some cases,
pushed kids toward drug use and lowered
self-esteem. Researchers suspected that the
intervention’s message made some kids
want to try drugs as a way of fitting in,
stating and the program’s message could be
misinterpreted by youth as conveying that
“peer pressure is around every corner,
because everyone is doing drugs but
you!” [34].
6 Scientific Considerations
We have authored this paper to initiate a scientific discussion on intervention back-

firing, and to encourage scholars to examine this phenomenon in greater depth. We
hope this will contribute to more transparency within the academic community, leading
to more well-rounded research on persuasive design and its application.
The scientific contribution of this paper includes the persuasive backfiring frame-
work, its two matrices (intention-outcome and likelihood-severity), and taxonomy.
These frameworks can be used to define, discuss, and further research behavior change
interventions that trigger unintended negative outcomes. Although we have not dis-
cussed ethics in this paper, it provides a system to further define ethical and unethical
uses of persuasive technology.
Perhaps the most important contribution of this paper, is raising awareness of the
elephant in the room, the well-known but rarely discussed fact that applied interven-
tions targeting good outcomes, occasionally produce negative outcomes. It also raises
questions on the social stigma attached to the people and organizations that deploy
backfiring interventions, and how this may contribute to under reporting negative
outcomes, or omitting them altogether.
More concerning, this investigation has identified a potentially large source of
publication bias, as we believe that there are many incentives and disincentives,
including stigma and embarrassment, that motivate researchers to avoid publishing
research on negative outcomes.
We believe that scholars will need to develop innovative new research methods to
overcome the research barriers that surround this subject, due to the stigma associated
with reporting on interventions that backfire, and the limited ability of scientific studies
to identify backfires that are more likely to become apparent in evaluations of
real-world interventions, and often informally known by intervention staff but not
publicly reported.
For instance, given the ability of technology to employ tailoring techniques, where
content can be personalized, persuasive technology scholars are better equipped to
undertake research on backfiring psychology, and use this knowledge to advise inter-
vention designers when they need to omit influence principles that may be counter-
productive to particular segments.
7 Practitioner Considerations
To assist practitioners, the taxonomy and tools presented in this paper provides a list of
risks that can be addressed, if identified before they occur. Moreover, these risks can be
used to build more effective interventions, by teaching practitioners how to avoid,
mitigate, or manage backfires. The examples in this study suggest that backfires may
originate from political tampering during the intervention design phase, evaluations
that do not distinguish between groups, overusing a principle to the point of triggering
mistrust in a message (“your brain on drugs”), misdiagnosis that leads to subscribing
the wrong intervention leading to garbage-in-garbage-out interventions.
However, one ethically questionable practice we discovered was the potential
intentional use of backfiring as a dark pattern, designed to deploy an intervention that
superficially appears to promote a healthy behavior, but which actually promote
unhealthy behavior. When corporations are obliged, or volunteer to carry out public
health interventions to warn the public against their product, these corporations can
easily benefit from the intentional use of backfiring interventions.
For instance, the “Talk: They’ll Listen” campaign is frequently cited as an example
of a cleaver antismoking ad campaign that on the outside appeared to be a legitimate
antismoking campaign, but which in practice caused an increase in youth smoking [19].
Consequently, policy makers and regulators who empower tobacco, alcohol,
and pharmaceutical companies to run their own interventions, need to be fully informed
about the potential intentional use, and abuse of backfire-based campaigns, that
superficially look effective, but at a deeper level, have been engineered to encourage
the opposite effect.
Finally, we believe that backfires are normally present to some degree in all
interventions, and that they can never be fully eliminated. It may be more practical to
focus on strategies to reduce the riskiest backfires, to manage those that cannot be fully
eliminated, and to continually monitor and improve interventions over time.
8 Conclusions
The stigma associated with reporting behavior change interventions that trigger neg-
ative outcomes, has relegated the topic of intervention backfiring to an informal
observation that is widely known, but rarely discussed or reported. This has created a
climate where scholars routinely overemphasize positive outcomes, while failing to
report the fact that the same principle, can also lead to unforeseen negative outcomes.
In this paper, we discussed multiple ways how behavior change interventions can
backfire. We provided a framework to help facilitate the discussion of this topic,
presented tools to aid academics in the study of this realm, and offered advice to
practitioners about potential risks. We encourage researchers to build on this work, and
take a more systematic look on approaches involving the design of behavior change
interventions.
In the future, researchers will need to innovate new ways to study this subject, and
extend our scientific and practical knowledge of what pitfalls need to be avoided when
designing technology-supported behavior change interventions. We advocate that
researchers and practitioners adopt an honest and open attitude towards identifying and
removing backfires as soon as possible, and to disseminating strategies to reduce their
occurrence, before they cause more harm than good.
References
1. Andreasen, A.: Social marketing in the 21st century. Sage Publications Inc, Thousand Oaks
(2006)
2. Bartholomew, L.K., Parcel, G.S., Kok, G., Gottlieb, N.H.: Planning health promotion
programs: an intervention mapping approach. Wiley, San Francisco (2011)
3. Beer, M., Eisenstat, R.A., Spector, B.: Why change programs don’t produce change (1990)
4. Beshears, J., Choi, J.J., Laibson, D., Madrian, B.C., Milkman, K.L.: The effect of providing
peer information on retirement savings decisions. J. Finan. 70(3), 1161–1201 (2015)
5. Chambers, A.: Africa’s not-so-magic roundabout (2009). http://www.theguardian.com/
commentisfree/2009/nov/24/africa-charity-water-pumps-roundabouts
6. Charmaz, K.: Constructing grounded theory. Sage, Thousand Oaks (2014)
7. Chatterjee, S., Price, A.: Healthy living with persuasive technologies: Framework, issues,
and challenges. J. Am. Med. Inf. Assoc. 16(2), 171–178 (2009)
8. Crano, W.D., Burgoon, M., Oskamp, S. (eds.): Mass media and drug prevention: Classic and
contemporary theories and research. Psychology Press, Mahwah (2001)
9. Cugelman, B., Thelwall, M., Dawes, P.: Online interventions for social marketing health
behavior change campaigns: A meta-analysis of psychological architectures and adherence
factors. J. Med. Internet Res. 13(1), e17 (2011). http://doi.org/10.2196/jmir.1367
10. Cunningham, J.A., Wild, T.C., Cordingley, J., Van Mierlo, T., Humphreys, K.: A
randomized controlled trial of an internet-based intervention for alcohol abusers. Addiction
104(12), 2023–2032 (2009)
11. Curtis, P.: Anti-bullying wristband scheme backfires (2004). http://www.theguardian.com/
education/2004/dec/08/schools.uk2
12. Davidson, K., Goldstein, M., Kaplan, R., Kaufmann, P., Knatterud, G., Orleans, C.,
Whitlock, E.: Evidence-based behavioral medicine: What is it and how do we achieve it?
Ann. Behav. Med. 26(3), 161–171 (2003)
13. Embry, D., Biglan, A.: Evidence-based kernels: Fundamental units of behavioral influence.
Clin. Child Family Psychol. Rev. 11(3), 75 (2008)
14. Farvolden, P., Denisoff, E., Selby, P., Bagby, R.M., Rudy, L.: Usage and longitudinal
effectiveness of a Web-based self-help cognitive behavioral therapy program for panic
disorder. J. Med. Int. Res. 7(1), e7 (2005). http://www.ncbi.nlm.nih.gov/pmc/articles/
PMC1550639/
16. Fraser, J.: Anti-smoking ads cleverly boost smoking among teens (2006). http://www.
naturalnews.com/020996.html
17. Glaser, B.G., Strauss, A.L.: The discovery of grounded theory: Strategies for qualitative
research. Transaction Publishers, New Jersey (2009)
18. Graham, J.: Father leaves nine children at Nebraska hospital (2008). http://newsblogs.
chicagotribune.com/triage/2008/09/father-leaves-n.html
19. Hastings, G.: Social marketing: why should the devil have all the best tunes?.
Butterworth-Heinemann, Oxford (2007)
20. Healey, B., Zimmerman, R.S.: The new world of health promotion: New program
development, implementation, and evaluation. Jones & Bartlett Learning, Sudbury (2009)
21. Hinkley, K.: 5 Ridiculous Anti-Drugs Posters (2014). http://www.talkingdrugs.org/5-anti-
drugs-campaigns
22. Holpuch, A.: Truvada has been called the ‘miracle’ HIV pill–so why is uptake so slow
(2014). http://www.theguardian.com/world/2014/sep/18/truvada-mircle-pill-prevent-hiv-
controversy
23. Kietzmann, J.H., Hermkens, K., McCarthy, I.P., Silvestre, B.S.: Social media? Get serious!
Understanding the functional building blocks of social media. Bus. Horiz. 54(3), 241–251
(2011)
24. Kirkland, A., Metzl, J.M. (eds.): Against health: How health became the new morality. NYU
Press, New York (2010)
25. Kotler, P., Roberto, N., Lee, N.: Social marketing: improving the quality of life, 2nd edn.
Sage Publications Inc, California (2002)
26. Lenert, L., Munoz, R., Perez, J., Bansod, A.: Automated e-mail messaging as a tool for
improving quit rates in an internet smoking cessation intervention. J. Am. Med. Inf. Assoc.
11(4), 235–240 (2004)
27. Linehan, C., Harrer, S., Kirman, B., Lawson, S., Carter, M.: Games against health: a
player-centered design philosophy. In: Proceedings of the 33rd Annual ACM Conference
Extended Abstracts on Human Factors in Computing Systems, pp. 589–600. ACM (2015,
April)
28. London, L.: Guiltless Girls: Unpacking 100-calorie Snacks (2009)

29. Marshall, A., Leslie, E., Bauman, A., Marcus, B., Owen, N.: Print versus website physical
activity programs A randomized trial. Am. J. Prev. Med. 25(2), 88–94 (2003)
30. Meijers, M.H., Verlegh, P.W., Noordewier, M.K., Smit, E.G.: The dark side of donating:
how donating may license environmentally unfriendly behavior. Social Influence, 1–14
(2015). doi:10.1080/15534510.2015.1092468
31. Michie, S., Abraham, C.: Interventions to change health behaviours: evidence-based or
evidence-inspired? Psychol. Health 19(1), 29–49 (2004)
32. O’Keefe, D.: Persuasion: Theory and Research. Sage Publications Inc, London (2002)
33. Prochaska, J.O., Velicer, W.F.: The transtheoretical model of health behavior change. Am.
J. Health Promot. 12(1), 38–48 (1997)
34. Reaves, J.: Just Say No to DARE (2001). http://content.time.com/time/education/article/
0,8599,99564,00.html
35. Schultz, P.W., Nolan, J.M., Cialdini, R.B., Goldstein, N.J., Griskevicius, V.: The
constructive, destructive, and reconstructive power of social norms. Psychol. Sci. 18(5),
429–434 (2007)
36. Stibe, A.: Towards a framework for socially influencing systems: Meta-analysis of four
PLS-SEM based studies. In: MacTavish, T., Basapur, S. (eds.) PERSUASIVE 2015. LNCS,
37. Stibe, A.: Advancing typology of computer-supported influence: moderation effects in
socially influencing systems. In: MacTavish, T., Basapur, S. (eds.) PERSUASIVE 2015.
38. Webb, T., Joseph, J., Yardley, L., Michie, S.: Using the internet to promote health behavior
change: a systematic review and meta-analysis of the impact of theoretical basis, use of
behavior change techniques, and mode of delivery on efficacy. J. Med. Internet Res. 12(1),
e4 (2010). http://doi.org/10.2196/jmir.1376
Captology and Technology Appropriation:
Unintended Use as a Source
for Designing Persuasive Technologies
Alina Krischkowsky ✉ , Bernhard Maurer, and Manfred Tscheligi

( )
Center for Human-Computer Interaction, Department of Computer Sciences,

University of Salzburg, Salzburg, Austria
{alina.krischkowsky,bernhard.maurer,manfred.tscheligi}@sbg.ac.at
Abstract. In this paper we theoretically reflect upon persuasive technology usage

under the light of technology appropriation. The intended usage of technology often
fails, meaning that the designers’ intended use is not always translated into user
behavior. This is also true for persuasive technology, since technology will always
be used within a context involving users’ own intentions that may not always be
anticipated by designers. This clashes with Fogg’s framing of captology, which
explicitly focuses on endogenous intent, i.e., a persuasive intent that is designed into
a technology. With this paper we open up an initial theoretical discourse around
these two concepts, highlighting how the design of persuasive technologies can be
informed by existing knowledge around technology appropriation. This is done by
reflecting upon three identified ‘action points’: (1) learning from appropriation,
(2) designing for appropriation, and (3) designing for personal differences and ambi‐
guity of interaction.
Keywords: Technology appropriation · Captology · Intentionality · Unintended use
1 Introduction
In this paper we open up a discourse around the design and use of interactive persuasive
technology (PT) by reflecting upon conceptualizations around technology appropriation and
unintended use. Since persuasion requires intentionality (e.g., [1, 2]), we argue that these
technological ‘built-in’ intentions will always be used within a context involving users’ own
intentions that may not always be accordingly anticipated by designers. Therefore, we
discuss in how far designing for the unexpected [8], in terms of unintended use, can be a
valuable source for designing PTs. In particular we will reflect upon Fogg’s captological
design [2] and on technology appropriation in terms of related design perspectives and prin‐
ciples. In doing this theoretical reflection, we highlight that also unintended use of PT does
not necessarily have to be a sign of ‘failure’ but rather a sign that the technology has become
“the user’s own” (e.g., [8, 17]). Such phenomena need to be named and taken seriously and
not neglected by considering them ‘side effects’ [2]. Especially, for the design of PTs the
study of unintended use is critical to understand the manifold ways of how technology can
be appropriated, to also counteract more systematically undesired negative consequences.

DOI: 10.1007/978-3-319-31510-2_7
Captology and Technology Appropriation 79
2 Intended and Unintended Use
The following section details related work on PT and intentionality complemented with
remarks on technology appropriation and unintended use to outline the very character‐
istic foundations of these two concepts.
2.1 Intended Use and Persuasive Technology
As already defined in Foggs’ early work [1], PT is an “interactive technology that

changes a person’s attitudes or behaviors” (p. 225). A central aspect of this definition is
that persuasion always implies an intent to change attitudes and/or behaviors. So to say,
persuasion requires intentionality (e.g., [1, 2]). The concept of captology (i.e., the study
of computers as PT [2]) specifically focuses on planned persuasive effects of technology
and not on potential side effects, i.e., that persuasive intent is designed into a computing
product, also referred to as– “built-in”– persuasive intent [2].
However, Fogg’s work has also been criticized (e.g., [3, 4]). For example, Johnson [4]
emphasizes that captological design neglects user-centered design philosophies, since it
excludes unintended consequences and instead focuses on attitude and behavior changes
that are intended by the designer. Here, Atkinson argues that an “ethical design impasse is
created” ([3], p. 171), since responsible user-centered design necessities and a sound
examination of intended and unintended consequences of technology usage are missing.
Verbeek [6] agrees with that and highlights that unintended consequences need to be
fundamentally incorporated into design decisionmaking processes by better understanding
and predicting them. These complementary viewpoints illustrate to explicitly include unin‐
tended use of PTs in research, design and theoretical reflections thereof to emphasize their
innovation potential, but also systematically counteract undesired negative consequences.
2.2 Unintended Use and Technology Appropriation
Verbeek talks about the uncertainty that surrounds eventual effects of PT, as persuasion
cannot be seen as an intrinsic property of technologies [6]. Therefore, technologies may
be approached as entities that have no fixed identities, rather embody interpretative
flexibility [16] and/or multistability [7]. These approaches towards understanding tech‐
nologies in their actual contexts of usage, derive from various research disciplines and
theorists, such as e.g., phenomenology [7], or mediation theory [6]. The theoretical core
idea is that technologies need to be interpreted and appropriated by the user in order to
be used [6]. There exist many examples of unintended use, coming from people’s
everyday life experiences and practices, but also from observations in various research
disciplines, such as HCI (e.g., [17]). Technology that is used differently than initially
intended by the designers and not simply what they gave to them [9], may not be a sign
of ‘failure’ rather than a sign that the technology has become “the user’s own” (e.g.,
[8, 17]). In regard to ‘failing’ technologies, Dix [8] and Carroll [9] counter that “design
can never be complete” as it is impossible to design for the unexpected, but that “you
can design to allow the unexpected”. Thus, users complete the design as they adapt and
appropriate it [9], fitting them into their working practices [10].
80 A. Krischkowsky et al.
3 Deriving an Informed Basis: Appropriation and Persuasion
An important question deriving from this discourse around persuasion and appropriation
is, in how far these concepts clash or complement with each other? In this section, we
discuss this question on basis of their theoretical foundations.
Captological design focuses on planned persuasive effects and not on potential side
effects [2]. Simply said, if a technological artifact is designed with an intent in mind to
change a specific attitude or behavior (e.g., persuade drivers to change their driving style
towards a more economic one), and you observe in your studies that your intended attitude
and behavior change is met (e.g., people adopted your designed fuel-saving technology to
drive more economical), your design has worked out. But what happens if the planned
outcome is different than intended (e.g., no or different attitude/behavior change adopted)?
We agree that unintended consequences categorically do not belong with having been
persuaded [3]. However, is it also a sign of failure from a design perspective? In many
disciplines, appropriation is considered as a positive phenomenon in system design and a
source for technological innovations [8, 9]. Atkinson [3] and Berdichewsky et al. [5] state
that it is central for the persuasive discipline itself to show (moral) responsibility and name
such phenomena, i.e., side effects that are neglected in captological design.
Thereby, appropriation holds several advantages, such as, situatedness, dynamics
and ownership [8]. Situatedness means that the end point of design is the intervention
[11], and not just an artefact or an artefact and its immediate ways of interacting with
it. It is rather the way it changes the environment in which it is set [8]. Therefore, in
designing PTs, we cannot expect to understand each usage context fully and meet every
possible need to change a certain behavior or attitude. It can rather be about the general
changes/interventions (also unintended) a new technology brings into a given usage
context that may be seen as initial leverage points to innovate and (re-)design.
Concerning dynamics, it is critical for the design process, to envision that usage envi‐
ronments and needs change. Over time, the use context of technology may change (e.g.,
a persuasive app that was designed for individual use, is used by multiple users). The
technology therefore needs to be responsive to this change (e.g., explicitly allowing for
collaborative use). With appropriation a certain sense of ownership evolves. When
designing technologies with a specific intent, people using these systems may feel a loss
of control, i.e., not doing things their own way. This aspect has already been discussed
in persuasive literature (e.g., [3, 5]), in terms of moral and ethic questions (e.g., human
autonomy) surrounding PT design and use. In contrast, appropriation argues for owner‐
ship, to allow people to do things their own way to raise positive feelings that can be as
important as the things that can be achieved with the very technology [8]. Therefore,
designing situated and dynamic PTs that also allow for ownership, hold the potential to
create technologies that persuade by being situated in peoples practices.
4 Action Points
Based on this general discussion around appropriable PTs that allow for situatedness,
dynamic use and ownership, the following section details action points of how we can
leverage the users’ different ways of appropriating PTs and thereby, complete the design
through their actions. This is done by outlining how designers and users of PTs can learn
from appropriation and unintended use practices as well as how designers can apply
appropriation principles to fit the users’ respective appropriation ‘needs’.
Learning from Appropriation and Unintended Use. The knowledge on how a user
appropriates a specifc design or artifact is a valuable source for reflection for both, the
user (i.e., learning from own behaviour) as well as the designer (i.e., learning based on
the user’s unintended use in context). By deliberately designing for open-ended use, the
act of appropriation itself can become a reflective practice for the user, leading to new
layers a persuasive design can address. For instance, technology can allow for
confronting the user with his/her appropriations (i.e., how s/he uses the technology
differently on a daily basis). This confrontation can be used by designers of PT as a way
to induce behaviour change by allowing the user to reflect upon his/her own actions and
usage appropriations.
We further argue for open ended persuasive designs in order to address long term
persuasive goals, rather than short term. PT that is appropriable can potentially be more
deeply rooted in a user’s everyday practices. In that sense, persuasive techologies that
are integrated into users’ everyday routines, by the users themselves, can also become
natural for the user, making such technologies inherently part of their lives to address
long term persuasive goals. In that sense, technologies can become mediatory artifacts
constantly adapting and communicating with the user.
Designing for Appropriation: Intended vs. Unintended Use. There exist several
design approaches with a strong focus on appropriation, such as “continuous design and
redesign” [14], “continuing design in use” [13], and metadesign “designing for design
after design” [15]. What unites all of these approaches is that the main challenge for
designers is to design malleable technologies that can be adapted to users’ organiza‐
tional, social and personal practices to harvest users’ needs in order to improve its design
[9]. The main question that arises is– How can we balance intentionality (for persuasion
purposes) with open, flexible and dynamic design to allow for unintended use (for
appropriation purposes)? We believe that designing appropriable PTs is a challenge, but
the fact that design for appropriation is possible is made most clear by realizing that
some sorts of design make appropriation difficult or impossible [8]. This might be true
for PT as well. In the following we outline some major design principles from appro‐
priation literature that can provide an initial step towards designing appropriable PTs.
Interpretation allowed! (e.g., [6–8, 16]) It is important that not everything in a system
or product does have a fixed meaning. It is rather critical to include elements where users
can add their own meanings, contextualize their own interactions, to make them adapt‐
able to their own needs and surroundings of interaction.
Provide visibility! (e.g., [8]) It is central that the systems’ functioning is clear to the
users so that they can anticipate likely effects of their actions. This is how users are
empowered to make the system do what they would like. This is particularly relevant
for technologies when the effects of actions become visible at different time.
82 A. Krischkowsky et al.
Reveal Intentions! (e.g., [8]) Appropriation can be used to subvert systems (i.e.,
deliberately using something different than intended). This is of particular relevance for
persuasion, as designers may not try to prevent such subversion but aim to expose the
intention behind the system. This supports users in choosing appropriations that may
subvert the system rules but still preserve the intent. This means that persuasion may
work in its inherent usage and not exclusively with the given functions.
Learn! (e.g., [4, 8]) A user’s appropriation practices can lead to new technologies
that incorporate these adaptations created by the user, supporting the development of
new innovative technologies. In observing and documenting in our research the many
different ways in which technology has been appropriated, technology may be re-
designed or new technologies emerge to support newly discovered uses and resulting
behaviors. This form of co-design, has often been criticized as being lacking in persua‐
sive and specifically captological designs.
By considering interpretative flexibility, visibility and the exposure of intentions in
the design of PTs, we argue that users may be given the opportunity to complete the
design and persuasive intent with their own actions and appropriations.
Designing for Personal Differences and Ambiguity of Interaction. Besides allowing

for appropriated use based on e.g., everyday practices of a user, we further argue for
allowing appropriation and unintentionality on an interactional level. As physical action
is an important component of human cognition and self-awareness, it is strongly related
to how a human creates reason and perceives himself within the environment. This
notion of thinking through doing [12] is for instance very prominent in body-based
interactions. However, every body is different and individual movement differences are
a matter of personal traits. Embracing this ambiguity of interactions can be a promising
source for interaction designs of PT. If a PT allows for body-based interactions (e.g.,
interacting via a physical artifact), a user’s personal way of doing and interacting with
it can potentially foster the act of appropriation, leading to higher user engagement.
Ideally, the act of using PT can then become a matter of self expression, which makes
such technology much more likely to be adopted and used on a daily basis. This presents
new design opportunities for PT that are deeply embedded into the user’s everyday
practices and routines.
5 Conclusion and Future Work
With this paper we took an initial step towards designing for appropriable PTs. In our
discussion around persuasion and appropriation and our three derived ‘action points’
we have highlighted how conceptualizations around appropriation can potentially be a
source for the design of PTs. Nevertheless, with this theoretical discourse new important
questions emerge, such as, how to balance precise persuasive intent and open-ended
design at the same time? Such questions will be part of our future work, aiming to derive
concrete principles for the design of appropriabale PTs on basis of empirical ‘over time’
research and theoretical reflections thereof.
Acknowledgements. The financial support by the Austrian Federal Ministry of Science,

Research and Economy and the National Foundation for Research, Technology and Development
is gratefully acknowledged (Christian Doppler Laboratory for Contextual Interfaces).
References
1. Fogg, B.J.: Persuasive computers: perspectives and research directions. In: Proceedings of
the CHI, pp. 225–232. ACM (1998)
Morgan Kaufmann, Amsterdam (2003)
3. Atkinson, B.M.: Captology: A critical review. In: IJsselsteijn, W.A., de Kort, Y.A., Midden,
C., Eggen, B., van den Hoven, E. (eds.) PERSUASIVE 2006. LNCS, vol. 3962, pp. 171–182.
Springer, Heidelberg (2006)
4. Johnson, R.R.: Book reviews: Persuasive technology. JBTC, J. Bus. Tech. Commun. 7, 251–254
(2004)
5. Berdichewsky, D., Neuenschwander, E.: Toward an ethics of persuasive technology’.
Commun. ACM 42(5), 51–58 (1999)
6. Verbeek, P-P.: Persuasive Technology and Moral Responsibility: Toward an ethical
framework for persuasive technologies. In: Proceedings of PERSUASIVE 2006, Springer
(2006)
7. Ihde, D.: Technology and the Lifeworld. Bloomington/Minneapolis, Cambridge (1990)
8. Dix, A.: Designing for appropriation. In: Proceedings of the BCS-HCI 2007. UK, pp. 27–30
(2007)
9. Carroll, J.: Completing design in use: Closing the appropriation cycle. In: Proceedings of the
ECIS 2004, Turku, Finland, 11 p. Paper 44 (2004)
10. Dourish, P.: The appropriation of interactive technologies: Some lessons from placeless
documents. JCSCW 12(4), 465–490 (2003)
11. Dix, A., Finlay, J., Abowd, G., Beale, R.: Interaction design basics, 3rd edn. Human Computer
Interaction. Prentice Hall, Upper Saddle River (2004)
12. Klemmer, S.R., Hartmann, B., Takayama, L.: How bodies matter: Five themes for interaction
design. In: Proceedings of the DIS 2006, pp. 140–149. ACM (2006)
13. Henderson, A., Kyng, M.: There’s no place like home: Continuing design in use, pp. 219–240.
Lawrence Erlbaum, USA (1991)
14. Jones, J.C.: Continuous design and redesign. Des. Stud. 4(1), 53–60 (1983)
15. Ehn, P.: Participation in design things. In: Proceedings of the Participatory Design 2008,
pp. 92–101 (2008)
16. Bijker, W.E., Pinch, T.J.: The Social Construction of Facts and Artifacts: Or How the
Sociology of Science and the Sociology of Technology Might Benefit of Each Other. MIT
Press, Cambridge MA (1987)
17. Bødker, S., Christiansen, E.: Poetry in motion: Appropriation of the world of apps. In:
Proceedings of the ECCE 2012, pp. 78–84. ACM (2012)
Prevention and Motivation
Self-Reflecting and Mindfulness: Cultivating Curiosity
and Decentering Situated in Everyday Life
Ralph Vacca ✉ and Christopher Hoadley

( )
Educational Communication and Technology, New York University, New York, USA
{ralph.vacca,tophe}@nyu.edu
Abstract. Research on the use of mobile to promote mindfulness states is still rela‐
tively nascent, especially when exploring how such states can be cultivated in
everyday life, outside of meditation-based approaches. In this study we investigate the
design of a mobile app that seeks to cultivate mindfulness states situated in everyday
life. Using reminders to prompt self-reflection and breathing exercises to prompt body
awareness, we sought to address the overarching question – how can we design
towards mindfulness situated in everyday living and how might it change what we
mean by mindfulness? Our findings suggest that mobile-based approaches can
promote curiosity and decentering through self-reflection, and that the valence and
likelihood of experiencing certain mental events may influence how self-reflection
is experienced, which in turn influences curiosity and decentering factors of mind‐
fulness states.
Keywords: Mindfulness · Situated context · Self-reflection · Curiosity ·

Decentering · Emotional health · Persuasive design · Mobile learning
1 Introduction
Traditionally, mindfulness focuses on complete freedom from suffering and cultivating

positive qualities of the mind characterized by a state of altruistic omniscience. Wester‐
nized adoption of mindfulness is largely removed from the spiritual origins and focused
on the therapeutic benefits – happiness and wellness. Despite these differences, both
share a characterization of mindfulness as a transient state of non-appraisal in which
mental experiences and sensory information are meta-cognitively monitored without
evaluation or interpretation [1, 2]. Furthermore, both share a view of mindfulness
training as a cultivation of dispositional traits that eventually will impact all aspects of
one’s everyday life [3].
Building on the momentum of the quantified-self and persuasive technology move‐
ment is the opportunity to use mobile devices to support mindfulness as a practice that
is situated in everyday life. Mobility is increasingly being understood as the mediation
of one’s relationship with situated contexts such as location and those around us [4].
The original version of this chapter was revised.

An erratum to this chapter can be found at 10.1007/978-3-319-31510-2_31

DOI: 10.1007/978-3-319-31510-2_8
88 R. Vacca and C. Hoadley
In this study we investigate the design of a mobile app that seeks to activate and
support mindfulness states situated throughout a person’s everyday life. Using reminders
to prompt self-reflection and breathing exercises to prompt body awareness, we sought
to address the overarching question – how can we design towards mindfulness situated
in everyday living and how might it change what we mean by mindfulness? Specifically
we sought to explore how self-reflection throughout one’s day serves as a mediating
process that influences the way one experiences curiosity and decentering – two factors
commonly used to operationalize a mindfulness state [5].
2 Related Work
2.1 Mindfulness States

Mindfulness can be understood both as a dispositional trait and as a temporary state that
can be induced. The two perspectives are interlinked in that consistent induction of
mindfulness states may lead to long-term changes in mindfulness as a dispositional trait
[6]. When understanding mindfulness as a state, two factors of curiosity and decentering
are often used to operationalize the construct [5]. Lau et al. [5] characterize curiosity as
present-moment awareness with an investigative interest, while decentering is defined
as reflecting a shift from identifying personally with thoughts and feelings to relating to
one’s experience of a wider field of awareness. In other words, decentering is about
seeing one’s thoughts and feelings as passing mental events in the mind rather than
reflections of reality [7].
There is debate as to how such states can be cultivated that stem largely from one’s
interpretation of what a mindfulness state actually is. For instance, as we mentioned, the
curiosity factor of a mindfulness state is characterized as being present-moment aware‐
ness with an investigative interest. There however are differences in the extent to which
such awareness is unencumbered by language or conception. For instance, Buddha’s
earliest teachings describe mindfulness as a form of moment-to-moment application of
bare attention that does not linguistically or conceptually elaborate one’s observed
experience [8]. This may stand in contrast to approaches to mindfulness that see aware‐
ness as a form of dealing with the narratives themselves such as therapeutic interventions
that advocate for forms of managing the interpretations (e.g., acceptance) we assign
through mindfulness-like approaches [3, 9]. Furthermore, ongoing work on under‐
standing the neurological underpinnings of mindfulness indicate that early stage mind‐
fulness practitioners may experience top-down attentional control processes (i.e.,
conceptual), while more experienced practitioners are able to attenuate in a more non-
conceptual experiential manner [10].
2.2 Self-Reflection as a Mediating Process

In the quantified-self movement, self-monitoring is instrumental to changes in behavior.
By shifting awareness to patterns of behavior one can better self-regulate either through
increased motivation, deeper insights, or other persuasive aspects. While there are a variety
of semantic variations in terminology, self-reflection can be considered a specialized subset
of self-monitoring focused on internal mental events rather than behaviors or sensory data.
Self-Reflecting and Mindfulness 89
Adopting the Self-Reflection and Insight Scale (SRIS) operationalization of self-

reflection we are defining self-reflection as having certain characteristics. First, while
self-reflection can be seen an involuntary process as is the case with rumination that is
correlated with a variety of mood disorders [11] we are viewing self-reflection as a
voluntary effortful process. Second, the focus of self-reflection is understood as one that
is conceptual – there is a top-down attentional focus that involves language and
appraisal, rather than a focus on experiential attention. We are assuming self-reflection
involves dealing with the contents of mental events (conceptualization). Third, self-
reflection is being seen as free from any regulatory or problem-solving orientation. In
the SRIS instrument, the second factor of insight captures one’s ability to understand
and make sense of their thoughts and emotions, while self-reflection seeks only to
capture one’s awareness of such mental events.
2.3 Designing for Self-Reflection

There are a variety of design approaches to promoting self-reflection while situated in
everyday life. Reminders to self-reflect on current mental events or particular ones are
a popular persuasive approach. Examples include, Mood Panda’s [12] reminders to track
current moods and Conscious app’s [13] initial daily directive to self-reflect on and
ongoing reminders to reengage effort to self-reflect on that particular directive. Another
approach goes beyond tracking and attempts to provide guidance on the self-reflective
process, e.g., the Mindfulness App [14] prompts users to engage in short 1-minute guided
meditations. Lastly, there is an approach that does not attempt interject into the everyday
experience of the user, but provides an on-demand library to help mediate one’s inter‐
action with their situated context. In the Buddhify mobile app [15], users can select from
meditations categorized by goals and context (e.g., traveling or feeling sick).
3 Study Design
3.1 Theoretical Framework
Our study’s goal was to obtain a rich understanding of how specific design embodiments
in our tool influence curiosity and decentering through self-reflection.
In designing our tool, we laid out a conjecture map that outlined key design embodi‐
ments, their relation to mediating processes, and their relation to our target outcomes
(Fig. 1). The relationship we anticipated between embodiments and mediating processes
are our design conjectures, and the relation between mediating processes and outcomes
are our theoretical conjectures. The objective of specifying design conjectures is to trace
the observed effects back to the embodiments. This can be challenging in that each
conjecture is implemented in tools and activities in ways that their operation is not
isolated – in turn making the study of each of the pieces separately not possible [16]. In
short, our design seeks to provide specific mental events for users to self-reflect on and
explore how situating this self-reflective process in everyday life in a particular way
may influence how mindfulness states are experienced.
Fig. 1. Conjecture map of mobile tool design
3.2 Mobile Mindfulness App Design
The mobile app engages users in five different challenges each targeting different mental
event. All of the challenges shared the same design embodiments but they differed in
the specific mental event targeted (see Table 1).
Table 1. A list of challenges, properties, and sequence of interactions

Mental Event Valence Scope Sequence of Interactions
Gratitude Positive Broad Recall moment of gratitude in your mind
Focus on breathing for 15 s
Conflict Negative Relational Recall moment of conflict
Focus on your breath for 15 s
Self-Compassion Positive Broad Recall moment of self-compassion in your mind
Self-Criticism Negative Broad Recall moment of self-criticism
Envy Negative Narrow Recall moment of envy in your mind
Given we are viewing the self-reflective process as conceptual, the type of mental
event is especially relevant. A review of mindfulness-based interventions [3, 17] and
prior pilot testing led us to outline five different mental events categorized by valence
and scope. Valence describes whether the mental event is likely to be linked to a positive
or negative affective state. Scope describes how broad or narrow the mental event is
likely to be characterized by the user. For instance, self-criticism is a broad mental event
Fig. 2. The daylong life cycle of a challenge start and completion
that can encompass other events such as envy. In addition we have a relational scope
that is both narrow and entails focusing on “conflict” with another person.
As illustrated in Fig. 2, a challenge is started by the user at the beginning of the day
and entails setting the intention to observe a specific mental event (e.g., self-criticism)
by accepting the challenge.
Fig. 3. Check-in. Step 1 – Recalling moment. Fig. 4. Check-in. Step 2 – Focus on breath.
After accepting the challenge, users are prompted throughout the day via reminders to
make an effort to observe that mental event. No action needs to be taken on part of the user
when the mental event is observed. At the end of the day, the user is prompted to check-in,
which consists of two-steps. First they are prompted to recall one moment where they
observed the mental event (Fig. 3), and second to focus on their breath for 15 s and let that
mental event recalled go as they shifted their focus (Fig. 4). The attentional shift to breath
is a common attentional control exercise used in mindfulness activities to promote decen‐
tering [18]. While throughout the day the reminders seek to prompt self-reflection and
curiosity on specific mental events, the end of the day interaction seeks to use the observed
mental events to engage users in a decentered form of self-reflection.
3.3 Measures
Mindfulness State. Participants’ achieved levels of curiosity and decentering was
assessed both quantitatively and qualitatively. Quantitatively the 6-item Curiosity
subscale and 7-item Decentering subscale of the Toronto Mindfulness Scale (TMS) [5]
was used. The curiosity subscale asks participants to express how well what they expe‐
rienced is described by items such as “I was curious about my reactions to things”, while
the decentering subscale asks participants items such as “I experienced myself as sepa‐
rate from my changing thoughts and feelings”. Items on the subscales are rated on a
5-point Likert scale (0 = “not at all”, 4 = “very much”). Scores on the subscales summed.
Qualitatively, participants engaged in a semi-structured interview that prompted them
to describe the details of their curiosity and decentering experiences with interview
questions adapted from the TMS subscale questions.
Self-Reflection. Self-reflection for each challenge was measured primarily through a
semi-structured interview that focused on the how self-reflection was experienced. In
addition, the 12-item self-reflection subscale (SRIS-SR) of the Self-Reflection Insight
Scale (SRIS) [19] was used. Self-reflection items include; “It was important to me to try
to understand what my feelings mean”. The items are on 7-point scales (1 = Strongly
Disagree, 7 = Strongly Agree). Lastly, we added some customized survey questions that
sought to elicit insights on the frequency of moments of self-reflection.
Utility. At the end of the study each user completed ratings of how useful the different
design embodiments were to supporting self-reflection (See Table 3).
3.4 Participants
We recruited 11 participants from several New York City universities. Participants were
required to have never previously meditated or engaged in cognitive behavioral therapy,
as well as have a personal iPhone they can use for the study. The average age was
(M = 25.09, SD = 3.86), and 7 out of the 11 participants were female (64 %).
3.5 Procedure
Participants were asked to complete an initial survey when signing up to ensure they
had not engaged in any form of meditation or cognitive-behavior therapy prior to this
experience. Each user was prompted to complete all of the five challenges over one
week, and complete a survey each time a challenge was completed. The survey included
questions that span the TMS and SRIS-SR measures, as well as customized questions
that focused on the frequency of self-reflection. The order of the challenges completed
by users was counterbalanced to minimize any learning effects. After completing all of
the challenges, users completed a utility survey and engaged in an hour-long semi-
structured interview. The interview protocol was structured to address each of the design
and theoretical conjectures.
4 Results
This mixed method study was structured as sequential explanatory [20] in which the first
phase was to analyze our quantitative data so as to inform subsequent semi-structured
interviews conducted with all of the eleven participants (Table 2).
4.1 Phase I: Quantitative Analysis
Curiosity, decentering and SRIS-SR scores for each mental event type were all normally
distributed, as assessed by Shapiro-Wilk’s test (p > .05). In addition, when running a
repeated measures ANOVA on curiosity, decentering, and SRIS-SR, we used Green‐
house-Geisser correction. For curiosity, post hoc tests using the Bonferroni correction
revealed that conflict significantly differed from every other mental event type in curiosity
(p < .005). For decentering, post hoc tests using the Bonferroni correction revealed that all
of the challenges differed from each other by mental event type (p < .05), except between
self-compassion and self-criticism. Lastly, for SRIS-SR post hoc tests using the Bonfer‐
roni correction revealed that all of the challenges differed from each other by mental event
type on SRIS-R scores, except for Gratitude and Self-Compassion (p < .0005).
Table 2. Summary of Quantitative Analysis

Mental Event Curiosity Decentering SRIS-SR
All (F(3.056, 30.557) = (F(2.913, 29.135) = (F(2.804, 28.039) =
5.462, p < .005) 4.114, p < .05) 12.937, p < .001)
Self-Criticism M = 17.64, SD = 4.86 M = 16.73, SD = 3.23 M = 49.27, SD = 3.38
Envy M = 17.27, SD = 4.13 M = 15.73, SD = 2.70 M = 47.00, SD = 3.85
Self-Compassion M = 17.27, SD = 3.29 M = 16.64, SD = 3.78 M = 51.91, SD = 4.34
Gratitude M = 17.36, SD = 4.01 M = 18.27, SD = 3.29 M = 51.73, SD = 4.13
Conflict M = 12.64, SD = 3.11 M = 13.09, SD = 3.81 M = 44.27, SD = 3.26
The self-reflection in-action (SR-InAct) question served the purpose to inform subse‐
quent interview questioning only, rather than be used as part of an inferential analysis.
As such only the means were calculated. Negative thinking (M = 3.91, SD = 1.221) and
gratitude (M = 3.91, SD = .701) were slightly above neutral. Self-compassion had the
highest mean (M = 4.45, SD = .688), and envy (M = 1.36, SD = .505) and conflict
(M = 1.27, SD = .467) were on the lower end.
The mean ratings on the utility of the various design embodiments were 4 or above,
indicating that no embodiment did not contribute to self-reflective practices. In looking
at the response rate of end-of-day check-ins that required recalling the moment and
engaging in breath awareness, about 96 % of end-of-day checkin reminders led to a
completed challenge, while the remaining were incomplete and required an repeated
enrollment in the challenge – two challenges in the entire study.
Table 3. Utility for Design Conjectures

Question Rating
By accepting the challenge I felt I was setting an intention to self-reflect M = 4.18, SD = .60
Setting an intention to self-reflect made it more likely I would self-reflect M = 4.09, SD = .70
Without the reminders I would have not self-reflected throughout the day M = 4.18, SD = .75
My self-reflection experience was different based on the mental event M = 4.27, SD = .46
Recalling a moment was easier when I self-reflected throughout the day. M = 4.45, SD = 52
Focusing on breathing was helpful in changing the focus of my attention M = 4, SD = .77
4.2 Phase II: Qualitative Analysis
The eleven semi-structured interviews were transcribed and analyzed using analytic
codes that were directly mapped to the five design conjectures and two theoretical
conjectures illustrated in Fig. 1. Each coded interview was treated as a case with the unit
analysis as each user’s single experience with the mobile app. Cases were individually
analyzed for themes and concurrently arranged in a word table for cross-comparisons
and shared themes [21] across the seven different conjectures. From the word table a
review of themes was conducted and a list of the major themes contributing to the initial
research questions were outlined and are discussed below.
5 Discussion
The goal of this study is to better understand how specific design embodiments enacted
self-reflection and in turn influenced curiosity and decentering experiences. While
overall users scores on the TMS and SRIS-SR indicated high levels of curiosity and
self-reflection, and varied levels of decentering, we focused on how such processes were
experienced, and how specific embodiments shaped these experiences.
5.1 Perceived Control
The theme of perceived control is about the extent to which users felt they could handle
the degree of self-reflection asked by the challenges. All of the users felt strongly that
initially accepting the challenge felt like they were formally setting the intention to self-
reflect. In addition, users felt “strongly” or “very strongly” that setting an intention for
the day was essential to feeling they would self-reflect.
In our interviews, the length of the intention (i.e., day), and clarity on what was
expected was highlighted as important to their perceived control over self-reflection.
For instance one user shared that he, “would have lost steam if I was committing to
something for more than a day, but I took it one day at a time.” Another user shared she,
“liked how simple it seemed at first” and another user said, “I could understand what I
was supposed to do, like, if I see it, just acknowledge I observed it.”
The use of reminders played into this perceived control in that forgetting was consis‐
tently mentioned, as a reason why they thought self-reflecting throughout the day would
be challenging. Over 90 % of users felt strongly that without the reminders they would
not have self-reflected throughout the day on specific mental events. This is line with
research that suggests that when situated in everyday contexts there may be a lack of
cognitive resources that can serve as an obstacle to challenging automatized cognitive
habits, such as not being aware of specific mental events [22].
The implications for design suggest there is value in the use of reminders to prompt
situated self-reflection in general. In addition, our findings suggest that short temporal
commitments (e.g. daylong) and clear implementation intentions may help create a
strong sense of control over being able to self-reflect throughout one’s often busy and
attention-demanding day. In addition, the use of reminders when motivation obstacles
are not apparent may be effective in supporting one’s perceived sense of control and in
turn strengthen one’s intention to observe internal mental events.
5.2 Likelihood of Opportunities to Observe

Across all of the challenges curiosity scores were high yet we did notice differences in
self-reflection scores (SRIS-SR). The pattern was that challenges that targeted narrow
mental events had lower self-reported engagement in self-reflection. In other words,
there seemed to be difficulty with self-reflecting on mental events we categorized as
“narrow.” Mean scores of our self-reflection in-action question asking users to rate the
degree to which they self-reflected throughout the day supports this findings. At the
lower end of all the challenges was envy (M = 1.36, SD = .505) and conflict (M = 1.27,
SD = .467).
In our interviews users consistently shared that self-reflecting was hard for envy
because it did not seem to come up too often in their day-to-day interactions. One user
did state that outside of the study they began to notice mental events he would charac‐
terize as “envious.” One user shared that, “I feel like I used to feel envy a lot, but now
I have this job and I’m not about that.” Another user described the challenge as, “looking
for something and [I] didn’t know if I was making it happen because I was looking for
it.” In other words, for most users envy just did not occur too often and as a result they
experienced difficulty in self-reflecting and rated the activity of recalling observed
moments challenging.
However, in exploring why curiosity levels remained high despite low self-reflection
scores we understood that for this narrow mental event, the reminders served less as
reminders to observe, but more as an invitation to think deeper on what they might be
missing or how they may be interpreting events. In other words, as one user put it, “I
just didn’t notice moments of envy. But then I was like, maybe I’m just not realizing
their envy, you know? So I just started thinking back, like maybe I’m missing something.
Maybe I’m just totally not labeling it right.” Other users shared similar sentiments of
addressing the difficulty with self-reflecting on envy by retrospectively looking back
and trying to see if they were incorrectly characterizing certain thoughts and feelings
they had. The low scores on the SRIS-SR subscale which removes motivation from
picture [23] was, most likely, sensitive to challenges in observing, while the curiosity
subscale picked up on the investigative aspect of one’s reaction to the challenge. We

noticed the same thematic pattern with the mental event type, conflict, which we labeled
as “relational”, although to a lesser degree.
The implications for design here are that personalization or on-demand interactions
may better address differences in the likelihood of engaging in particular kinds of mental
events. In addition, future research on how low likelihood mental events may positively
impact motivation to self-reflect may be worth exploring.
5.3 Self-Regulation Impulse

In our phase I quantitative analysis we saw variations in decentering scores between
challenges by mental event type. The post-hoc analyses did not yield any additional
insight in that all of the types significantly differed from each other. In looking at means
scores we noticed that gratitude (a positive valence event type) hovered above all the
rest (M = 18.27, SD = 3.289). Our initial thought was that all positive valence types
might have higher decentering scores, yet in the middle were self-criticism and envy
and self-compassion with means all hovering around 16, and the lowest decentering
score was the conflict challenge (M = 13.09, SD = 3.81).
From our interviews what emerged was a theme of self-regulation impulse that was
connected to how negative the user felt the particular mental event was. For instance,
one user shared, “I felt like the point of the gratitude one was to feel good, but I didn’t
see what the point of the other one – the criticism one. I was just wanting to, you know,
change it.” This is in contrast with another user who described their experience on self-
reflecting as, “I was okay just sitting with that thought and moving on.”
All of the users shared the sentiment that focusing on self-criticism made them want to
be proactive in some way. This is inline with research that suggest there may be two different
kinds of self-reflection – one that is problem focused (PF) and the other that is self-focused
(SF) [23]. In describing how this impulse to self-regulate influenced their sense of thoughts
as passing or as separate from themselves, users repeatedly described feeling attached to
negative thoughts, while being able to let go of positive ones. For instance one user shared,
“In the gratitude one I felt it in my breath and I moved on. But in the other one. The critical
one. I was like, I should stop doing that.” We sought to understand why self-compassion,
which we labeled as positive, did not fall into our hypothesized pattern that positive valence
mental events did not elicit strong impulses to self-regulate. Users shared that they felt self-
compassion overlapped with self-criticism in that many were self-compassionate about
mistakes or things they were saying to themselves. This indicates that the valence of mental
events may be more complicated to clearly delineate as positive or negative.
In discussing the conflict challenge with users, which had the lowest decentering
score was the conflict challenge (M = 13.09, SD = 3.81), there was by far dominant
focus on problem-solving rather than acknowledging the presence of specific mental
events. This suggests that prompting users to self-reflect on relational instances, such as
conflict between people, may bring the focus to problem solving and exacerbate
impulses to self-regulate.
The implications for design are that focusing on mental events that are likely to
induce negative affect, may benefit from embedded emotion regulation activities that
can leverage impulse to self-regulate. In addition the findings also suggest the need to
explore differentiated interactions to support decentering. Given all five mental event
types used the same recall and breathing exercises, perhaps extended or guided decen‐
tering activities are needed to counteract ones desire to self-regulate.
5.4 Shifting Attention Through Body Awareness
Overall breath awareness was described easy to engage in. In our utility ratings, most
users reported that, “Focusing on breathing was helpful in changing the focus of [their]
attention.” Interviews indicated that body awareness provided a general sense of calm
and detachment before dealing with negative mental events. One user stated, “I felt like
I had to remember something annoying or shameful, and then just let it go by focusing
somewhere else.” Another described it as, “the breathing was like at school when you
have to take a minute before you do something you’ll regret. Except here I think you do
it so you could just be okay with what went down.” The design implication here is that
body awareness may be a powerful design embodiment in shifting attention during an
end-of-day reflection and leading to a sense of “letting go” and “being with it”, which
are in line with decentered experiences.
6 Conclusions and Future Work
A limitation of this study is the limited data points on users’ engagement with the chal‐
lenges and mental event types. An increased number would have provided a larger lens
and reduction in confounding situational factors (e.g., a rough day at work) in the data.
Furthermore, there was no experimental design that may have allowed for different
causal claims to be made about specific design embodiments. In addition, the use of
SRIS-SR and TMS have their limitations. SRIS-SR is not written as a state-based retro‐
spective instrument, and TMS is intended for primarily meditation-based experiences.
Lastly, there was limited control over situational factors during sample selection (e.g.
heavy commuters, demanding work schedules).
Future studies may want to explore the potential for such situated approaches to
harness self-regulatory impulses to connect to more traditional forms of cognitive
behavioral strategies such as cognitive reappraisal. Recent research has made the case
that mindfulness states may support cognitive reappraisal [24]. Furthermore, an exper‐
imental design comparing different forms of decentering, and reflection prompts may
provide deeper insights into how variations in language, frequency, and body awareness
interactions can influence curiosity and decentering through self-reflection.
References
1. Van Dam, N.T., Earleywine, M., Borders, A.: Measuring mindfulness? An item response
theory analysis of the mindful attention awareness scale. Pers. Individ. Dif. 49, 805–810
(2010)
2. Bodhi, B.: What does mindfulness really mean? A canonical perspective. Contemp.
Buddhism. 12, 19–39 (2011)
3. Kabat-zinn, J.: Mindfulness-Based Interventions in Context: Past. Present, and Future 10,
144–156 (2003)
4. Sharples, M., Taylor, J., Vavoula, G.: A theory of learning for the mobile age. In: Sage
Handbook of Elearning Research, pp. 221–247 (2010)
5. Lau, M., Bishop, S.R., Segal, Z.V., Buis, T., Anderson, N.D., Carlson, L., Shapiro, S.,
Carmody, J., Abbey, S., Devins, G.: The toronto mindfulness scale: development and
validation. J. Clin. Psychol. 62, 1445–1467 (2006)
6. Garland, E.L., Hanley, A., Farb, N.A., Froeliger, B.: State Mindfulness During Meditation
Predicts Enhanced Cognitive Reappraisal. Mindfulness (N. Y) (2013)
7. Teasdale, J.D., Moore, R.G., Hayhurst, H., Pope, M., Williams, S., Segal, Z.V.: Metacognitive
awareness and prevention of relapse in depression: empirical evidence. J. Consult. Clin.
Psychol. 70, 275–287 (2002)
8. Kuan, T.: Mindfulness in Early Buddhism: New Approaches Through Psychology and
Textual Analysis of Pali. Routledge, Chinese and Sanskrit Sources (2007)
9. Hofmann, S.G., Asmundson, G.J.G.: Acceptance and mindfulness-based therapy: new wave
or old hat? Clin. Psychol. Rev. 28, 1–16 (2008)
10. Chiesa, A., Serretti, A., Jakobsen, J.C.: Mindfulness: top-down or bottom-up emotion
regulation strategy? Clin. Psychol. Rev. 33, 82–96 (2013)
11. Watkins, E., Brown, R.G.: Rumination and executive function in depression: an experimental
study. J. Neurol. Neurosurg. Psychiatry 72, 400–402 (2002)
12. MoodPanda: MoodPanda (2015). http://www.moodpanda.com/
13. Makan Studios: Consciouss (2014). http://www.consciousday.com/
14. Mindapps Inc.: The Mindfulness App (2014)
15. Everywhere, M.: Buddhify (2015)
16. Sandoval, W.: Conjecture mapping: an approach to systematic educational design research.
J. Learn. Sci. 23, 18–36 (2014)
17. Cullen, M.: Mindfulness-based interventions: an emerging phenomenon. Mindfulness 2,
186–193 (2011)
18. Tang, Y.-Y., Posner, M.I.: Attention training and attention state training. Trends Cogn. Sci.
13, 222–227 (2009)
19. Grant, A.M., Franklin, J., Langford, P.: The self-reflection and insight scale: a new measure
of private self-consciousness. Soc. Behav. Personal. an Int. J. 30, 821–835 (2002)
20. Creswell, J., Clark, V.P.: Choosing a mixed methods design. In: Designing and Conducting
Mixed Methods Research, pp. 53–106. Sage Publications (CA) (2007)
21. Yin, R.: Case Study Research: Design and Methods. Sage Publications, Thousand Oaks
(2009)
22. Holland, R.W., Aarts, H., Langendam, D.: Breaking and creating habits on the working floor: a
field-experiment on the power of implementation intentions. J. Exp. Soc. Psychol. 42, 776–783
(2006)
23. Grant, A.M., Unit, C.P., Wales, S., Franklin, J., Langford, P., Psychology, C.: The self-
reflection and insight scale: a new measure of private self-consciousness. Soc. Behav. Pers.
30, 821–836 (2002)
24. Garland, E.L., Hanley, A., Farb, N.A., Froeliger, B.: State mindfulness during meditation
predicts enhanced cognitive reappraisal. Mindfulness 6, 234–242 (2013)
Alcohol Behaviour Change: Lessons Learned
from User Reviews of iTunes Apps
Omar Mubin1(B) , Abdullah Al Mahmud2 , and Muhammad Ashad Kabir3

1
Western Sydney University, Parramatta, Australia
[email protected]
2
Swinburne University of Technology, Melbourne, Australia
[email protected]
3
Charles Sturt University, Bathurst, Australia
[email protected]
Abstract. Mobile based persuasive technology can help us to shape

positive behaviour and induce habit cessation. This paper reports on
the preliminary content analysis of the comments made by users on a
set of 18 iTunes apps that were designed to attempt to reduce the over
consumption of alcohol and ultimately cut down drinking. In total 204
comments were retrieved from the set of 18 applications using data gen-
erated from a custom batch script. Our main results from the content
analysis show that an efficient user interface is imperative to facilitate
the user acceptance of persuasive mobile systems that attempt to inhibit
consumption of alcohol. Furthermore, we noted more positive comments
towards apps that adopted a self control behavioural change strategy,
particularly as they followed a subtle and not abrupt interaction style.
We conclude our analysis by providing a list of design recommendations
for mobile apps that can assist in inhibiting alcohol consumption. Our
analysis indicated that customisation and the possibility of maintaining
incremental milestones were amongst the more sought after app features.
Keywords: Alcohol · Behaviour change · Mobile computing · Mobile

apps · iTunes
1 Introduction
Binge drinking, drink driving, alcohol related violence, alcohol related illness
are some of the many ill effects of the over consumption of alcohol [17]. Var-
ious intervention strategies operate on a governmental level or through non-
governmental organisations with the ultimate aim of promoting awareness and
reminding people of the side effects of consuming alcohol in large amounts. In
addition, technology (such as computing, mobile, smart systems, etc.) has the
potential to transform how we deliver alcohol interventions and promote behav-
iour change from harmful alcohol for example [21]. In today’s digitally connected
world with a high penetration of smart phones, mobile technology can be utilised
as an efficacious tool as it provides users with the right information at the right

DOI: 10.1007/978-3-319-31510-2 9
100 O. Mubin et al.
time allowing them to make informed decisions; in particular within the health
domain [19]. This ultimately promotes sustained behaviour change. Newer and
newer mobile apps are emerging day by day and they can provide obvious merits
to change behaviour by providing appealing interfaces and appropriate features.
Persuasive technology or captology is broadly defined as a stream of technology
that is designed to change attitudes or behaviors of the users through persuasion
and social influence, but not through coercion [7]. According to BJ Fogg [6] tech-
nology can persuade users by adopting certain guidelines; for example: increasing
self efficacy, providing tailored information, triggering decision making, guiding
users through a process, etc. In sum the goal of persuasive technology is to ini-
tially provide motivation and consequently the ability to change behaviour [8].
Technologies have had a significant impact in alleviating various social issues
(obesity, well being, sustainability, environmental awareness, etc.).
Accessing information and support for alcohol related issues through technol-
ogy deviates away from traditional rehabilitation procedures and therapy as it
can be done autonomously, anonymously and conveniently. The option of remain-
ing anonymous is important in sensitive areas such as substance abuse because
it is often easier to get information anonymously via an interactive computing
program, than face another human being [7]. Furthermore, the accessibility of
information is not confined to a physical location nor is it tied to a specific
time-frame. A self-administered and stand-alone intervention helps to eliminate
various barriers such as the stigma of attending a treatment facility, the high
cost of treatment, and scheduling and transportation issues. Currently there are
mostly web/desktop-based interventions in place to assist in persuading indi-
viduals to reduce their consumption of alcohol and inform them on the effects
of alcoholism [3]. The downfall with web/desktop-interventions is firstly their
feasibility and inability to be conveniently accessed at the end-users own discre-
tion. Secondly, customisation is usually low for such web based interventions -
personalisation is now being considered as an integral component of mobile based
behavioural change applications [11]. Consequently, it is imperative to build effi-
cacious intervention programs (possibly mobile based and on the go) to stimulate
or help sustain positive behaviour change for drinkers who do not connect with
support service as there is evidence that mobile intervention is suitable to recover
from alcoholism [12].
We have noticed that governmental organisations such as the National Health
Service England (NHS) [15] are slowly beginning to recognise and advocate the
importance of providing mobile apps for promoting reduced consumption of alco-
hol. However, little is known about the effectiveness of such apps especially for
behaviour change and user insights are relatively unexplored [5]. As the target of
those apps are a specific user group, it is very crucial to acknowledge their feed-
back in order to judge the quality of those apps. It has been observed that some
mobile apps are not designed according to the established rules so that they can
effectively change a particular behaviour. For example, a study of iPhone apps
for smoking cessation found that certain apps are not designed based on the
recommended guidelines for smoking cessation [1]. Similar violations of design
principles can be seen in apps from other domains [22]. Within Human Computer
Alcohol Behaviour Change: Lessons Learned from User Reviews 101
Interaction (HCI) and Mobile Computing the role of mobile apps in promoting
health and well-being is an acknowledged and promising research area. However
more recently [18] efforts are being dedicated in achieving a deeper understand-
ing of the implications of health monitoring through mobile applications, more
specifically recognising that the solution is not in simply providing digital infor-
mation but rather carefully designing the visualisation, presentation and modal-
ity of the information. In sum, although researchers have begun to understand
behavioural change techniques that result in a reduction of alcohol consumption
through mobile apps [13]; it is still necessary to look for user preferences and
opinions concerning those and other similar mobile apps.
End user review is very important in order to understand users’ perception of
a particular product. It has been found that end user reviews play a vital role in
order to buy items from web stores such as Amazon [14]. End user reviews appear
on app store as ratings, comments and sometimes with visual marks of goodness
or badness, which may influence the number of downloads of those apps. We
are not aware of any study which particularly focused on understanding mobile
app reviews on different apps designed to change alcohol consumption. Such an
investigation can provide us with some answers as to what design features are
the most efficient in incurring behavioural changes towards alcohol intake. There
are some studies which have investigated mobile apps in order to understand the
process of user ratings and user opinions [9]. Using analysis of user reviews on
social media (specifically iTunes, youtube, etc.) to determine and elicit user
insights, comments, feedback and complaints is slowly gathering interest in the
field of Human Computer Interaction and Mobile Computing [10]. With the
proliferation of Apple devices such as the iPhone and the popularity of the App
Store (as evidenced by 100 billion app downloads [2]) we proceeded with iTunes
as the platform of choice for our analysis.
As the first step in our design process towards creating a persuasive mobile
application (app) to control the consumption of alcohol we aim to understand
the impact of existing mobile applications to support the abstinence of over con-
sumption of alcohol. We also aim to realise what particular design features and
persuasive strategies are most preferred by the users. In order to determine such
insights from users we extracted comments from iTunes on a set of short listed
existing mobile applications of varying persuasive strategies that were aimed
to control drinking behaviour of its users. The comments were then analysed
using content analysis. In conclusion we discuss design guidelines and recom-
mendations for developing such mobile apps and attempt to ground our findings
through an overview of commercial apps recommended by the NHS in the UK
which are meant to control alcohol consumption.
2 Method
Our research methodology was primarily based on the prior work of [4] where
the authors conducted a needs assessment to examine mobile smart-phone apps
that promoted alcohol behavior change. Furthermore, in their research a meta
102 O. Mubin et al.
level categorical analysis of iTunes apps designed to mitigate and inhibit alcohol
addiction was conducted to understand the various behavioural strategies they
employ to control drinking behaviour. However, an analysis on user opinions,
feedback and insights was missing and they did not present any user interface
guidelines to inform the design of such mobile apps which is vital given the
explosion in the number of smart phones. Therefore, we build on top of the work
of [4] by conducting content analysis on a set of user reviews on relevant iTunes
apps with a goal of identifying recommendations to inform the design of mobile
apps that inhibit the over consumption of alcohol.
2.1 App Selection

The apps were retrieved through iTunes by utilising a set of predefined keywords.
We utilised the following set of keywords to search for applications: alcohol self
help (4), alcohol abstinence (3), alcohol recovery (53), substance abuse (24), quit
drinking (31), alcohol harm reduction (73), quit alcohol (22), alcohol moderation
(0), alcohol anonymous (65). All keywords were adapted from those discussed
in [4]. The number of unique applications retrieved for each keyword are provided
in brackets. There maybe additional keywords that could have been employed
in the search however it is our opinion that through the existing list of keywords
a fairly exhaustive set of apps was retrieved. In fact the appearance of an app
via specific keywords was not mutually exclusive and apps would be listed for a
number of different keywords.
In the work of [4], two primary categories of apps were considered for analy-
sis; firstly apps which facilitated alcohol consumption and secondly apps which
provided an intervention. Overall, in iTunes there are a number of applications
relevant to alcohol consumption however our focus was on applications of mainly
a persuasive type that can change the behaviour of people in a positive way
both in the long term and short term. Therefore we ignored any app that would
facilitate alcohol consumption (such as recipe apps, etc.) or that would provide
general information about alcohol without providing the feature of assisting in
overcoming a drinking habit. Categorisation of an app into one of these cate-
gories was enabled by reading the basic description of the app in iTunes and
through the definitions of each description in [4] it was ensured that this was a
mutually exclusive process. This filtration resulted in a set of 26 applications. In
order to automate the extraction of comments, we wrote a simple batch script
that would query the iTunes API for the comments on each app on the basis of
an app ID. 8 apps were depicted by the iTunes API as not having any registered
comments.
Each application from the set of 18 remaining applications was then cate-
gorised into one of the following evidence-based subcategories as suggested in
[4,5]: (1) motivational or counselling and (2) self coping or self control. Moti-
vational or counselling applications would provide insightful feedback to drive
behavioural changes whereas applications employing self control or self cop-
ing strategies utilised mainly metrics such as the Blood Alcohol Concentration
(BAC) to provide information to the user with little or no initiative towards
decision making. The number of apps that adopted a self control behavioural
change strategy were 6 in number. The intervention strategy employed by the
app (or simply the category of the app) was the independent variable of our
analysis as we wished to ascertain which of the two strategies were the most
effective and preferred by the users.
2.2 Coding Process
In total 204 comments (excluding comments that were junk, spam or not in
English) were retrieved from the set of 18 applications using data generated
from the script on March 27, 2015. 29 comments were from self control apps - see
details in Table 1. The average length of a review was 35.4 words. The maximum
length was 225 words and the minimum length was 2 words. The entire process
Fig. 1. Sequence of steps/process of shortlisting apps and comments
Table 1. Details of the chosen apps which were used to analyse user reviews
Keyword # of Keyword # of
Name Category Name Category
used Comm- used Comm-
of App of App of App of App
on iTunes ents on iTunes ents
Alcohol Self control Step Alcohol Motivational
Self Help 1 16
Self Help Self coping Away Reduction Counselling
How to Stop
Substance Alcohol Self control Alcohol Motivational
4 Drinking 1
Abuse Self Help Self coping Reduction Counselling
Alcohol
Stop Drinking
Alcohol Motivational Alcohol Motivational
With Andrew 36 Quit n Save 1
Reduction Counselling Reduction Counselling
Johnson
Alcohol Motivational Alcohol Motivational
iRecovery 13 I am Sober 50
Reduction Counselling Recovery Counselling
Alcohol Motivational Alcohol Self Control
Quit That 25 Addicaid 21
Reduction Counselling Recovery Self Coping
Addiction Alcohol Self Control New to Substance Self Control
1 1
Counter Reduction Self Coping Recovery Abuse Self Coping
Saying Alcohol Motivational Addiction Substance Motivational
3 12
When Reduction Counselling Avert Abuse Counselling
Alcohol Motivational Quit Self Control
Sober Tool 11 No Go 1
Reduction Counselling Drinking Self Coping
Quit Drinking Alcohol Motivational
7
Hypnosis Reduction Counselling
104 O. Mubin et al.
of shortlisting apps and comments is summarised in a block diagram (see Fig. 1).
Each comment was treated as one unit of data. Each comment was then coded
to determine:
1. valence (positive, negative or irrelevant) towards the effectiveness of the

behavioural change strategy and feedback by the user in actually being able
to carry out a behavioural change in their life.
2. valence (positive, negative or irrelevant) towards the application features,
design, user interface, usability, etc.
In order to verify our coding scheme two authors separately coded ≈ 20 %

of the comments. Cohen’s Kappa for valence for behavioural change strategy
(κ = 0.69) and for valence for application features (κ = 0.65) was computed
which indicated moderate agreement. Discrepancies in coding were resolved after
discussion.
3 Results and Discussion

A table of the categorised results (see Table 2) from the coding is provided sum-
marising the total number of coded comments across the two valence types and
the category of the application. A Chi-Square Test revealed that there was a
significant but weak association between valence attribution towards the appli-
cation features and the category of apps (χ2 (N = 204,2) = 9.64, p = 0.008,
Cramer V = 0.22). Apps that adopted a self control paradigm elicited signif-
icantly fewer negative comments on the application features as compared to
apps that followed motivational strategies (Z = −2.5). A second Chi-Square
Test revealed that the category of apps did not have a significant association
with the valence attribution towards the effectiveness of the behavioural change
strategy employed by the app (χ2 (N = 204,2) = 1.62, p = 0.45).
Table 2. Table summarising total number of coded comments
Valence RE effectiveness Valence RE

of the behavioural strategy application features
Positive Negative Irrelevant Positive Negative Irrelevant
Category of Motivational 82 8 85 82 41 52
application Self control 10 2 17 22 1 6
One of the main trends observed in our analysis was that in general users
preferred the user interface and design features of self control apps as compared
to apps that followed a motivational strategy. Prior work [5] has depicted sim-
ilar trends (albeit in health-based behavioural change applications), primarily
due to the user feeling agitated by over prompting or by an over zealous sys-
tem. It can be suggested that in such situations the user would desire control
and edit rights over the addiction control plan and strategy. Some of the negative
comments on the features of a motivational style of behavioral change referred
to the demanding nature of such applications, for example: “I have tried this
app several times it did nothing for me except give me some anxiety from the
fast pace and demanding tone used it gave me no positive results at all” or to
the inability of the strategy convincing them to incur a behavioural change,
for example: “It hasn’t really helped me to stop drinking”. Further research is
required to determine if the preference of the design features of self control and
self coping style of applications was due to a cause and effect relationship with
user interface inadequacies. It would not be naive to assume that implementing
motivational apps would be more challenging than apps of simple self reporting
or self recording nature. In our overview of the iTunes comments we noticed that
many users thought that motivational apps were “slow”, “useless”, “inaccurate”,
etc. The difficulty in implementing persuasive applications that facilitate behav-
ioural change explicitly and directly are known to be a technical challenge [20];
primarily due to real time monitoring requirements.
Many users preferred the simplicity in design of the self control and self
coping style of apps as indicated in the following comment: “The app is intuitive
and the design is bright and refreshing - nothing clinical or scary about it”. It
goes without saying that simple, engaging and motivational interaction (i.e., the
user interface) is imperative for the success of a persuasive mobile application -
as also emphasised in [16]. There were only 3 comments which were coded to
be positive in terms of the evaluation of their behavioural change strategy and
negative in terms of their application features. Overall, when an app had a usable
interface most users were positive about the prospects of being able to change
their behavior (only 1 comment which indicated otherwise).
3.1 Design Recommendations

We analysed user comments across both categories of apps to elicit specific yet
emerging patterns and we also extracted design features that were preferred and
recommended. A number of acquired design features that we present as design
recommendations are now listed with some sample comments:
– Customisation with respect to layout, options, features, alerts, plans, etc.

was one of the key guidelines. Ultimately users wished to have the ability to
modify both user interface and behavioral strategy aspects at their own will
(“personalisation feature is great and very effective”).
– Real time feedback was also deemed to be important, as users wished to
visualise the result of their behavioural changes instantly (“The app provides
immediate accountability”).
– The most preferred mode of interaction was relaxing, gradual, not abrupt and
soothing intervention. This was a key feature found in most self control apps
as the following comments illustrate: (“It not only helps me to relax and fall
asleep, but it has helped to reduce my desire for alcohol”) and (“This app has
such a soothing and intuitive design”).
106 O. Mubin et al.
– Incremental goals and targets also allowed users to set short term objectives
and perceive success on their completion: (“I love getting the alert every night
telling me I am one more day sober”) and (“I love the milestones. I set each
new one as my wallpaper”). This feature (i.e. milestones, daily goals, etc.) was
one of the most popular and sought after and was one of the key elements
appreciated in self control apps.
– Many users thought social interaction and the ability to communicate progress
through social media was important, as evidenced by the following: (“The only
thing I wish it did was allow me to post to a social network my progress, but
I worked around that by taking screenshots and posting those”) and (“Unless
I’m missing it, I’m really surprised that I can’t tweet or post anywhere for
that matter directly from app”).
– The ability to control other cravings besides alcohol through the same app
was also a desired aspect: (“Would be really nice if I could change it from
alcohol to something else”). Though we have seen several apps for alcohol
behaviour change a few apps were more generic. For example, they could
be used for multiple purposes such as smoke cessation, cutting down alcohol
consumption or quitting other bad habits.
As a final step in our analysis we sourced apps from the National Health Ser-
vice (NHS) England Health Apps Library [15] to benchmark, compare, ground
and contrast our findings. The Health Apps Library is aimed at people living in
England who want to use apps to manage their health; and health professionals
who may want to recommend an app to someone if needed. When we searched
the library we found in total eight apps which were associated with the keyword
alcohol and which have been reviewed by clinicians in the UK to ensure that they
are safe and trustworthy. Out of those eight apps there were three apps, which
were not related to alcohol behavior change rather they were either for general
health tracking or for creating awareness for other people. The apps, which are
related, to track or cut down alcohol consumption were 1. Leaf-your discrete
drink tracker, 2. DrinkCoach, 3. Change4Life drinks tracker, 4. Type 1 diabetes
friend, and 5. Drink meter. No user reviews were found for those apps except for
Change4Life drinks tracker. That particular app had a 4.5 start average rating
and had the following features: a. track drinks each day, b. view drinking levels
over a week or a fortnight to spot when someone is at an heightened risk, c. learn
more about the specific risks, d. get daily tips and feedback, e. add daily notes,
f. share with friends via Facebook or Twitter. In general users found the app
motivational and one person had been using it for two years. We noticed that
the overall goal of these five apps was to help people drink within the safe limits
rather than motivate them to completely quit. Clearly these five apps were of a
self control type relying on the user to him/herself manage their alcohol intake.
This is consistent with our earlier findings that motivational apps are not widely
implemented or deployed yet, probably due to user interface or user perception
issues.
3.2 Limitations and Future Work

There are a number of limitations of our study some of which are inherent to the
structure and setup of iTunes. The retrieval of comments for a particular app
through the iTunes API did not synchronise with what was seen as the com-
ments for that API through the iTunes website/software. In this paper, we have
only considered one source (i.e., the API) and aim to merge the two sources in
our future research. However, we did notice that the sample size of comments
retrieved from the API were much larger in number than what was being seen
through the iTunes software/website. The actual number of downloads and pop-
ularity of an app is also a measure that cannot be retrieved through iTunes. The
popularity can also not be determined by the level of usage of an app; we cannot
tell from a particular comment if that particular user is an expert or a novice
user of that app.
Furthermore, it is evident that self control apps were fewer in number and
hence the corresponding set of comments was smaller than motivational apps.
We aim to address this constraint by considering Android apps found on Google
Play. The limited number of user reviews across apps in general was one of
the reasons why we did not restrict the extracted reviews to a specific time
period. Another interesting observation while searching the apps was that we
found a number of varying behavioural change strategies (beyond the two we
discussed: motivational and self control) in the apps which were aimed to change
behaviour. In the future, it will be equally interesting to extend our analysis to
a wider variety of persuasive strategies. We would also like to acknowledge the
possibility that some apps could comprise of both motivational and self control
features. That may actually be quite beneficial allowing further opportunity for
users to customise their interaction. In our sample of 18 apps we did not observe
any to merge both types of behavioural strategies.
Although some comments were quite large we did not observe any comment
which had a merge of positive and negative emotions; most users were very
assured as to how they evaluated either the interface or the behavioural strategy
of the app. However the subjectivity of the coding process must be acknowledged.
One of the key limitations of our process of content analysis was that we did
not benchmark or crosscheck the accuracy of the comments made on an app by
actually analysing its interface. Despite a number of limitations we are of the
opinion that our design recommendations, trends and lessons can be of help for
future implementations of apps that attempt to counter the over consumption
of alcohol.
4 Conclusion
In this paper we have presented the preliminary content analysis of the com-
ments made by users on both self control and motivational type iTunes apps
that were designed to attempt to reduce the over consumption of alcohol and
ultimately cut down drinking. Our main results show that an self control apps are
more popular and were rated more positively primarily due to being less invasive
108 O. Mubin et al.
and more supportive of customisation. Users also found issues in the user inter-
face of motivational apps which may be due to the difficulty in implementing such
apps and sourcing real time information. The most sought after and preferred
feature was the ability of apps to support incremental progress using elements
such as daily milestones and goals. We have also summarised some key design
guidelines as a prospective checklist which would be useful for other researchers
and mobile app designers to incorporate in systems that aim to inhibit the con-
sumption of alcohol.
References
1. Abroms, L.C., Padmanabhan, N., Thaweethai, L., Phillips, T.: iphone apps for
smoking cessation: a content analysis. Am. J. Prev. Med. 40(3), 279–285 (2011)
2. App. Store (2015). http://www.theverge.com/2015/6/8/8739611/apple-wwdc-
2015-stats-update
3. Bewick, B.M., Trusler, K., Barkham, M., Hill, A.J., Cahill, J., Mulhern, B.: The
effectiveness of web-based interventions designed to decrease alcohol consumptiona
systematic review. Prev. Med. 47(1), 17–26 (2008)
4. Cohn, A.M., Hunter-Reel, D., Hagman, B.T., Mitchell, J.: Promoting behavior
change from alcohol use through mobile technology: the future of ecological momen-
tary assessment. Alcohol.: Clin. Exp. Res. 35(12), 2209–2215 (2011)
5. Dennison, L., Morrison, L., Conway, G., Yardley, L.: Opportunities and challenges
for smartphone applications in supporting health behavior change: qualitative
study. J. Med. Internet Res. 15(4), e86 (2013)
6. Fogg, B., Cuellar, G., Danielson, D.: Motivating, influencing, and persuading users:
An introduction to captology. Hum. Comput. Interact. Fundam. 9, 109–122 (2009)
7. Fogg, B.J.: Persuasive technology: using computers to change what we think and
do. Ubiquit. 2002(December), 5 (2002)
8. Fogg, B.J.: A behavior model for persuasive design. In: Proceedings of the 4th
international Conference on Persuasive Technology, pp. 40. ACM (2009)
9. Hoon, L., Vasa, R., Schneider, J.G., Mouzakis, K.: A preliminary analysis of vocab-
ulary in mobile app. user reviews. In: Proceedings of the 24th Australian Computer-
Human Interaction Conference, OzCHI 2012, pp. 245–248. ACM, NY, USA, New
York (2012)
10. Khalid, H., Shihab, E., Nagappan, M., Hassan, A.: What do mobile app. users
complain about? a study on free ios apps (2014)
11. Masthoff, J., Vassileva, J.: Tutorial on personalization for behaviour change. In:
Proceedings of the 20th International Conference on Intelligent User Interfaces,
pp. 439–442. ACM (2015)
12. McTavish, F.M., Chih, M.Y., Shah, D., Gustafson, D.H.: How patients recover-
ing from alcoholism use a smartphone intervention. J. Dual Diagn. 8(4), 294–304
(2012)
13. Michie, S., Whittington, C., Hamoudi, Z., Zarnani, F., Tober, G., West, R.: Iden-
tification of behaviour change techniques to reduce excessive alcohol consumption.
Addict. 107(8), 1431–1440 (2012)
14. Mudambi, S.M., Schuff, D.: What makes a helpful review? a study of customer
reviews on amazon. com. MIS Q. 34(1), 185–200 (2010)
15. NHS Health Apps Library (2015). http://apps.nhs.uk/apps/alcohol/
16. Oinas-Kukkonen, H., Harjumaa, M.: Persuasive systems design: Key issues, process
model, and system features. Commun. Assoc. Inf. Syst. 24(1), 28 (2009)
17. Organization, W.H., et al.: Global status report on alcohol and health-2014. World
Health Organization (2014)
18. Panel Mobile HCI 2015 (2015). http://mobilehci.acm.org/2015/panel.html
19. Toscos, T., Faber, A., An, S., Gandhi, M.P.: Chick clique: persuasive technology
to motivate teenage girls to exercise. In: CHI 2006 extended abstracts on Human
factors in computing systems, pp. 1873–1878. ACM (2006)
20. del Valle, A.C.A., Opalach, A.: The persuasive mirror: computerized persuasion
for healthy living
21. Wang, K.C., Hsieh, Y.H., Yen, C.H., You, C.W., Chen, Y.C., Huang, M.C., Lau,
S.Y., Kao, H.L.C., Chu, H.H.: Soberdiary: A phone-based support system for assist-
ing recovery from alcohol dependence. In: Proceedings of the 2014 ACM Interna-
tional Joint Conference on Pervasive and Ubiquitous Computing, pp. 311–314.
ACM (2014)
22. Wolf, J.A., Moreau, J.F., Akilov, O., Patton, T., English, J.C., Ho, J., Ferris, L.K.:
Diagnostic inaccuracy of smartphone applications for melanoma detection. JAMA
Dermatol. 149(4), 422–426 (2013)
Persuasive Strategies to Improve Driving Behaviour
of Elderly Drivers by a Feedback Approach
Perrine Ruer ✉ , Charles Gouin-Vallerand, and Evelyne F. Vallières

( )
LICEF Research Center, Télé-Université du Québec, Montréal, Canada

{pruer,cgouinva,evallier}@teluq.ca
Abstract. We are witnessing a growing aging population who wishes to live

independently. In a driving context, the elderly want to maintain an active life‐
style, but they may suffer from impairments due to aging. New intelligent trans‐
portation systems can be beneficial for drivers to assist them with driving. Hence,
new intelligent technologies have to be accepted and have to persuade the driver
of adopting safer driving behaviours. In this paper, we present a persuasive driving
feedback for elderly drivers. The feedback objectives are to evaluate the drivers’
perception of the driving fatigue and to compare it with their driving behaviour.
Our first results from an exploratory field experiment with twenty elderly drivers
support the use of the feedback with that category of drivers relative to fatigue
perception. The originality of this paper is to enable progress in the area of
persuasive technologies applied to road safety and for elder people.
Keywords: Elderly drivers · Persuasive technology · Feedback · Contextual

information · Road safety
1 Introduction
Aging is a growing phenomenon in almost all countries of the world. The percentage of
people of 65 and older will more than double between 2010 and 2050 [1]. Elder people
want to continue to drive to maintain their independence, even more so for elderly drivers
living in rural or remote areas. Mobility is fundamental to them because it allows them
to maintain an active aging. Thus, it is in the best interest of societies to maintain elderly
adults driving as long as they can safely do so. In the last decade, there has been much
attention given to technologies focusing on elderly drivers’ mobility [2]. Indeed, the
development of transport technologies can address their mobility need by allowing them
to use safely their personal car as long as possible. For the past few years, there has been
a growing interest in intelligent vehicles. A notable initiative on intelligent vehicles was
created by the U.S. Department of Transportation with the mission of preventing
highway crashes [3].
Technologies can be useful as countermeasures to driving fatigue among elderly
drivers, as long as they induce behaviour changes. Persuasive technologies have this
feature. The association of persuasion strategies to technology is essential because
persuasive technologies can help changing user’s behaviour. To do so, it is essential to
take into account the characteristics of elderly people and to educate them on how to

DOI: 10.1007/978-3-319-31510-2_10
Persuasive Strategies to Improve Driving Behaviour 111
use technology and when [2]. However, in the road safety scientific literature, there is
a lack of knowledge about persuasive technology applied to elderly drivers.
Therefore, this paper presents a persuasive system that was designed for helping
elderly drivers to manage their fatigue at the wheel. The paper is structured as follow.
Section 2 presents an overview of persuasive technologies, in light of elderly drivers
and driving self-regulation. In Sect. 3, we present our methodology to design the persua‐
sive system. Section 4 reports our first results. Finally, Sect. 5 concludes with the limits
of this exploratory study and suggestions for future research in that field.
2 Related Works
2.1 Persuasive Technologies
In [4], Fogg defines persuasion as “an attempt to change attitudes or behaviours or both
(without using coercion or deception)” (p. 16). The aims of persuasive technologies are
to induce attitude change and behavioural outcomes. Changing these outcomes is central
(e.g. reinforce attitudes, act of complying, etc.). And the concept of motivational afford‐
ance (motivation to use the system) is affiliated with psychological outcomes and target
behaviours. All persuasive technologies focus on these final effects. The results can be
positive (partially or fully), negative or with no effect [5].
Oinas-Kukkonen and Harjumaa [6] define the context analysis as the grouping of the
persuasion’s intent, the understanding of the persuasion event and the definition or recog‐
nition in the use of ongoing strategies. First, persuasive technology involves a voluntary
change by the user of his/her behaviours [4]. Indeed, persuasive systems have to influ‐
ence, motivate or convince the user to take action. Besides the intent, the event of persua‐
sion and the strategy have to be considered for the design and the use of a persuasive
technology. The event concerns the usage and the user of the persuasive system. First, the
usage (or the context of use) refers to the problem domains’ attributes (for instance, promo‐
tion of health or ecological consumption). Then, an analysis of the user is also important.
Indeed, there are individual differences between people on how to interpret information.
For instance, there are differences in cognitive processes between younger and elder
people. These differences have an impact on the persuasion strategy which is defined as the
message and how it is delivered [5, 6]. Once the context analysis has been determined, the
next phase is to design the persuasive system. To do so, the information content and
the software functionalities need to be understood to design and evaluate the software
system persuasiveness.
2.2 Existing Persuasive Technologies in the Automotive Context
During the last 30 years, there has been growing progresses in the domain of in-vehicle
technologies. These technologies are likely to have used persuasive strategies. The best-
known persuasive technologies in automotive context are eco-driving technologies [7, 8].
In 2015, persuasive technologies have been designed in order to improve users’ safe driving
behaviours and their motivation. There are two basic workings. The first one is to give
real-time feedback to a current individual driving behaviour, for instance such as
112 P. Ruer et al.
recommendations. The second basic working is to accumulate information about a driver

behaviours and to give an accumulated feedback to the driver [7].
The most prevalent persuasive systems in automotive context are in-car interfaces
and mobile systems, like smartphone applications [7, 9]. Persuasive systems integrated
in a vehicle give real-time feedback with optical advices (for instance, a bar chart or
pictograms can appear on the dashboard to give a driving behaviour feedback). The
inconvenience of in-car technologies is that it is not installed in many cars, even though
in the next few years, we expect they will be increasing. As for persuasive mobile
systems, several applications have been developed to give driving feedback to the user
after the driving. The example presented by Schätzl [7] was an application which gives
information about speed curve over time and abrupt accelerations and decelerations.
The disadvantages of using mobile app is that it has to be downloaded, or to be manually
started before each driving session and it can be used only with a compatible smart‐
phone [9].
Nevertheless, persuasive technologies affect the driving behaviour and have a posi‐
tive influence, for instance on the fuel consumption [7]. Persuasive technologies can
promote driver behaviour and improve road safety, provided that the persuasive tech‐
nologies address appropriately the drivers’ characteristics and are well-accepted by them
[9]. Indeed, acceptance of technology is necessary, in order to avoid rejection by the
user [8]. But technologies could also improve the quality of life by overcoming age-
related deficits [10].
In sum, persuasive technologies can not only improve driving safety, save resources,
protect the environment but they can also improve the quality of life of elderly drivers
by overcoming age-related deficits.
2.3 Elderly Drivers
Elders have specific driving characteristics. They are prudent, drive at low speed and
respect the traffic laws and their driving style is low risk taking [11]. Their main causes
of accidents are errors which are defined as judgment and observation failures dangerous
to other drivers [12]. When driving, a lot of information is asked from the driver in a
time constraint (for instance: changing direction, insertion in traffic, intersection, etc.).
Elderly drivers have difficulties to manage the unexpected and accidents happen when
they are making decisions under this time constraint [12].
Drivers have to respond rapidly to risks, which requires good abilities like attention,
perception, motor abilities, information treatment, etc. With aging, some physiological
impairments appear and have negative impacts on driving skills [13]. The cognitive
skills are affected and lead to a longer reaction time, a diminution of attention, and a
shorter memory [14]. Physical abilities are worsening with a deterioration of psycho‐
motor skills, development of arthritis, which causes neck problems, or also the vulner‐
ability of the body [10]. And sensory functions decline with a diminution of visual acuity,
diminution of hearing or even, perceptual ability [14, 15]. These impairments are
common among the elderly with normal aging. In addition, there is an increase of
diseases, associated with medication in-take and a higher risk of accidents. Medication
alters the driving skills and reduces sensorimotor performance (for example: decreased
alertness, impaired vision, etc.) [10]. Therefore, some authors propose license restric‐
tions to manage elderly drivers’ safety [16], even though empirical research has clearly
showed that driving cessation had adverse negative consequences [10, 15, 17] as it is a
stressful experience having a negative impact on their quality of life.
2.4 Driving Fatigue Self-Regulation

Fatigue is one cause of elders’ accidents. Fatigue refers to a transitional state between
wakefulness and sleep leading to sleep if it isn’t interrupted [18]. Different models and
theories have been proposed in order to explain the safe and risky behaviour of elderly
drivers. Among them, a multifactorial model by Anstey and her colleagues [14] suggests
that both self-monitoring beliefs about one’s driving capacity and one’s real capacity to
drive safely are predictors of driving behaviour. Self-monitoring is determined by
cognition, whereas the capacity to drive safely is determined by the sensorial, physical
and cognitive variables. For instance, the driver whose capacity to drive safely is reduced
due to a visual deficit, while at the same time his or her beliefs about the capacity to
drive match his or her real capacities, then the driver is susceptible to self-regulate and
drive safely. However, if that same driver is unaware of the deficit, decides to ignore it
or overestimates his or her visual capacities, that driver is susceptible to have unsafe
driving behaviours.
Research has shown that elderly drivers, like other groups of drivers, continue to
drive even though they are drowsy or nodding off [19]. Furthermore, drowsy or fatigued
drivers use different strategies in order to keep awake, such as opening the window,
putting on the radio, talking to passengers, stopping to eat, to exercise or to relax, without
napping or sleeping. Yet, these strategies are the least efficient ones in order to counter
fatigue at the wheel, while the most efficient strategies, such as stopping to nap or sleep
or asking a passenger to drive, are generally avoided even though drivers know they are
the efficient ones [20, 21].
Persuasive technologies can be applied to feedback concerning specific driving
behaviours, for instance, those relative to fatigue, and for a category of persons, such as
elder people.
2.5 The Applicability of a Design Persuasive Method Through a Real Driving

Feedback
The design method we chose to apply to the development of a real driving feedback is
the Persuasive Systems Design (PSD) Process Model, proposed by Oinas-Kukkonen
and Harjumaa [6]. The advantages of this methodology are its usefulness to study feed‐
back with users in real-life contexts and its applicability to on-road situations.
Oinas-Kukkonen and Harjumaa have established a set of design principles that have
been classified in four categories adapted partially from Fogg’s work [4, 6]. They are
primary task, dialogue, system credibility, and social support. For each of them, Oinas-
Kukkonen and Harjumaa give the design principles, the requirements and offer imple‐
mentation examples. Primary task is defined by the authors as “the carrying out of the
114 P. Ruer et al.
user’s primary task”. For instance, the design-principle self-monitoring in the automo‐
tive context would be that the system should provide a feedback of the driving perform‐
ance (speed, braking, vigilance, etc.). The users can track their performance or status,
through a tool (in-car interface or mobile phone application), to improve decision
making.
Dialogue is the user’s feedback, via verbal-information or other kinds of summaries
to support and help the users to move closer to their target behaviour. An example of
the application of the praise principle to an elderly driver would be an image or some
words to motivate the person to adopt a better driving style. The other goal could be to
increase safe driving for the driver and the community (i.e. other drivers).
As for the system credibility, it allows designing a system that could be more
convincing for the users. For example, the fact that feedback can be transmitted to the
point of view of some authorities, such as a government transport ministry, could have
more persuasive power.
The fourth category, social support is about designing a system that will motivate
the users by leveraging social influence. An example of the social learning principle
applied to feedback for elderly drivers is a shared event log for encouraging good
behaviours.
3 Methodology
In our study, the intent is to persuade the elderly driver to manage appropriately his/her
fatigue on his/her subsequent conduct of the vehicle. We did a pilot study to evaluate
the perception of the driving fatigue with elderly drivers and compared it with their
driving behaviour. To do this, we did experimental work and we measured perceived
fatigue, speed, sleep habits, etc. During the experimentation, each participant drove
around 50 km on a closed driving circuit with the LiSA car (in French: Laboratoire
intelligent de Sécurité Automobile. In English: Intelligent Laboratory on Automobile
Safety) (Fig. 1). Participants were asked to respect a 60 km/h limit.
For the event, we took into account the driving context and the driver. In the context
of our research, we used an instrumented vehicle, the LiSA, a Nissan Versa 2008. The
car is equipped with a data logger AIM Evo4 which can collect data on speed, steering
wheel movements, acceleration, braking, 3-axis acceleration and GPS location from the
car embedded computer. LiSA also includes an eye tracking faceLAB 5.0 system, a
Microsoft Kinect (with a head tracking software [22]) camera and several other cameras
to monitor the driver. Finally, all the data are recorded by a computer installed on-board.
Thus, LiSA is able to collect a wide range of contextual information.
To complete the contextual information on the user profile, questionnaires were
filled before, during and after the experimentation to collect sociodemographic vari‐
ables, such as age, gender, driving experience, opinions toward safe driving, etc.
Moreover, in the context of our experiment, we collected the driver’s level of
perceived fatigue by using a scale ranging from ‘0’ not at all fatigued to ‘10’ very
much fatigued at four times during the experimental session. Each participant was
asked his or her level of fatigue just before starting driving (Time 1), after 15 min
Fig. 1. The LiSA instrumented vehicle
of driving once the experiment had started (Time 2), after 30 min (Time 3), and when
the driving experiment ended (around 45 min) (Time 4). The focus of the experi‐
ment was to measure the level of perceived fatigue, not the physiological level of
fatigue. A 45-minutes driving task on a monotony road low (e.g., 60 km/h) has been
found to induce fatigue in prior laboratory experiments [18] and a field experiment
pretest. The pretest shows a significant increase in the assessed fatigue level for the
majority of the participants, without reaching a hazardous level.
Finally, in the present study, the data analysis strategies adopted was the use of a
tool with data mining and statistical evaluation algorithms. For example, one algorithm
transforms the data on the most common measure by aggregating the data from sensors
with an higher sensing frequency by computing the average values [23]. We also use a
quasi-monotonic segmentation algorithm [24] to reduce the noise and compute the rele‐
vant acceleration and deceleration behaviours of the participants. We used it to allow
an easy visualization of the data collected. This tool was used first to test algorithms
offline and to stock the contextual information collected for each participant. We applied
this tool to all data collected from the field experiment. The tool enabled us to visualize
the collected data with graphs and maps. It then proposed a feedback solution to promote
awareness of the fatigue impact on driving (particularly speed) and bring the drivers to
better manage their fatigue and, ultimately, change their driving behaviours accordingly.
For designing the feedback, we selected the most suitable persuasive design princi‐
ples from Persuasive Systems Design (PSD) Process Model [6]. Following an analysis
of the persuasion context, in terms of intent, event and strategy, systems qualities were
applied to the design of the feedback (Fig. 2). We explain each of them in the next page.
The goal of this intervention was to provide participants with information on their
driving style relative to fatigue. The feedback consisted of three pages. The first one
showed illustrations, in the form of readily understandable pictograms and graphs. The
second page gave a written description of the driving experiment and definitions (in
order to facilitate understanding of terms that could be perceived too technical) and a
short summary for those who were less interested in too detailed explanations. Finally,
in the third page, different sections explained in more detail the first page’s illustrations,
including one section comparing the driver data to scientific research in the field. These
sections may be viewed as a kind of dialogue with the driver. We included praise, for
those who had identified well their fatigue, and suggestions, for those who perceived it
116 P. Ruer et al.
Driving
Analyzing the Design of system behaviour
persuasion qualities and/or
context attitude
change
• Intent: •Primariy task support:

•Persuade the elderly •Tailoring
driver about his/her •Personalization
driving and his/her •Self-monitoring
fatigue perceived
•Dialogue support:
• Event:
•Suggestion
•Driving context and
•System credibility:
driver information
•Trustworthiness
• Strategy:
•Real-world feel
•A Data Analysis Tool
•Social support:
•Social comparison
Fig. 2. Phases adapted from PSD process model
less well. For praise, words were used to provide personal feedback information based
on the user good driving behaviours. And for the suggestions, the feedback proposed
different comparison between the speed limit and driver’s speed, between perceived
fatigue and driver’s speed, etc.
A fictitious example of the first page with illustrations is presented in Fig. 3.
Fig. 3. Fictitious example of the feedback first page for one participant
In this example, at the top left, there are pictograms depicting participant gender,
driving time, kilometers travelled, mean speed and type of activity. Because of the
exploratory nature of the study we selected different pictograms in order to verify which
ones were evaluated as most useful and easily understandable. The content of the infor‐
mation was tailored according to the context in which the experimental driving was done
and in line with the interests of elder people. And each feedback is personalized
depending on the driving behaviour.
At the center of the example, the graph depicts the participant speed during the 45 min
experiment and the graph on the far right, the perceived level of fatigue of the participant
and his or her the reference age group at four moments during the driving period. These
curves allowed a better fit to the reality of the driver experience, for system credibility.
Because objectives information from real-world context is obtained through data coming
from sensors into the instrumented vehicle and questionnaires, the tool can be viewed
as trustworthy and without bias.
The table at the bottom of the page indicates mean speed and standard deviation
every 15 min, for the participant and for the reference age group allowing social compar‐
ison and some form of social support. Each elder person can compare his/her driving
performance with his/her age group’s reference, mainly for speed and fatigue. After the
design of our feedback, we tested it with elderly drivers.
4 First Results
Twenty older drivers participated in the experiment and received the feedback. Because
two of them were unreachable for the phone interview, the results presented here are for
the eighteen participants who completed the whole experiment. The mean age was 66,7
years old (SD = 6,43). All participants had a valid driver’s license and they declared a
satisfactory medical condition during the phone recruitment. To complete the self-
declared medical condition and before the driving, each participant did a screening test,
the Useful Field of View (UFOV) test. This test is designed for people aged 55 and older
and is recommended as a control measure for the ability to drive [25]. During the
experiment, the measures taken were perceived fatigue, speed, etc. Each participant
drove around 50 km on a close driving circuit with the LiSA car. At the end, we offered
each participant his/her driving feedback. As the tool was offline, feedback was given
either in electronic format or in paper format if the person had no access to internet. The
eighteen participants received a financial compensation of 30 $ and 10 $ CAN respec‐
tively for their participation in the driving experiment and phone interview.
After a few weeks, participants were contacted by phone to answer a phone inter‐
view. Nine questions dealt with the design of the feedback and ten with the partici‐
pants driving performance and the impact of the feedback received about the driving.
Initially, we wanted to know if the participants were satisfied with the information
received through the feedback, if the information met their expectations and if they
had taken the time to read all the information received. Secondly, it seemed impor‐
tant to check whether the feedback received corresponded to what they remembered
and had retained from the experience (e.g. speed, perceived fatigue, etc.). Thirdly,
we wanted to know if they found the feedback useful and if it had had an impact on
118 P. Ruer et al.
their subsequent driving and management of driving fatigue. The results are
presented under the following categories: (1) satisfaction, expectations, information
and memory and (2) usefulness of the information received and changes in behaviour.
About satisfaction, expectations, information and memory, the vast majority of
participants said they were satisfied or very satisfied with the information received
(88.2 %) and over half of the participants (58.8 %) indicated that the information met their
expectations. Three in four participants (75 %) found that the speed curve, shown in the
graph, corresponded to their memory, and almost all (88 %) found that the figure showing
fatigue perceived during the experiment represented what they had felt. Table 1 presents
the mean score and standard deviation for each categories.
Table 1. Mean score and Standard Deviation (SD) for satisfaction, expectations, information and
memory
Satisfaction Expectations Information Memories

Speed curves Fatigue curves
Mean 4,35/5 3,64/5 9,11/10 4,35/5 4,29/5
SD 0,86 0,93 0,92 0,86 0,84
Relative to the usefulness of the information received and behaviour changes, the
comparison with the reference age group was considered useful by all participants and
the general impression of feedback, content and graphic presentation was very positive
(Table 2). Despite the positive impression left by the feedback, only about one in two
(52.9 %) said that it had an impact on their driving afterwards and 70 % of participants
said that it would have an impact in the future. Three out of five (64,7 %) believed that
this feedback would foster better management of their fatigue in the future.
Table 2. Mean score and Standard Deviation (SD) for usefulness of the information received
and behaviour changes
Usefulness of the Behaviour change
information
Impact on the driving Impact on the driving Impact on their fatigue
afterwards in the future management
Mean 4,58/5 2,47/5 2,64/5 2,94/5
SD 0,50 1,41 1,32 1,29
About the feedback design, fourteen participants said they read the feedback (the
three pages) and found it very useful, relevant and understandable. Finally, as for the
other feedback elements, which one they most and least liked. The fatigue curve was
the most liked and the least liked was the pictograms.
5 Discussion and Conclusion
Overall, persuasive technology in automotive context promotes safety for drivers. With
the aging, elder people are more concerned about driving fatigue on the wheel. Persua‐
sive technologies appear as a solution to help them to improve their behaviour and drive
safely. We proposed in this paper the results of an exploratory study of a persuasive
technology designed for elderly drivers, in the form of feedback. The goal of the feed‐
back was to promote awareness of the fatigue impact on driving (particularly speed) and
bring the drivers to better manage their fatigue and, ultimately, driving behaviours. We
tested it with twenty elderly drivers after an experimentation of 45 min driving time.
Eighteen elder persons answered a phone interview about the satisfaction, expectations,
information and memories and about the usefulness of the information received and
changes in behaviour. The preliminary results are encouraging. The feedback as
conceived for elderly drivers was well received and was considered useful. 70 % of
participant said that the feedback would have an impact on their driving behaviour in
the future. The design of persuasive feedbacks for elderly drivers seems promising as
our results suggest that some elderly drivers appreciate receiving constructive informa‐
tion on their driving.
However, further works are needed. Firstly, we proposed the feedback between one
and two months after the driving session, and the interview took place several weeks
after the feedback was given. It would be important, in a future study to verify if feedback
given to elderly drivers on their driving has inducing real changes in their driving
behaviours and/or in their attitude toward driving fatigue. In addition, in our current
experiment, the number of participants was relatively small. It would be important in
future research to increase the number of elder participant substantially.
Secondly, the feedback design for elderly should be further improved. For example,
some of our participants suggested an improvement could be done by increasing the size
of curves in the illustrations. They didn’t see well due to vision problems. Others
proposed different colors than those we had used like a darker green color to accentuate
contrasts. So, the design has to be considered because it might play a role in persuading.
Furthermore, it would be interesting to expand and test feedback which would
include other real-world information, such as physiological data (blinking movement,
heartbeat, etc.) or the environmental context (weather, traffic, etc.). Some driving behav‐
iours due to fatigue are associated with environmental context, for instance, monotony
road or night condition. The objective would be to increase context awareness and induce
more behavioural changes among elderly drivers when fatigued at the wheel, using
efficient persuasive tools designed specifically for that category of drivers.
Acknowledgments. We want to particularly thank the Canadian Automobile Association (CAA)

Foundation, Section of the Quebec Province, for funding the research works behind this paper.
120 P. Ruer et al.
References
1. Sivak, M., Schoettle, B.: Recent changes in the age composition of drivers in 15 countries.
Traffic Inj. Prev. 13(2), 126–132 (2012)
2. Reimer, B.: Driver assistance systems and the transition to automated vehicles: A path to
increase older adult safety and mobility? Publ. Policy Aging Rep. 24(1), 27–31 (2014)
3. Emery, L., Srinivasan, G., Bezzina, D., Leblanc, D., Sayer, J., Bogard, S., Pomerleau, D.:
Status report on USDOT project – an intelligent vehicle initiative road departure crash
warning field operational test. In: 19th International Technical Conference Enhanced Safety
of Vehicles, Washington DC (2005)
Morgan Kaufmann Publishers Inc., San Francisco (2003)
5. Hamari, J., Koivisto, J., Pakkanen, T.: Do persuasive technologies persuade? - a review of
empirical studies. In: Spagnolli, A., Chittaro, L., Gamberini, L. (eds.) PERSUASIVE 2014.
6. Oinas-Kukkonen, H., Harjumaa, M.: Persuasive systems design: key issues, process model,
and system features. Commun. Assoc. Inf. Syst. 24(1), 28 (2009)
7. Schätzl, J.: How Effective are Persuasive Technologies in Automotive Context? Persuasive
Technologies and Applications (2015)
8. Meschtscherjakov, A., Wilfinger, D., Scherndl, T., Tscheligi, M.: Acceptance of future
persuasive in-car interfaces towards a more economic driving behaviour. In: Proceedings of
the 1st International Conference on Automotive User Interfaces and Interactive Vehicular
Applications, pp. 81–88. ACM (2009)
9. Vaezipour, A., Rakotonirainy, A., Haworth, N.: Reviewing in-vehicle systems to improve
fuel efficiency and road safety. Procedia Manuf. 3, 3192–3199 (2015)
10. Eby, D.W., Molnar, L.J.: Has the time come for an older driver vehicle? (2012)
11. Gruau, S., Pottier, A., Davenne, D., Denise, P.: Les facteurs d’accidents de la route par
somnolence chez les conducteurs âgés: Prévention par l’activité physique. Recherche-
Transports-Securite 79, 134–144 (2003)
12. Assailly, J.P., Bonin-Guillaume, S., Mohr, A., Parola, A., Grandjean, R., Frances, Y.M.: Les
conducteurs âgés en bonne santé font plus d’erreurs et d’oublis que d’infractions: Enquête
auprès de 904 volontaires. La Presse Médicale 35(6), 941–947 (2006)
13. Oxley, J., Langford, J., Koppel, S., Charlton, J.: Senior Driving Longer, Smarter, Safer:
Enhancement of an Innovative Educational and Training Package for the Safe Mobility of
Seniors. Technical Report, Monash University Accident Research Centre, Monash Injury
Research Institute (2013)
14. Anstey, K., Wood, J., Lord, S., Walker, J.G.: Cognitive, sensory and physical factors enabling
driving safety in older adults. Clin. Psychol. Rev. 25(1), 45–65 (2005)
15. Owsley, C., Mcgwin, G.: Vision and driving. Vision. Res. 50(23), 2348–2361 (2010)
16. Langford, J., Koppel, S.: Licence restrictions as an under-used strategy in managing older
driver safety. Accid. Anal. Prev. 43(1), 487–493 (2011)
17. Rakotonirainy, A., Steinhardt, D.: In-vehicle technology functional requirements for older
drivers. In: The 1st International Conference on Automotive User Interfaces and Interactive
Vehicular Applications, pp. 27–33. ACM, New York (2009)
18. Thiffault, P., Bergeron, J.: Monotony of road environment and driver fatigue: a simulator
study. Accid. Anal. Prev. 35(3), 381–391 (2003)
19. Beirness, D.J., Simpson, H.M., Desmond, K.: The Road Safety Monitor 2004: Drowsy
Driving. Traffic Injury Research Foundation, Ottawa, Ontario (2005)
20. Gershon, P., Shinar, D., Oron-Gilad, T., Parmet, Y., Ronen, A.: Usage and perceived
effectiveness of fatigue countermeasures for professional and nonprofessional drivers. Accid.
Anal. Prev. 43, 797–803 (2011)
21. Vanlaar, W., Simpson, H., Mayhew, D., Robertson, R.: Fatigued and drowsy driving: a survey
of attitudes, opinions and behaviours. J. Saf. Res. 39, 303–309 (2008)
22. Kedowide, C., Gouin-Vallerand, C., Vallieres, E.F.: Recognizing blind spot check activity
with car drivers based on decision tree classifiers. In: 28th AAAI Conference on Artificial
Intelligence (AAAI 2014), Québec (2014)
23. Ruer, P., Gouin-Vallerand, C., Zhang, L., Lemire, D., Vallières, E.F.: An analysis tool for
the contextual information from field experiments on driving fatigue. In: Christiansen, H.,
Stojanovic, I., Papadopoulos, G.A. (eds.) Modeling and Using Context. LNCS, vol. 9405, pp.
172–185. Springer, Heidelberg (2015)
24. Lemire, D., Brooks, M., Yan, Y.: An optimal linear time algorithm for quasi-monotonic
segmentation. Int. J. Comput. Math. 86(7), 1093–1104 (2009)
25. Clay, O.J., Wadley, V.G., Edwards, J.D., Roth, D.L., Roenker, D.L., Ball, K.K.: Cumulative
meta-analysis of the relationship between useful field of view and driving performance in
older adults: Current and future implications. Optom. Vis. Sci. 82(8), 724–731 (2005)
Creating Awareness of Sleep-Wake Hours by Gamification
Ezgi Ilhan1 ✉ , Bahar Sener2,3, and Hüseyin Hacihabiboğlu4

( )
1
Department of Industrial Design, Atilim University, Ankara, Turkey
[email protected]
2
Department of Industrial Design, Middle East Technical University, Ankara, Turkey
3
School of Engineering, University of Liverpool, Liverpool, UK
[email protected]
4
Department of Modeling and Simulation, Middle East Technical University, Ankara, Turkey
[email protected]
Abstract. Gamification can be used to motivate people to carry out hard-to-

perform tasks. It can help in changing undesirable habits and in improving a
person’s subjective well-being. Sleep-wake behaviors are important determinants
of day-to-day well-being. This study aims to find out whether it is possible to
modify sleep-wake habits using gamification. To this end, a gamified alarm clock
app, Sleepy Bird, was designed and tested in a user study with thirteen participants
using gamified and thirteen participants using non-gamified versions for two
weeks. The results indicate that the participants of the gamified version were more
motivated to start the day at required times than the participants of the non-gami‐
fied version. The participants of the gamified version were also observed to have
made desirable modifications to their sleep-wake habits.
Keywords: Gamification · Sleep-wake habits · Habit modification · Subjective

well-being
1 Introduction
As one of the critical dimensions of subjective well-being, management of sleep-wake

habits of people is an area that can benefit from gamification. Waking-up and sleeping
at required hours can be considered as difficult-to-perform tasks (at least for some). This
study aims to find out whether it is possible to modify sleep-wake habits and improve
people’s subjective well-being using gamification. A gamified alarm clock app, Sleepy
Bird, was designed and tested in a user study to investigate whether the app can create
positive feelings and better awareness of sleep-wake habits.
1.1 Gamification
The popularity of gamification derives from the popularity of playing games. The Global
Games Market Report [17] indicates that globally, 1.2 billion people had played games
by 2014. Playing has the power to trigger new behaviors against previous habits [3].
Gamification is a useful strategy that can influence people positively in the activities
that they ignore, find boring, tiresome or even difficult to achieve. Gamification is based

DOI: 10.1007/978-3-319-31510-2_11
Creating Awareness of Sleep-Wake Hours by Gamification 123
on the usage of game design elements in non-game contexts [6]. It has the potential to
persuade people to take actions about undesirable but necessary activities [2] making
them more appealing to people by creating extrinsic and intrinsic motivations [21]. An
extrinsic motivation [16] satisfies needs with tangible outcomes such as reputation, titles,
money and real rewards, whereas an intrinsic motivation symbolizes the satisfaction
from an interesting and enjoyable activity. The success of a gamified tool comes from
the satisfaction of intrinsic motivation and self-defined goals of players [21]. In this
study, gamification is used as a motivation tool to regulate sleep-wake habits of people
in order to improve their well-being.
1.2 Sleep-Wake Habits

Sleep-wake habits play an important role in happiness and long-term well-being. The
more efficiently the subject sleeps, the higher the emotional, psychological and social
well-being becomes. Thus, if sleep problems are reduced, both emotional and psycho‐
logical well-being will improve [8, 9].
Sleep duration is also related to psychological health. Although it may differ for
different age groups, nominal adequate sleep duration is 6 to 8.5 h. Optimal sleepers are
known to have lower rates of incidence for depression due to possessing higher amounts
of mastery, self-confidence and social relations, whereas insufficient sleep causes
depression, stress and both physical and cognitive fatigue during the following day [15]
and is known to deteriorate the learning performance of adolescents [19].
There are examples of mobile apps or alarm clocks that use gamification to track
people’s sleep-wake habits, and to ease waking-up times and prevent snoozing. For
instance, ‘Gun O’Clock’ aims to wake people up with a game which the user has to shoot
the clock with a toy gun; or the users of ‘Helicopter Alarm Clock’ need to catch a flying
helicopter and put it on its nest to stop the alarm [4]. ‘Zeo Sleep Manager’ is another
example that tracks sleep cycles using wearable devices [14]. It has sensors on its head‐
band to measure the electrical current during sleep phases. Then, it wirelessly sends the
information to the mobile platform to make users aware of their sleep-wake habits and
health conditions.
Some of the apps come with gamified features as well. People need to follow instruc‐
tions offered by the mobile apps in order to go-to-bed, wake-up on time or to have enough
sleep. ‘Early Bird’, developed for Starbucks, tries to persuade people to start their day
on time with feedback messages and a reward system on a social platform [20]. When
people use this alarm app without snoozing, it gives achievements to users. The achieve‐
ment system is implemented with points, stars, social sharing and a cup of coffee, which
is available for an hour after getting out of the bed.
Existing hardware/software solutions either compel people to wake-up at adjusted
hours or use wearable electronics rather than using the power of persuasive design.
Without using persuasive design and a ‘fun factor’ effectively, gamified apps remain
inadequate in influencing sleep-wake habits in a profoundly positive way. Tracking sleep
through an app can help people to become aware of their sleep-wake habits. This can
help to improve their sleep-wake behaviors positively in the long run [11].
124 E. Ilhan et al.
2 Methodology
With the evidence from literature that sleeping and waking-up are important everyday
behaviors and that gamification could potentially increase subjective well-being, the
present study explored the effects of gamification on the attainment of the following
goals: (1) increasing the motivation of participants for waking-up, (2) creating better
awareness of sleep-wake habits, and (3) creating positive attitudes towards sleeping and
waking up at optimal times.
The study required an understanding of people’s existing sleep durations and
waking-up times. A mobile alarm clock app, which allows remote data collection, was
considered to be an appropriate tool to track sleep-wake habits of the participants.
A specially designed gamified alarm clock app–Sleepy Bird– was used to track and
collect sleep-related data in order to determine the effects of gamification.
2.1 Development of the Mobile App
Sleepy Bird was designed specifically for the present study, by modifying an existing
popular game with alarm features. It was considered that in platform games, the quan‐
titative progression of players can be observed easily. Flappy Bird [7] is a popular game
which was downloaded up to 3 million times per day. Owing to this popularity, Flappy
Bird was selected as a starting point for a gamified alarm clock app.
For the technical development of the app, Java language and libgdx graphics
library were used. The entire Flappy Bird game was redesigned and redeveloped for the
Android platform with embedded alarm functions, feedback, and calculations.
Both a gamified (G) and a non-gamified (N-G) version of Sleepy Bird were used in
the study. The differences between the versions can be seen in Table 1. Since gratifying
feedback was suggested to motivate participants intrinsically, both versions of the app
offered personal feedback messages and insights about wake-up times, snooze actions,
and sleep durations.
Table 1. Differences between G and N-G versions of Sleepy Bird app
Gamified Version (Experiment Group) Non-Gamified Version (Control Group)

Alarm time and sound Alarm time and sound
Going to bed hour and sleep duration Going to bed hour and sleep duration
Feedback about wake/sleep actions Feedback about wake/sleep actions
Revised Flappy Bird game N/A
Game elements (e.g. visual elements, N/A
lives, points, feedbacks, leader board)
As Sleepy Bird aimed to create positive feelings while starting the day, color choices
and graphics design were carefully made. The Sleepy Bird character was designed to
have a big red eye to emphasize its level of drowsiness. The game also used light to set
an ambience, informing participants that the sun had risen that morning.
The app was designed with a Turkish language interface to fit the target user group
of the study. At the top of the game screen, the score and remaining lives were displayed
(see Fig. 1). For the gamified version, Sleepy Bird was designed to have four buttons on
the main screen, which were grouped into two with color coding. OYNA (play) and (EN
İYİLER) (leader board) buttons were related to gaming; ALARM KUR (set alarm) and
UYKUYA DAL (sleep) buttons were related to sleeping-waking up. For the non-gami‐
fied version ‘play’ and ‘leader board’ were removed.
Fig. 1. Example screenshots from ‘Sleepy Bird’
The game provided personal feedback messages when, for example, the participants
set the alarm or wanted to sleep. The messages aimed to create awareness about sleep-
wake habits and were shown in red color to draw attention. If the feedback message was
relatively long, a pop-up screen opened just as Sleepy Bird communicated with the users
about their action.
The Sleepy Bird character appeared on the main screen to remind users to set their
alarms for the following day and not to sleep late in order to wake-up easily next
morning. Depending on the wake-up time set by the participants, the app showed them
the score award for the next day. In order to record the sleep duration, the users needed
to tap on the sleep button. After the alarm went off and awakened them in the morning,
they played the game whenever they wanted during the day. The interface enabled the
participants to check the leader board as a way to self-assess their standing among other
players. The non-gamified version only required setting the alarm and sleep hours.
Decisions Behind the Design of ‘Sleepy Bird’. In the development process of the app
versions, the following considerations were taken into account:
126 E. Ilhan et al.
• ideal wake-up time was set to be between 6.00-8.00 am and at least 1.5-2 h before
the work time [5].
• adequate sleep duration was set as 7 h for the target group [22].
• game elements such as points, lives and leader boards were used to motivate partic‐
ipants [13].
• continuous feedback messages about personally required sleep-wake hours were
provided [16].
The game rules were set as:
• The game gave ‘10 lives’ by default.
• When participants woke-up at their optimum hours they won the maximum number
of lives. Reward with ‘30 lives’ was found as an appropriate solution to alleviate the
tradeoff between enhancing the fun and keeping the value in the gameplay. The worst
condition of wake-hour rewarded the user with ‘3 lives’.
• Work/school time was important for the background calculations of the app. It gave
maximum game lives when there was a 1-2 h time gap between wake-up and work
start hours. If participants did not wake-up at the right time they lost ‘1 life’ for every
5 min of delay.
• Snoozing the alarm negatively affected participants in the competition. Each time the
participants snoozed the alarm, they fell 25 m back in the game.
• If participants went to bed too early or too late, the speed of Sleepy Bird (character)
increased to make the game more difficult.
2.2 Data Collection Tools
Data were collected via pre-use questionnaire (delivered prior to the use of the app),
post-use questionnaire (delivered on the completion of the predetermined usage period),
and usage logs. The pre-use questionnaire aimed to select and group the participants
according to their scores. The post-use questionnaire aimed to collect usage-related
information. The questionnaires also helped to compare relevant data between G and
N-G versions of the app. Sleep-wake habits, moods and awareness levels were collected.
Four scales were implemented in the questionnaires to collect personal evaluations.
These were Epworth Sleepiness Scale (ESS), Pittsburgh Sleep Quality Index (PSQI),
Morningness-Eveningness Questionnaire (MEQ), and Delighted-Terrible Scale.
2.3 Selection and Distribution of the Participants
The call for recruiting participants was made via word of mouth, e-mails and social
media channels. Eligible participants were required to fit the following criteria:
• using of an alarm to wake-up.
• being within the age group of 18–35 (with similar sleep-wake habits).
• possessing average awaking difficulty.
• having a mobile phone running on Android platform.
• having no particular aversion to the Flappy Bird game.
A pre-use questionnaire was sent to qualifying volunteers. Candidates without sleep-

related problems or with serious medical problems were excluded from the study.
Selected participants were divided into two groups to use either G or N-G versions of
the apps. Since the game in the app was a modified form of Flappy Bird, participants’
approaches to the game played a significant role in deciding the group divisions. If
participants had positive attitudes towards Flappy Bird, they were assigned to the G
version of the app. If they were not interested in Flappy Bird, they were given the N-G
version. An equal number of participants were allocated to each group. In total, 26
participants took part in the study (13 G; and 13 N-G users). Prior to finalizing the details
of the main study, as a piloting, different user groups utilized the app in different time
periods during the development stages. This helped to detect and correct possible func‐
tional errors, to improve the app’s performance, choose appropriate statistical analysis
methods, and decide on the number of participants that would be recruited for the study.
3 Results and Analysis
The results of the pre-use questionnaire, computer logs and post-use questionnaire were
analyzed as described in the following sections.
3.1 Pre-use Questionnaire
The pre-use questionnaire is organized in three parts to assess the suitability of the
candidates for the study and then to collect relevant information about them.
( i) Part 1. In Part 1, the Epworth Sleepiness Scale (ESS) was used to gather information
on the candidates’ subjective evaluations of sleep levels [12]. A total score was evaluated
as participants having ‘average sleep problems’ (scores between 4 to 9); ‘sleep problems’
(10–14); and ‘requiring professional medical help’ (15 +). If the score was medium level
(4–14), the participants were considered to have the right level of sleep problem to take
part in the study. In the ESS results, the minimum total score was 4, and the maximum
score was 13, with a mean score of 7.76 within the selected group of participants. Of
the volunteers, six received ESS scores lower than 4, and were not included in the tests
as they presented no sleep problems.
(ii) Part 2. In this part, the participants’ sleep, work and snooze behaviors were eval‐
uated. Some questions included in this part were adapted from the Pittsburgh Sleep
Quality Index (PSQI) to obtain information about participants’ subjective sleep-wake
evaluations [18]. This part of the study aimed to assess pre-use sleep patterns of the
participants. They were also asked the time of the day that they start working, to correctly
calculate their appropriate wake-up hours. It was observed that 92 % of the participants
started their school or work between 8.00–10.00 am.
(iii) Part 3. This part of the questionnaire focused on the participants’ personal assess‐
ments about their wake-up attitudes and motivations. The questions regarding sleep-wake
habits were obtained using the Morningness-Eveningness Questionnaire (MEQ) [10]. The
128 E. Ilhan et al.
26 participants assigned scores ranging from 1 to 5 (1 being ‘strongly disagree’ to 5 being

‘strongly agree’) to each of the following six questions. All scores given in brackets are out
of 130 (the sum of G and N-G = 26 × 5).
– Q1 asked whether participants believed it was easy to wake-up in the morning, and
the answers showed the lowest total score (54/130).
– Answers to Q2 proposed that G app would help to wake-up more easily (*81/130).
– Q3 (63/130) and Q4 (74/130) focused on the degree of being wakeful or tired after
waking-up.
– Q5 received the highest total score (99/130) indicating that the participants had posi‐
tive feelings when they started the day at a correct time.
– Q6 received a high score (87/130) implying that G app would motivate participants
to start the day at a correct time.
The results of Part 3 of the pre-use questionnaire are later compared with counterparts
from the post-use questionnaire, to examine whether a positive change in participants’
attitudes could be established.
3.2 Computer Logs

Computer logs were used to collect each of the 26 participant’s sleep-wake actions
quantitatively. After the installation of Sleepy Bird on their phones, the usage data were
automatically gathered in a database, and all the actions taken in the app were remotely
recorded in the system. The logs were related to wake-up time, sleep time, sleep duration
and number of snooze actions.
The participants were required to set the alarm clock every day for the next morning
before going to bed. When the alarm setting was forgotten, participants’ data were not
useful for the evaluations of that day. Therefore, at 10.00 pm every evening, reminders
about alarm-setting were displayed by both versions of the app.
At the end of the study, a total of 325 samples (i.e. number of total actions related
to wake-up hours, sleep duration, and number of snoozes) were gathered from the
participants. Data comparisons were made based on the app version (G or N-G), and
each participant’s habits relative to the first- and second-half usage periods and the usage
time (weekend or weekdays). Differences between the following were determined:
• number of snoozes,
• wake-up time (deviation from personalized optimum time),
• going to sleep time (deviation from personalized optimum time),
• sleep durations (deviation from personalized optimum duration).
The usage frequency between G and N-G versions showed a slight difference: G alarm
provided a total usage data of 170 (52.3 %) samples, whereas N-G alarm provided a total
of 155 (47.7 %) data.
Analysis Strategy for the Logs. In order to analyze the computer logs, data were
analyzed in SPSS to find out: the mean, range, minimum and maximum values of wake
and sleep time, snooze actions, sleep durations, and difference from optimum cases.
In order to establish the presence of any statistically significant values, a full factorial
ANOVA model was obtained. The model included (i) game version, (ii) first and second
half of usage; and, (iii) weekend vs. weekday usage as variables. All the tests were
designed with 95 % confidence interval to have a margin of error less than 5 %, to reach
a high level of reliability.
(i) Game version. An important goal for the study was to investigate the possible effects
of gamification. Table 2 summarizes a descriptive statistics of snooze numbers, differ‐
ence from optimum wake/sleep time and difference from sleep durations for G and N-
G versions. The mean of snooze numbers (0.74) and difference from optimum sleep
duration (which is 0) of the G version (24.33 min) was lower than mean values for the
N-G version (46.33 min). Moreover, going-to-sleep time was closer to optimum (which
is 0) for the G version, with a mean deviation of -11.75 min, whereas it was -36.85 min
for the N-G version.
Table 2. Group statistics of actions by app version (G/N-G) for 26 participants

Version* Total no. of Samples Mean Std. Deviation Std. Error
Mean
Snooze G 170 0,74 1,373 0,105
N-G 154 1,19 1,604 0,129
Difference_optimum_wake G 170 13,18 56,324 4,320
N-G 154 9,77 59,607 4,803
Difference_optimum_sleep G 169 -11,75 84,303 6,485
N-G 150 -36,85 76,252 6,226
Difference_optimum_duration G 169 24,33 88,391 6,799
N-G 149 46,33 84,928 6,958
*G: gamified; N-G: non-gamified.
In order to see whether there was a statistically significant difference of sleep-wake

habits between the G (experiment) and N-G (control) users, one-way ANOVA was used.
The difference was statistically significant for three actions according to post-hoc
comparisons. As can be observed from Table 2 (text in bold), Sleepy Bird resulted in
significant differences between the G and N-G versions in terms of snooze action; go-
to-sleep hours; and sleep durations.
Both subjective questionnaire evaluations and computer log results showed that
wake-up hours changed in a positive way for participants of the G version. Participants
of N-G also changed their wake-up hours with the help of feedback from the app.
Participants of G version (7.24 h) became closer to optimum sleep durations (7 h) than
participants of N-G version (7.46 h) at the end of the study. Going-to-sleep time changed
in a personally-positive way for participants of both G (from 01.13 am to 00.19 am) and
N-G versions (from 00.32 am to 23.53 pm).
(ii) First and Second Half of Usage Period. Since the participants used the apps for
different numbers of days, the median value was different for each participant. The app
usage ranged from 8 to 22 days between the participants. Average values of the date set
130 E. Ilhan et al.
(i.e. wake-up time, sleep time, sleep durations, snooze numbers) of the first half of the
usage period was compared to that of the second half. In the second half, mean values
for the tested dependent variables approached zero. Thus, all of the actions had a positive
progression in the duration of the study.
For G users, all the mean values became closer to zero in the second half, which
indicated the optimal case. At the beginning of tests, participants snoozed the alarms
more (0.99 times in average), woke–up later (17 min later than their personalized
optimum time), slept earlier (15.49 min earlier than their personalized optimum time),
and slept more (32.86 min more than their personalized optimum durations). According
to post-hoc comparisons, gamification resulted in statistically significant differences in
terms of snooze action (p = 0.008) between first half usage of gamified and second half
usage of gamified versions.
(iii) Weekend vs. Weekdays. Participants used Sleepy Bird most frequently on Tues‐
days and Wednesdays (18 %) and least frequently on Saturdays (9.5 %). The alarm was
set 70 times (21.8 %) during the weekends compared to 250 times (78.2 %) in school/
work days. For G users, 133 action data events on weekdays and 37 action data events
at weekends were collected. According to post-hoc comparison results, differences in
the snooze action while using the gamified version of Sleepy Birds was statistically
significant (p = 0.036) between weekdays and weekend usage. For N-G users, 120 data
events in weekdays and 34 data events at weekends were gathered. According to post-
hoc comparisons, the N-G version of Sleepy Bird caused statistically significant results
in terms of difference from optimum going-to-sleep time (p = 0.035) between weekdays
and weekend.
3.3 Post-use Questionnaire
The post-use questionnaire enabled data comparisons between G and N-G versions of
app users in response to whether a gamified alarm app created positive feelings and
awareness for sleep-wake habits.
(i) Part 1. This part revealed the emotional experience related to the usage of the app.
Delighted-Terrible Scale [1] (one question to select one of the facial expressions) was
used to measure short term subjective well-being after the usage of Sleepy Bird. One
participant (4 %) felt very delighted due to Sleepy Bird. Six participants (23 %) felt
delighted during usage and 15 participants (58 %) pointed out that they felt happy due
to this special alarm design. Four participants (15 %) felt neutral. The average feeling
for the gamified version was between a delighted and happy, whereas the average feeling
for the non-gamified version was between a happy and neutral.
(ii) Part 2. This part was used to measure changes in participants’ awareness levels
regarding their sleep-wake habits after the usage of the app. It consisted of four questions
focusing on: wake-up hours (Q1), sleep durations (Q2), the interval between waking-up
and starting to study/work (Q3), and number of snoozes (Q4). G users scored a total of
177 points, whereas N-G users scored 124 points, indicating that awareness levels of
sleep-wake habits reached a higher level for the participants of G version. Q2: in relation
to participants’ awareness on sleep duration, both G and N-G users scored a total of 80
points, which was the maximum score received within this part of the questionnaire. Q3
& Q4: total scores for both questions were identical for G and N-G users (75). This
indicated that the Sleepy Bird app helped to increase participants’ awareness for the time
interval between wake and work time, and snooze action. Q1: total score for this question
had the minimum value (71), which was about awareness level to start the day at correct
hours.
When results were analyzed in SPSS using a t-test, awareness levels on sleep-wake
habits between G and N-G showed some statistically significant differences for all of
the questions (Q1: p = 0.010; Q2: p = 0.042; Q3: P = 0.004; Q4: p = 0.011). Participants
of the G version gained a significant awareness as a result of app usage.
(iii) Part 3. This part was used to analyze participants’ self-evaluations of their wake-
up attitudes and motivations after using Sleepy Bird. In contrast with the same part (Part
3) of the pre-use questionnaire results, the scores were higher in total as indicated in
Table 3 (text in bold). G users evaluated themselves as relatively more successful, more
motivated and feeling more positive in terms of waking-up at early hours. These partic‐
ipants gave the highest score to Q1 (54 points), indicating how much they thought that
they were waking-up at the correct time in the morning. G group also scored higher than
N-G for Q5 (52) and Q6 (50) in the post-use questionnaire indicating that the app made
them feel psychologically better and more motivated to wake-up at required hours. Q2
(40) showed that G group could not wake-up so easily during app usage. Score of Q3
and Q4 (36) showed that G group did not feel that awake, well-rested and lively after
getting out of the bed.
Table 3. Total scores received for six questions in pre- and post- questioannires
Pre-Questionnaire Post-Questionnaire
No. G N-G G N-G
Q1 27 27 54 49
Q2 45 36 40 41
Q3 31 32 36 34
Q4 37 37 36 33
Q5 53 46 52 43
Q6 49 38 50 44
Total Score 242 216 268 244
Although the total scores for G and N-G groups increased in the post-use question‐
naire, only one question was calculated at the statistically significant level. According
to separate t-test analysis, Q1 showed a significant change in G (p = 0.00) and N-G
(p = 0.001) groups. Moreover, the total score of questions for both groups (p = 0.012)
showed statistically significant changes between pre-use and post-use of the app. This
result showed that the app provided motivation and positive attitudes on sleep-wake
habits, especially for (Q1) waking-up at correct hours.
132 E. Ilhan et al.
(iv) Part 4. This part was used to understand participants’ tendency to change their
sleep-wake habits as well as their suggestions for new app features. Open-ended ques‐
tions were used to collect participants’ comments on positive and negative aspects of
Sleepy Bird. In total, 17 participants (65 %) stated that a gamified app would change
sleep-wake behaviors in the long term. The ‘leaderboard’ was interpreted as the most
attractive part for competition. Example comments from the participants of the gamified
version were:
– “I decreased my snooze habit for the game success. I confess that I dismissed the
alarm of Sleepy Bird once and set the alarm of my phone again in order not to lose
game lives in the app.” (Participant 8)
– “During the usage period, the app said to me that I will be either fresh or fatigued the
following day. When it said that I will be fresh, I was really fresh. Conversely after
the notification of tiredness, I felt tired the next day.” (Participant 13)
4 Discussion and Conclusions
This paper investigated the practical outcomes of gamification informed by domain

knowledge in sleep medicine. It provided a broader attention to the relationship between
gamification, and sleep-wake habits. The aim was to examine whether gamification can
create motivation and awareness to improve sleep-wake habits, so as to positively influ‐
ence subjective well-being in the long term.
The study utilized gamification with a shift from an extrinsic reward system to an
intrinsic feedback system. Sleepy Bird app was designed to focus on ‘intrinsic motiva‐
tions’ with its gratifying feedback system. With its feedback and notifications, the app
gave information on optimum wake and sleep hours to improve users’ awareness of
sleep-wake habits. The results revealed that gamification can be useful for the creation
of positive feelings and better awareness about sleep-wake habits. Gamification also
helped the study participants to start their day at required hours, since the app notified
them about their required sleep-wake hours and sleep durations. A gamified alarm app
also encouraged participants to not snooze their alarms.
The results provide insights for game developers. Computer logs allowed collection
of game-related data concerning the numbers of game lives, game points and top scores.
Time, sequence and frequency of actions taken by ‘players’ can be tracked through such
an app design. The feedback messages about competition and leaderboard results were
useful ways to concentrate users’ attention to their sleep-wake habits indirectly so that
they may become better in the game.
References
1. Andrews, F.M., Withey, S.B.: Social Indicators of Well-being: Americans’ Perceptions of

Life Quality. Plenum Press, New York (1976)
2. Baranowski, T., Buday, R., Thompson, D.I., Baranowski, J.: Playing for real: Video games
and stories for health-related behavior change. Am. J. Prev. Med. 34(1), 74–82 (2008)
3. Chan, K.: Playing in traffic: Pervasive gaming for commuters. In: Paper presented at the
Interactive Entertainment, New Zealand, 22 November 2011
4. Chou, Y.: How gamification can help you get out of bed by “killing” time. Retrieved from
http://www.yukaichou.com/product-gamification/gamification-bed-killing-time/
#.VCgqz_l_sUg (2012). Accessed 10 October 2014
5. Ciftci, B.: Personal communication (Associate Professor at Department of Thoracic Medicine
in Sanatoryum Hospital, Ankara, Turkey), 24 February 2014
6. Deterding, S., Sicart, M., Nacke, L., O’Hara, K., Dixon, D.: Gamification: Using game design
elements in non-gaming contexts. In: Proceedings of the CHI 2011, Vancouver, Canada, pp.
2425–2428. ACM (2011)
7. Gayomali, C.: A quick guide to flappy bird, the super-addicting, incredibly popular game.
Retrieved from http://www.fastcompany.com/3025959/fast-feed/a-quick-guide-to-flappy-
bird-the-super-addicting-incredibly-popular-game (2014). Accessed 14 September 2014
8. Gray, E.K., Watson, D.: General and specific traits of personality and their relation to sleep
and academic performance. J. Pers. 70(2), 177–206 (2002)
9. Hamilton, N.A., Gallagher, M.W., Preacher, K.J., Stevens, N., Nelson, C.A., Karlson, C.,
McCurdy, D.: Insomnia and well-being. J. Consult. Clin. Psychol. 75(6), 939–946 (2007)
10. Horne, J.A., Ostberg, O.: A self-assessment questionnaire to determine morningness-
eveningness in human circadian rhythms. Int. J. Chronobiol. 4(2), 97–110 (1976)
11. Hurd, R.: 8 Best sleep tracking apps and devices now that Zeo is gone. Retrieved from http://
dreamstudies.org/2013/07/16/best-sleep-tracking-apps-devices/ (2013). Accessed 7 July
2014
12. Johns, M.W.: A new method for measuring daytime sleepiness: The Epworth Sleepiness
Scale. Sleep 14(6), 540–545 (1991)
13. Montola, M., Nummenmaa, T., Lucero, A., Boberg, M., Korhonen, H.: Applying game
achievement systems to enhance user experience in a photo sharing service. In: MindTrek
2009 Proceedings of the 13th International MindTrek Conference: Everyday Life in the
Ubiquitous Era, pp. 94–97. ACM, New York (2009)
14. Orlin, J.: Sleep tracking startup Zeo says goodnight. Retrieved from http://techcrunch.com/
2013/05/22/sleep-tracking-startup-zeo-says-goodnight/ (2013). Accessed 7 June 2014
15. Roberts, R.E., Roberts, C.R., Chen, I.G.: Functioning of adolescents with symptoms of
disturbed sleep. J. Youth Adolesc. 30(1), 1–18 (2001)
16. Ryan, R.M., Deci, E.L.: Intrinsic and extrinsic motivations: classic definitions and new
directions. Contemp. Educ. Psychol. 25(1), 54–67 (2000)
17. Schutte, W.: Global games market report infographics. Retrieved from http://www.
newzoo.com/infographics/global-games-market-report-infographics/ (2013). Accessed 15
March 2015
18. Smyth MSN, C.: The pittsburgh sleep quality index (PSQI). N.Y. Univ. Coll. Nurs. 29(6), 1–
2 (2012)
19. Wolfson, A.R., Carskadon, M.A.: Sleep schedules and daytime functioning in adolescents.
Child Dev. 69(4), 875–887 (1998)
20. Young, K.: Starbucks early bird offers discounted coffee for the dedicated. Retrieved from
http://www.trendhunter.com/trends/starbucks-early-bird (2012). Accessed 12 September
2014
21. Zichermann, G.: Intrinsic and extrinsic motivation in gamification. In: Paper presented at the
Gamification Summit, San Francisco, 27 October 2011
22. Zohar, D., Tzischinsky, O., Epstein, R., Lavie, P.: The effects of sleep loss on medical
residents’ emotional reactions to work events: A cognitive-energy model. Sleep 28(1),
47–54 (2005)
Methods and Models
Cicero: Middleware for Developing Persuasive
Mobile Applications
Antonello D’Aloia1, Matteo Lelli1(&), Duckki Lee2, Sumi Helal3,

and Paolo Bellavista1
1
Department of Computer Science and Engineering,
University of Bologna, Bologna, Italy
{antonello.daloia,matteo.lelli4}@studio.unibo.it,
[email protected]
2
Creative Innovation Center, Advanced Convergence R&D Lab,
LG Electronics, Seoul, South Korea
[email protected]
3
Mobile and Pervasive Computing Lab, University of Florida, Gainesville, USA
[email protected]
Abstract. We present Cicero – a middleware solution to support developers

design and implement persuasive mobile apps. Based on the Action-Behavior
Model (ABM), Cicero provides developers with powerful class libraries and
collaboration methodology to streamline the development of mobile persuasive
apps without requiring a steep knowledge of behavior science theory or ven-
turing into domain-specific knowledge and artifacts. Cicero guides the devel-
opers in following the ABM steps, provides APIs for cyber sense and cyber
influence, and embodies the necessary model computations including measuring
end-user compliance and response to influence and persuasion. Cicero also
facilitates the engagement of domain experts in a clearly defined collaborative
role. Here we also originally detail the design and implementation of an Android
version of the Cicero middleware and we present a use case to practically
exemplify how Cicero can facilitate the application developers’ work.
Keywords: Persuasive computing Middleware Mobile app development

Android Sensing Actuation Situation Activity recognition
1 Introduction
Mobile smart technologies and products have become extremely pervasive and are an
essential part of our daily lives. Their almost ubiquitous presence is influencing us and
has begun to change our way of life and behaviors, either intentionally or uninten-
tionally. Among the many digital gadgets and products, smartphones and smart
watches are the most popular personal devices that we use today. They are getting
much attention lately as potentially powerful persuasive tools that could change human
behavior due to their accessibility, sensibility, connectivity, and triggerability.
However, recent research work in the field has shown that it is not easy to create
persuasive mobile apps without adequate knowledge of behavior and persuasion theory.

DOI: 10.1007/978-3-319-31510-2_12
138 A. D’Aloia et al.
In addition, assessing user behavior response to persuasive influencing factors is another

non-trivial challenge. App developers usually lack such knowledge and, therefore, to
bridge this gap, we propose Cicero – a middleware and associated toolkit to support
developers in designing and programming persuasive mobile apps. Based on the
Action-based Behavior Model (ABM) [4, 5] and situation-based user compliance
assessment algorithms [1–3], Cicero is designed to provide developers with powerful
class libraries and a programming/collaboration methodology to streamline the devel-
opment process of persuasive apps without requiring a steep knowledge of behavior and
persuasion theory or venturing into domain-specific knowledge and artifacts. Cicero
guides the developers in following the ABM model, provides API for cyber sense and
cyber influence, and embodies the necessary model computations including the effective
measurement of end-user response to influence and persuasion.
The remainder of the paper is structured as follows. In Sect. 2, we briefly de-scribe
the ABM model and discuss related work. Section 3 describes how Cicero is designed
as a mobile efficient specialization of the general ABM. Cicero Android implemen-
tation follows in Sect. 4, together with a use case in Sect. 5. After reporting experi-
mental performance results of our Cicero middle-ware, conclusive remarks and
directions of future work end the paper.
2 Background
Existing behavior theories and models are not adequate to mobile intervention [6]. In
order to fully utilize mobile technologies, behavior models are required to have
dynamic, regulatory system components to guide rapid intervention adaptation based
on the individual’s current and past behavior and situational context [6]. ABM is
designed to meet these requirements by providing adaptive iterative process based on
domain-specific behavior assessment. ABM provides specific step for a cyber system to
follow to persuade a user to change behavior. ABM also gives technological guidelines
on how a cyber system should act to provide persuasive influence by providing specific
cyber actions, which could be easily converted to APIs that developers understand and
utilize to create apps. We briefly describe ABM here. Further explanation on the
specific choices and inner workings of the ABM can be found in [4, 5].
ABM [4, 5] is a persuasion
template designed by incorporating
relevant persuasion theories into a
“cyber model” that app developers
can understand and utilize in a
broad range of applications that
require persuasion. As shown in
Fig. 1, the model consists of a cyber
system (cyber sense and cyber
influence) and a set of user actions.
Each step contributes and relates
to one or more of the three elements
of persuasion: motivation, ability Fig. 1. Action behavior model
Cicero: Middleware for Developing Persuasive Mobile Applications 139
and trigger [7]. The cyber influence affects and helps the user to take each action within
ethical guidelines [8]. As the user works toward her goals, the cyber sense monitors,
analyzes and observes her actions. After the initial rounds of steps, the Assess action
evaluates the achievements of the goals and either rewards or rolls the user back
through reinforcement loops to the appropriate model step based on the measured
deficits of each action. For positive achievements, the Reward action may give
intrinsic, extrinsic, or virtual rewards to reinforce the motivation.
A specific definition of situation that involves contexts, activities, and device
interactions, which was introduced in [3], is utilized by cyber sense. Based on this
definition, Situation-based Assess Tree (SAT) [1, 2] is defined as a formalism, struc-
ture, and algorithm for domain-specific behavior assessment under the ABM model.
The Cicero middleware implements a strictly mobile specialization of ABM where
cyber sense and influencing are limited to Android smart phones/watches. Cicero also
implements the Assess step through an optimized, cloud-based implementation of the
SAT algorithm as detailed in the following.
2.1 Related Work

Despite the increasing number of applications with persuasion features, only few
middleware and frameworks nowadays exist to support and streamline persuasive app
development. Moreover, some middleware solutions are specific for mobile crowd-
sensing lacking underlying persuasion theory. In this section, we analyze related work
on persuasion middleware, paying specific attention to their cyber sense influence
capabilities, and to their differences and similarities with our work.
Code In The Air (CITA) [10, 11] is a system developed in the Networks and Mobile
Systems group at MIT that simplifies the development of complex tasking applications
via a Web interface. CITA enables non-expert end users to easily express simple tasks on
their smartphones, and more sophisticated developers to code complex tasks by writing
purely server-side scripts in JavaScript. It is worth mentioning that CITA can sense and
recognize a number of low-level activities such as isWalking, isDriving, enterPlace, and
leavePlace. In addition, CITA allows developers and users to compose lower-level
activities using logical predicates to create high-level activities. Regarding the ability to
persuade the user, CITA provides limited support in the form of triggers and reminders. It
uses an asynchronous message delivery service to trigger users’ task fulfillment.
Funf [12] initially developed at the MIT Media Lab is another extensible sensing
and data processing framework for Android devices. It aims to help developers create
applications that need mobile sensing easily without having to access low-level APIs,
but with no support of cyber influence and rich persuasion features. There are also
some commercial applications similar to CITA and Funf, most notably Tasker1, which
allows users to perform tasks based on contexts such as time, date, location, event,
input gestures, or using information provided by other applications. If compared with
CITA and Funf, Tasker provides richer sentience as it is capable of providing sensor
1
http://tasker.dinglisch.net/.
values and more articulated/complex contexts, which are richer, modular, and easily
extensible in an app-specific way by developers.
The Pervasive Middleware for Activity Recognition (PEMAR) framework [13]
aims to increase the level of physical activity by creating a middleware for active games
on mobile devices. PEMAR can to recognize human motions based on human activities
and their contexts, but it lacks as well as an underlying theory of persuasion, though
games are used as a method of persuasion.
Google recently released Google Fit [14] - an open platform that allows users to
control their fitness data, and supports developers build health and fitness apps. It also
targets manufacturers and aims to influence their future device designs to take
advantage of. With Fit and the Google Play services location APIs2, developers should
be able to easily create health and wellness applications. Nevertheless, without proper
support for persuasion, developers have to create a lot of persuasion-related code
behavior from scratch, which is time consuming and involves a steep learning curve.
MoST [15] is a smartphone sensing library developed at the University of Bologna.
From the functional point of view, it is very similar to Google Play services location
APIs, but it is open source, lightweight, modular, and more efficient in terms of
resource usage, in particular battery.
Finally, DailyAlert [16] is a generic mobile persuasion toolkit for smartphones that
generalizes and automates persuasion as a service for mobile applications. It follows
the thin-client design to reduce power consumption on users’ devices and to simplify
integration with mobile applications. Although DailyAlert is a complete persuasive
toolkit, similar to Cicero, it is not based on explicit theoretical foundation and models.
3 Cicero Design
To design a mobile technology specialization (or

translation) of ABM, its cyber sense, cyber influence,
and its Assess Tree algorithm, we limit the cyberspace
and its devices to Android smartphones/tablets and
Android Wear. The set of available sensors on these
devices are identified and explicitly made accessible to
our middleware. Similarly, following a pyramid design
methodology [9] as shown in Fig. 2, a rich set of
potential activities and contexts are defined and
Fig. 2. Cicero design pyramid
explicitly supported in the middleware. Together, they
constitute the second layer of the pyramid. It is impor-
tant to note that information about activities could be used to recognize applicable
contexts (hence the larger representation of Activity compared to Context in this layer).
Finally, in the top of the pyramid, programmable situation structures that may utilize
any of the predefined contexts, activities or device interactions are supported. Situa-
tions are key to implementing the Assess step in the ABM.
2
https://developers.google.com/android/reference/com/google/android/gms/location/package-
summary.
It is worth noting that device interaction is co-located with the situation layer
because interactions with end users (e.g., through screen views, flash LED, vibration,
or integrated speakers) are often coordinated based on materialized situations.
3.1 Sensors
Most Android smartphones/watches have built-in sensors capable of providing high
precision/accuracy raw data to measure motion, orientation, low-level activities, and
various environmental conditions. We divide such sensors into three broad categories:
motion sensors (used to measure acceleration and rotational forces along three axes,
such as accelerometers, gyroscopes, and vector sensors), environmental sensors (for
various environmental parameters, such as humidity and pressure), and position sensors
(used to measure the physical position of a device, such as GPS or Wi-Fi)3.
3.2 Activity
Cicero maximizes the leverage of existing Android features especially Google Play
Services, by offering rich APIs such as ActivityRecognitionAPI and the FusedLoca-
tionProviderAPI for managing motion and position sensors4. Combined, they help
recognize several user activities including still, tilting (device angle related to gravity
change significantly), on foot, walking, running, on bicycle, in vehicle, and unknown.
To support a broader set of activities, we used additional libraries and APIs offered
by the Android platform including SensorManager5, which is used to manage envi-
ronment sensors, TelephonyManager6, which is used to sense the device status, and
MediaRecorder7 used to detect sounds around the user. We have supported the fol-
lowing activities based on these APIs in Cicero: “walking”, “running”, “still”,
“calling”, “going to < Loc>”, “being at < Loc>”, and “near < Loc>”. Developers are
provided with a very simple interface to register and detect occurrences of activities.
3.3 Context
In addition to the above activity recognition features, situations as defined in ABM
need more details – contexts and device actions. Context is determined by wiring
(combining) data that derive from different sensors. We denote gathered sensor data
with the symbol SEk ðxÞ which represents the sensed data by the kth sensor.
Using AND, OR, and NOT operators, we define the following three context models:
3
http://developer.android.com/guide/topics/sensors/sensors_overview.html.
4
https://developers.google.com/android/guides/overview.
5
http://developer.android.com/reference/android/hardware/SensorManager.html.
6
http://developer.android.com/reference/android/telephony/TelephonyManager.html.
7
http://developer.android.com/reference/android/media/MediaRecorder.html.
(1) Locational Model: context combines sensor data collected in different locations,
(2) Snapshot Model: context combines sensor data collected in different points in time,
and (3) Value Model: context as in (1) and (2) but defined over ranges of values rather
than specific single sensor values. This model is more practical to use in real-world
applications. Context under Value Model can be expressed as follows:
C ¼ SE1 ðva ; vb Þ op SE2 ðva ; vb Þ op. . .op SEk ðva ; vb Þ
Where va and vb represent the two extremities of the range of a given sensor SEi.
3.4 Situation
By recognizing different activities, contexts, and device actions, it is possible to rec-
ognize the occurrence of predefined situations that will enable the SAT assessment
algorithm and the ABM Access step. We utilize a simple quorum-based decision
method to recognize situations: if two or more elements of a situation are true (out of
three, i.e., device action, context, and activity), the situation is considered to have
occurred [3].
4 Cicero Implementation
In this section, we further explain the main characteristics of Cicero and its compo-
nents. Cicero works on mobile devices and in a layer underlying persuasive apps. For
the developer, Cicero is a middleware and a tool that simplifies the development
process of persuasive applications. For the end-user, Cicero is a run-time under which
Cicero-based apps execute. Developers can create persuasive applications by setting
cyber sense and cyber influence and, in general, by setting and programming each step
in ABM. This includes setting goals and contents relative to the Aware and Learn steps,
setting suggestions and reminders relative to the Recall step, and setting Rewards based
on available incentives. End users are transparently guided by ABM and its steps: they
can use the app by setting goals, checking their own progress, interacting with
reminders, and obtaining rewards.
The Cicero architecture, shown in Fig. 3, consists of two main managers: sentience
manager and cyber manager. The former is concerned with sensing in general and
allows the developer to select contexts and activities to monitor. The cyber manager
works to facilitate the developer’s work to set and program the various steps in ABM. In
addition, it is concerned with recognizing situations and monitoring the progression of
the various steps. Periodically, the cyber manager would evaluate how the user con-
forms to the settings of the model (progress towards goals) through the SAT algorithm.
SAT evaluations are done in the cloud whereas all other processing of the cyber manager
is done on the hosting mobile device. We give the design/implementation details of each
manager below.
4.1 Sentience Manager (SM)

SM accesses and gathers sensor data using listeners and filters data in order to evaluate
the achievement of goals. SM implements the Act step of ABM, thus it is initialized
and processed by the cyber manager. In order to manage the different sensors, SM is
further divided into several components:
• Location: used to manage position sensors. This
component is based on FusedLoca-
tionProviderAPI (an efficient module integrated
in Google Play Services).
• Motion: related to the motion sensors. This
module can detect in background the user’s
activities using ActivityRecognitionAPI offers
by Google Play.
• Environment: for handling environmental sen-
sors, such as amplitude or lightness.
• Social: related to social activities such as mak-
ing or receiving phone calls, and the frequency Fig. 3. Cicero architecture
of such activities.
• TimeService: service to manage and filter
sensed data about time context.
These components, in order to manage gathered data, interact with an Android
service running in the background that allows filtering data and the assessment of the
goals achievement. In order to do that, this service owns a pool of objects, each of
which is waiting to be concurrently updated from related components in SM regarding
a particular situation.
Another important responsibility of SM is sending events to the cyber manager in
order to detect the different situations that drive the SAT algorithm evaluation.
4.2 Cyber Manager (CM)

CM implements the ABM steps except for the Act (which is managed by SM). For
Aware, Plan, Learn, and Recall steps the developer can set own activities into adapters
offered by Cicero. Aware and Learn steps should inform and educate the end user, thus
they have to show influential contents. Cicero facilitates for such content to be placed
by a domain expert into a shared Google Drive folder accessible to the user through the
app. This enables the collaboration and participation of the domain experts, for
example physical therapists or doctors, who know best what content are appropriate,
and who are capable of using something as simple as Google Drive (but not any more
complex than that). The Plan step, depending on the goal-setting theory, should guide
the user during the configuration of variables that characterize the goals. After the Plan
step is set, CM initializes the SM using the scanning classes and SM’s own initial-
ization services.
Reminding the user is a very important aspect of persuasive apps so the user can
behave and converge towards the set goals. Thus, Recall step is realized by Cicero
using Android Notification triggered according to set schedules at configuration time,
and by SAT through reinforcement loops. Cicero sends notifications only when the
device is active to ensure user’s attention is secured. When the device is locked the
notification is queued and sent as soon as the user unlocks the device.
The last module of CM is the Assess step that implements the SAT algorithm to
evaluate the changes in user’s behavior and compliance (response to the persuasive
influences of the various steps). SAT tree is shown in Fig. 4. While the details of SAT
can be found in [1], here we point only to the facts that internal nodes in the level below
the root maps to the ABM steps, while the lowest level nodes which are positive or
negative user behaviors are essentially situations. Domain experts prescribe (describe)
the behaviors, while the developers use Cicero to implement behavior nodes into
situations using context, activity and device interactions. Evaluation is currently hosted
in the cloud as a Google App server, offering an Assess service. In fact, the domain
expert can configure all variables of SAT by an online service.
Fig. 4. Situation based Access Tree (SAT)
5 Cicero Use Case
In order to better describe Cicero and to demonstrate its benefits to developers, we

present a contrived use case in which we use Cicero to develop a mobile application to
help obese individuals lose weight. We present the use case and analyze how using
Cicero greatly eases the development tasks and reduces the total work effort required
by the developer. We describe the use case scenario and then describe the steps taken in
developing the application.
Scenario. San Raffaele is a local clinic which cares for numerous patients with obesity
related diseases. Clinic executive board decided to tap into mobile technology to help
reduce the impact of obesity among its patients and to better manage their chronic
conditions. San Raffaele contracts with a local software firm to develop a mobile app
for this purpose. The firm assigns Michael a chief mobile app developer. Recognizing
that his firm never possessed experience in user empowerment and persuasion theories,
he researched how a mobile application may be able to motivate users. Then, after a
brief search he finds Cicero, a middleware to support his team in developing the
persuasive mobile app to the local clinic.
Application. Following Cicero’s footsteps, Michael realizes that the app needs to
include an Aware step to make users aware of why it is so important to walk in order to
improve health. He follows Cicero guidelines and understands that content in this step
needs to be provided by the domain experts. Michael communicates with the San
Raffaele clinic and obtains a list of hyperlinks that provide awareness about this topic
as well as other information needed by other steps such as Plan step. Next, Michael
realizes that a user Profile needs to be captured (see Fig. 1). He implements the Profile
one-time step that prompts the users to insert personal information including the
location of their gym in order to be able to tailor the application to their needs. Next,
Michael completes the Plan step which sets a target schedule and instructs the users
about the proper way to walk (e.g., it is not healthy walking under a bright sunlight
midday), showing a list of links to useful web pages or videos (Learn step).
The final tasks in the development process involves collaboration and communi-
cation with San Raffaele’s experts to obtain descriptions of the negative and positive
behaviors which the developer will map into P- and N-nodes as per the SAT tree
structure in Fig. 4. Once these nodes are created, the application is essentially complete
and is able to sense users’ motions and locations which automate the recognition of
contexts, activities and eventually situations. This allows the SAT tree to assess
changes in users’ behaviors. Lack of response to persuasion automatically invokes
reinforcement loops. By rolling back the non-compliant user to the appropriate pre-
vious step. Also, based on the Plan step, the application automatically sends to users
reminders in order to assist them in reaching their set goals. Michael gets a firsthand
experience with Cicero and sees how it makes it extremely easy and fast for him to
develop the first prototype of the app to demo it to San Raffaele. As will be shown
below, Michael did not have to write the application from scratch, to the opposite, by
relying on Cicero, he ended up coding only small sections of code.
Implementation. The first task for Michael is to code four Android Activity classes,
one for each related ABM step. Then Michael sets these Activities in the Cicero
middleware, using the related setActivityAdapter method of CM. After that,
Cicero will start them when it is appropriate. For instance, Michael creates an Activity
called AwareActivity that shows the users a list of hyperlinks about the importance
of walking. Michael is free to develop that Activity as he wishes, such as using a list of
links to web pages or integrating it with documents stored in a shared Google Drive
folder. After that, he informs Cicero that his AwareActivity is the Activity that
implements the ABM Aware step, using the setAssessActivityAdapter

method. It is important to notice here how Cicero streamlines development and pro-
vides valuable reusable code yet at the same time allows the developer some freedom
and flexibility. Indeed, developers can use reusable Android Activity, only setting them
in the CM, allowing simplicity of use and reusability.
Next, we show how situations are defined and created. In dealing with delicate
pathologies (e.g., Diabetes or Bipolar diseases), collaboration between the developer
and domain experts (who are an authority in deciding on positive and negative
behaviors) is required. In particular, developers have to engage domain experts, asking
for behaviors that should be usefully tracked in that pathology. Situations must also be
categorized or mapped into sub-action nodes. The domain expert does not need to
understand the working of the SAT tree but should be fully aware of the concept of
positive and negative behaviors and also of the meanings of the ABM steps. Situations
related to this use case are shown in Table 1. They are described in simple phrases that
involve a metric or two (e.g., 20 min, daily, or a specific location). These phrases
should be the output from the domain expert. Cicero natively supports all the situations
identified, therefore, Michael can develop his application completely using the
middleware.
Table 1. Situations setup in the San Raffaele use case

Action Sub-action node P-or-N Situation
Node Node
Aware Behavior-Aware P Node P1: Walk more than before
N Node N1: Walk less than before
Plan Goal-Setting P Node P2: Complete set goals
Status N Node N2: Do not set goals
Learn Behavior P Node P3: Do not walk in bright sunlight
Knowledge N Node N3: Walk in bright sunlight
Act Goal-related P Node P4: Walk 20 min daily
Activity P5: Go to gym twice a week
N Node N4: Walk < 20 min daily
N5: Do not be at the gym twice in a
week
Now that situations are clearly defined, it is time for Michael to implement the
situation and to build up the SAT using his implementation. In order to do that,
Michael calls suitable methods of CM. For example, in order to add the P4 situation in
Table 1, Michael invokes the method addGoalRelatedActivityPNode of CM passing in,
among other parameters, ActivityName.Walking, ContextName.Duration, and 20 min
as reference value. Then, Cicero creates the related Leaf Node, adds it to the SAT tree,
and starts the sensing thread of user motions. This will begin the SAT algorithm
evaluation of user responses.
After all situations are added to SAT tree, Michael programs some notifications to
be displayed occasionally when the user switches on the device. In this way, the user
is more involved than popping out a notification when he or she is not actively using
the smartphone. Michael only creates the notification messages (subclass of Android
Notification) but Cicero schedules them according to the developer’s preferences
(Profile), such as frequency or preferred times. Moreover, other recalls based on sit-
uations are possible. Which notifications Michael implements is a matter of design
choices and developer creativity.
By completing the steps described above, developers have a ready-to-test and use
application. Clearly, it implies a great saving in time and amount of written code if
compared to creating the same application from scratch. Indeed, to develop from
scratch a quite complex application as in our use case, a developer would have to code
dozens of classes, not to mention the amount of time to research and understand
theories about persuasive technologies/applications. Nevertheless, relying on Cicero,
the same developer has to write just few classes, where only about few tens of code
lines relate to the interaction with Cicero to fully develop the persuasive application.
6 Some Experimental Performance Results of the Cicero

Middleware
To assess the performance of Cicero in different conditions, we utilized the use case as
a driver for several evaluations. An application developed relying on the use case was
used daily for a week by a 25-years-old user. The device used was a smartphone Nexus
5 featured by Android 6.0. By using Android Debug Bridge8 and Dev Tool9 by
Android Studio, we have thoroughly analyzed CPU and RAM usages. In particular,
considering the Act step, which is the only one that runs in the background, the CPU
usage reached a peak of 2 %, but demonstrated to be lower on average; the used RAM
was 19 MB with an assigned heap of 24 MB.
Battery is recognized as the most challenging resource for many types of mobile
devices. The Cicero battery consumption was measured by using Trepn Profiler to
assess the level of consumed energy10 of the mobile device: by using daily profiling
that mimics the usual daily behavior of the considered target users, we have found a
battery consumption of 7 % on average per day. Cicero is also supported by Google
Play Services; the energy consumption by Google Play has demonstrated to be an
additional 12 % daily.
Finally, we have also experimentally evaluated the Cicero latency and promptness
to answer user’s requests. By considering the relevant aspect of user’s activity
detection, the ActivityRecognitionAPI has shown performance indicators that depend
on the targeted accuracy of detection; for the sake of briefness, here we report that, in
the case of detection of a motionless device, latency is on average 550 ms from the
starting of the scan and the detected (motionless) activity, with a confidence of 100 %.
8
http://developer.android.com/tools/help/adb.html.
9
http://developer.android.com/tools/debugging/debugging-devtools.html.
10
https://developer.qualcomm.com/software/trepn-power-profiler.
7 Conclusion
We presented Cicero, a middleware and tool for developing persuasive mobile apps.
Cicero is based on the Action-Behavior Model and assessing user responses to per-
suasion. We described the design and implementation of Cicero and provided a use
case to explain how developers can use it to develop apps within a defined collabo-
ration framework with domain experts. Cicero saves significant development time and
waives the requirement that developers need to understand elements of persuasion
theory. We presented results of preliminary experiments that characterize Cicero per-
formance and energy use. Cicero software and documentations can be obtained from
the Cicero download site11. An important future work will provide for a
community-based shared resources that document domain experts contributions in the
form of short white papers describing positive and negative behaviors related to
common or special chronic conditions.
References
1. Lee, D., Helal, S., Sung, Y.S., Anton, S.: Situation-based assess tree for user behavior
assessment in persuasive telehealth. IEEE Trans. Hum. Mach. Syst. 45(5), 624–634 (2015)
2. Lee, D., Helal, S., Sung, Y.: Assessing behavioral responses in persuasive ubiquitous
systems. In: Biswas, J., Kobayashi, H., Wong, L., Abdulrazak, B., Mokhtari, M. (eds.)
ICOST 2013. LNCS, vol. 7910, pp. 176–186. Springer, Heidelberg (2013)
3. Lee, D., Helal, S.: From activity recognition to situation recognition. In: Biswas, J.,
Kobayashi, H., Wong, L., Abdulrazak, B., Mokhtari, M. (eds.) ICOST 2013. LNCS, vol.
7910, pp. 245–251. Springer, Heidelberg (2013)
4. Lee, D., Helal, S., Anton, S., De Deugd, S., Smith, A.: Participatory and persuasive tehealth.
Gerontology 58, 269–281 (2012)
5. Lee, D., Helal, S., Johnson, B.D.: An action-based behavior model for persuasive telehealth.
In: Lee, Y., Bien, Z., Mokhtari, M., Kim, J.T., Park, M., Kim, J., Lee, H., Khalil, I. (eds.)
ICOST 2010. LNCS, vol. 6159, pp. 121–129. Springer, Heidelberg (2010)
6. Riley, W.T., Rivera, D.E., Atienza, A.A., Nilsen, W., Allison, S.M., Mermelstein, R.: Health
behavior models in the age of mobile interventions: are our theories up to the task? Transl.
Behav. Med. 1, 53–71 (2011)
7. Fogg, B.J.: A behavior model for persuasive design. In: Proceedings of the Persuasive.
ACM, New York (2009)
8. Berdichevsky, D., Neunschwander, E.: Towards an ethics of persuasive technology.
Commun. ACM 42(5), 51–58 (1999)
9. D’Aloia, A.: Persuasive Computing: an Android-based Framework for Developing and
Managing Persuasive Applications (2015)
10. Ravindranath, L., Thiagarajan, A., Balakrishnan, H., Madden, S.: Code in the air:
simplifying sensing and coordination tasks on smartphones. In: Proceedings of Hot-Mobile.
ACM (2012)
11
http://www.icta.ufl.edu/cicero.
11. Kaler, T., Lynch, J.P., Peng, T., Ravindranath, L., Thiagarajan, A., Balakrishnan, H.,
Madden, S.: Code in the air: simplifying sensing on smartphones. In: Proceedings of SenSys.
ACM (2010)
12. Aharony, N., Pan, W., Ip, C., Khayal, I., Pentland, A.: Social fMRI: Investigating and
shaping social mechanisms in the real world. Pervasive Mobile Comput. 7(6), 643–659
(2011). ACM
13. Vaka, P., Shen, F., Chandrashekar, M., Lee, Y.: PEMAR: A pervasive middleware for
activity recognition with smart phones. In: PerCom Workshops. IEEE (2015)
14. Google Fit SDK: https://developers.google.com/fit/
15. Cardone, G., Cirri, A., Corradi, A., Foschini, L., Montanari, R.: Activity recognition for
smart city scenarios: Google play services vs. MoST facilities. In: Computers and
Communication (ISCC). IEEE (2014)
16. Zhan, A., Lim, J.H., Terzis, A.: DailyAlert: A generic mobile persuasion toolkit for
smartphones. In: PhoneSense. ACM (2011)
Formalization of Computational Human
Behavior Models for Contextual Persuasive
Technology
Tylar Murray1,2,3,4(B) , Eric Hekler1,2,3,4 , Donna Spruijt-Metz1,2,3,4 ,

Daniel E. Rivera1,2,3,4 , and Andrew Raij1,2,3,4
1
University of South Florida, Tampa, USA
[email protected]
2
Arizona State University, Tampa, USA
3
University of Southern California, Los Angeles, USA
4
University of Central Florida, Orlando, USA
Abstract. In theory, Just-in-Time Adaptive Interventions (JiTAIs) are

a persuasive technology which promise to empower personal behavioral
goals by optimizing treatments to situational context and user behaviors.
This paper outlines open challenges facing the development of JiTAIs
and discusses the use of modeling as a common ground between behav-
ioral scientists designing interventions and software engineers building
applications. We propose that Computational Human Behavior Mod-
eling (CHBM) has the potential to (1) help create better behavioral
theories, (2) enable real-time ideographic intervention optimization, and
(3) facilitate more robust data analysis techniques. First, a small set of
definitions are presented to clarify ambiguities and mismatches in ter-
minology between these two areas. Next, existing modeling concepts are
used to formalize a modeling paradigm designed to fit the needs JiTAI
development methodology. Last, potential benefits and open challenges
of this modeling paradigm are highlighted through examination of the
model-development methodology, run-time user modeling, and model-
based data analysis.
Keywords: Persusasive · Just-in-time · Adaptive · JiTAI · Modeling ·

Computational human behavior modeling
1 Introduction
A recent trend in the area of persuasive technology is the development of mHealth

applications which aim to deliver better, smarter, and more effective interven-
tions via mobile and wearable devices. One class of persuasive technologies with
this aim is the “Just-In-Time Adaptive Intervention” (or JiTAI) which describes
an intervention that adapts to an individual’s changing needs and circumstances
to deliver tailored support at the time when it is most needed. [12] Thanks to

DOI: 10.1007/978-3-319-31510-2 13
CHBM Formalization 151
great advances in wearable, mobile, and ubiquitous technologies increasingly rich

data is now available to characterize the context of the subject for use in JiTAI
applications.
Researchers theorize that an intervention which can tailor based on the user
and context may be an elegant solution to empower self-management of unhealthy
behaviors like substance abuse, overeating, sedentary behavior, and more [11].
These persuasive technologies aim to utilize contextual information — data col-
lected from the subject’s surroundings and history — to deliver personalized inter-
ventions at the optimal moment in time. Real-time monitoring of data to identify
states of special vulnerability to poor behavioral decisions or receptivity to inter-
vention at any given moment is possible [9], but “a major gap exists between the
technological capacity to deliver JITAIs and existing health behavior models.”
[11] Proof-of-concept applications have demonstrated the ability to adapt inter-
ventions to users [2,5] and context [3,4], but using current methods the complex-
ity of the behavioral model underlying a JiTAI application grows exponentially
as the complexity of the intervention design increases. The needs of a persuasive
technology are very different from the needs of extant behavioral research. While
the latter places emphasis on the study of the human system’s intricacies, the for-
mer needs a model which provides generalized insight and specific numerical pre-
dictions. Behavioral theories traditionally focus on nomothetic and static insights
that do not offer the granularity and specificity to support the full potential of
JiTAIs [14]. These methods are sufficient for analysis of traits which do not change
much over days or weeks, but data collection and intervention delivery timing is
now available to the microsecond for physiological data, behavioral features at the
minute-level, and psychological constructs (via EMA [15]).
The current development process for JiTAI-like persuasive technologies
requires close collaboration between behavioral scientists and application devel-
opers as they struggle to code-ify the model from extant behavioral theories for
each individual experiment. The models used by a programmer to describe a user
and the models used by behavioral scientists to describe a subject, have certain
key differences which can complicate the process of JiTAI design. In this work
we present a hybridization of the two modeling paradigms designed to emphasize
the strengths of each approach. As a part of this set of interdisciplinary terms,
we introduce the concept of a Computational Human Behavior Model (CHBM)
to describe this new class of models which aim to satisfy the demands of per-
suasive technology. Following definitions, we propose that by formalizing the
CHBM underlying persuasive applications, it will be possible to create better
behavioral theories, enable real-time ideographic optimization, facilitate more
robust data analysis, and reduce application development time. In this section
we present a look at how the concept of a CHBM would be applied to address
open issues holding back JiTAIs and we highlight the remaining issues which
must be addressed to make Just-in-Time Adaptive Interventions a reality.
2 Selected Definitions
This section presents definitions and design considerations relevant to human-
behavior modeling from a theory-agnostic standpoint so that different modeling
152 T. Murray et al.
paradigms can be described under a common foundation. This set of definitions

draws from both the area of HCI user-modeling and the extant paradigms of
human behavior modeling in behavioral science in an attempt to synthesize
a pragmatic language for use in the development of persuasive technology by
behavioral scientists and application developers alike.
Treatments are defined by M.C. Kaptein [10] as the set of messages or feedback
a user receives from a persuasive application. The term treatments seems syn-
onymous with interventions in usage, but a single treatment should be used to
unambiguously represent a single instance of user-interaction, whereas a single
intervention may represent a set of interactions given as a dose.
Just-in-Time (JiT) is a cross-disciplinary concept defined in the context of

behavioral interactions by Nahum-Shani et al. as “the effective provision of
timely support, operationalized by offering the type of support needed, precisely
when needed, in a way that minimizes waste (i.e., defined as anything that does
not benefit the person) and accommodates the real-life setting in which sup-
port is needed.” [11] Thus, for an intervention to be considered Just-in-Time
(JiT), it must attempt to deliver treatment immediately before or after an event
associated with the target behavior.
Adaptive interventions must utilize dynamic (time-varying) “information from

the person (e.g., changes in psychological distress, response to an intervention,
intervention adherence) [...] to make intervention decisions repeatedly in the
course of the intervention (e.g., changing the type, dosage, or timing of interven-
tion delivery).’ [11] An adaptive intervention is one that responds in real-time
to the changing needs of the participant by tailoring the treatment itself based
on situational context or the recent behavioral history of a user.
Individualization is defined by Nahum-Shani et al. as the “use of [static] infor-

mation from the individual to make decisions about when, where and how to
intervene.” [11] Thus, an intervention is individualized if “relatively stable infor-
mation from the person (e.g., gender, baseline severity of symptoms) is used
to make intervention-related decisions (e.g., to offer intervention package A or
B)” [11] For example, a stress-relief intervention regimen may utilize relaxing
music treatment based on the subject’s favorite songs at study initialization, or
a participant’s favorite color may be used as the basis for the user interface color
palette.
3 Computational Human Behavior Models
The following specification will allow for the formal description of a CHBM,
providing a standard approach to describing, designing, and visualizing human
behavior models for persuasive applications. A Computational Human Behavior
Model (CHBM) is defined here as a mathematical, explicit model which describes
how context is transformed into a behavioral outcome through the internal state
of the human system. In summary, a Computational Human Behavior Model
(CHBM) should have (1) a set of context, state, and behavior variables, (2) a
set of computations which define behavior variables as a function of state which
is itself a function of context, (3) a logical abstraction which allows researchers
to internalize the model’s behavior such that they will be better able to esti-
mate control of the human system in general, and (4) guidelines regarding the
applicable population and time-scale of the CHBM. The following section details
each of these CHBM components, followed by a methodology which makes use
of a graph representation to create and describe a particular CHBM.
3.1 Characteristics of a CHBM

User Features: Context, State, Behavior. A distinguishing feature of a
CHBM is the separation of the subject definition into environmental context,
internal state, and behavior variables. In reality, an individual represents an
inseparable component within the larger environment, but this simplification
segments out the human system for definition.
In most cases, it is sufficient to define context as a set of selected information
from the environment available for inflow into the human system, but contex-
tual information from the environment may be summarized and represented in
countless ways. In reality, consider context to be everything that is observed by
the senses. Some of this information will alter the internal state of the human
system, but some may not. When building a model based on theory alone, mod-
elers should make the selection and summary of contextual constructs to be
as generalizable, extensible, and reusable as possible. When utilizing a model
to simulate a particular experiment, efforts to connect avaialable data to that
which is available during the experiment may be needed, and contextual infor-
mation not available empirically may need to be simulated. The environmental
context influences the human system, which has an internal state represented
by a set of internal state variables. In reality, internal state includes all informa-
tion stored in the chemical and physical arrangement of our bodies. In order to
make the model tractable, the mass of information is summarized into a set of
meaningful constructs. Information flowing into a CHBM comes from the envi-
ronment around an individual (the context) as an inflow which is independent
of the individual’s state in this instant. Similarly, information flowing out of a
CHBM (as behaviors) represents actions the individual is taking to impact the
environment.
Relationships Between User Features. The relationships between context,

state, and behavior variables in a CHBM must be defined computationally. The
functional form of these computations is not constrained in this definition, the-
oretically allowing for the representation of any intervariate relationship. There
are numerous benefits to keeping the functional form of these relationships sim-
ple and homogenous across variables. Last, a simple formulation is more easily
understood, allowing for a straightforward interpretation and abstraction of the
model behavior.
Heuristic Interpretation. Statistical models trained on data do qualify as

CHBMs in that they can define the relationships between state and context,
but typically do not incorporate a logical abstraction of cognition and instead
treat the internal state as a black box. This abstraction is essential when con-
sidering the process of JiTAI design, since the search-space available to a JiTAI
designer can only be approached through heuristics guided by an understanding
of how the human system will generally behave under given conditions. Though
mathematical equations themselves reveal the nature of the system, naming and
describing the interpretation of specific constructs or coefficients which play piv-
otal roles in the model can aid in the process of internalizing model behavior.
Model Metadata. While a CHBM should strive to be as broadly applicable

as possible, this inevitably comes at the cost of increased complexity which can
make the CHBM’s nomothetic abstraction(s) intractable; there is a balance to be
struck between a CHBM’s inclusivity and the clarity of the abstraction. For this
reason, it may be important to specify the circumstances in which a given model
is valid. This is not analogous to the issue of over-fitting in machine learning,
as the model can remain accurate across the population; the primary reason for
limiting the number of variables or the functional complexity of relationships is
to preserve the heuristic understanding of the model.
3.2 Creating a CHBM

A network graph is an effective abstraction to describe the relationships (rep-
resented by arrows or “edges”) between variables (represented by the graph’s
“nodes”) in a CHBM. In this case a directed graph wherein edge arrows rep-
resent the flow of information between nodes is used. Thus, a directed graph
edge from node A to node B indicates that information flows from node A into
node B. This relation can be read as “A influences B”, “A informs B”, or similar.
This choice of notation is in agreement with graphs used in information theory,
communications models, and behavioral science. In contrast, some graphing par-
adigms (such as probabilistic graphical models and software design) prefer to use
notation wherein an edge is used to represent dependency.
While the network graph shows the connectivity of a model, it fails to indicate
the meaning of each connection. In the majority of existing applications, the
mathematical form of the relationship is implied or else it is neglected completely.
The most common analyses assess linear relationships between variables, and
thus it is perhaps reasonable to assume that this is the intention of most authors.
Assuming this is the case we can return to our simplistic example in Graph 1
and interpret the implied relationship as:
B(t) = coef fab A(t) + constb (1)
In this formulation coef fab represents the correlation coefficient which relates
A to B, and constb represents a scalar constant. For nodes with multiple inflow
edges, such as node B in the following graph:
A => B <= C => D (2)
Continuing with our assumption that node interrelations act as linear sums,
the resulting formulation is simply a sum of the inflows:
B(t) = coef fab A(t) + coef fcb C(t) + constb (3)
Using this formulation, the general form of the CHBM is expressed via the
network graph alone. The general solution of an CHBM does not require def-
inition of the constants, but a simulation cannot be run until some numerical
value is assumed. These constants often have theoretical significance in that they
often have meaningful influence upon system behavior. Scaling-coefficients, for
instance allow for relative weighting of each inflow. Similarly, the coefficients of a
dynamical equation define how quickly variables react to a change “upstream”.
This linear, homogeneous-graph representation is useful, but also very lim-
ited. One important feature which this formulation does not take into account is
the dynamics of the relationship. For instance, the above linear model assumes
that there is no delay between variables. This assumption is fine for some appli-
cations, but this is a very poor assumption for human behavior models.
Differential equations based on a fluid-flow analogy can be used to describe
the relationship between variables as described by Dong et al. [7]. Using the
differential formulation our equation for B in Graph 1 becomes:
dB
B(t) = coef fab A(t − θab ) − τb + const (4)
dt
Just as before, our general model is not expressed entirely through the graph,
and an ideographic example is specified by providing table of coefficient values.
Our table is now quite a bit larger, but these coefficients have meaningful defin-
itions which relate to our theory. While this formulation offers a huge improve-
ment over the linear formulation, we can still imagine relationships which it
cannot express.
It should be noted at this point that although the linear formulation is too
simple to express the dynamics of the differential formulation, the differential
formulation is capable of expressing linear relationships. This is accomplished
by setting coefficients of dynamical components to 0. One might think, then,
that there is some general formula which could express any functional form, and
that this form should be used to express the relationships between variables in
all CHBM graphs. While such formulations do exist (such as Taylor or Fourier
series approximations or even ANN-based relations), this usage tends to make
the model difficult to understand and to simulate with. Linear and differential
formulations are in such widespread use because of the relative ease with which
we can understand and solve them.
Let us now consider the case where a graph-wide assumption is NOT made.
That is, we will specify the functional form of each node individually so that each
edge on the graph may be linear in form while another may be differential. This
has the benefit of allowing for both complex relationships between variables as
well as simplistic ones. In this way one could craft a model in which two variables
are linearly related and a third is dependent on the variance of another variable
(a particularly odd formulation, but one which is relevant to behavioral theory).
Unfortunately, this approach also means that a table of formulations must now
be included with our graph to show the meaning of each edge in the graph.
Consider for example the table below for Graph 2:
Node Formulation
B coef fab A(t) + coef fcb C(t) + cosnstb
D coef fcd C(t − θcd ) − τd dD
dt
+ constd
If a fixed number of functional forms is adhered to, the graph can be made
to visually represent these functional forms through the use of different node
icon shapes. This approach quickly begins to resemble applications which use
flow-based programming. Indeed, they are quite similar in their approach, and
the specification of a CHBM is quite similar to the writing of a program.
In conclusion, we propose that an CHBM should be specified using the fol-
lowing rules (1) use a graph-wide formula assumption if possible, else specify
formulations for each node individually, (2) when choosing a formulation, con-
sistency between nodes is most important, (3) when choosing a formulation,
simplicity and clarity is second only to consistency.
4 Benefits of CHBM-enabled JiTAIs

This section discusses the utility of a CHBM throughout the lifecycle of a JiTAI
application. Hypothetical situations are posed to highlight the potential value
of CHBM use in the JiTAI development process and show open challenges
through establishment of a target user group model, application design, applica-
tion implementation, data analysis, model personalization, and model iteration.
4.1 A Priori CHBMs

Prior to development of a JiTAI, a mental model of the target user group is
established. This a priori model represents the researcher’s understanding of
the user group, and the design of the intervention utilizes the model in order
to predict user actions. This level of detail to which this model is documented
varies greatly between applications, and in some cases the causal descriptive
model has little grounding in existing behavioral theory [13]. Nevertheless, a
vague description of expected user behaviors and interactions with the persuasive
technology still represents a user model. Existing JiTAI-like applications may not
have a CHBM, but they always (sometimes informally) imply a CHBM. This
section highlights the benefits of defining a CHBM explicitly, rather than relying
on implicit behavioral theory.
Model Building. When model-building for a JiTAI, the planned system and
underlying model of human behavior becomes very complicated, and user
responses may be difficult to predict through thought experiments. Without
a concrete framework to describe the model, user behavior becomes oversimpli-
fied, giving an even less accurate picture of the complex human system. When a
model is under-developed, the application development process will open unad-
dressed questions and simple assumptions will be made. For instance, delivery
of a treatment may be limited to the waking hours or to the weekdays, but
this will not be reflected in the described user model. The mismatch between
the documented theoretical model and the actually implemented model further
muddle the process of study replication and analysis.
In addition to those assumptions knowingly made by application developers,
causal descriptive modeling often contains implicit assumptions which are eas-
ily overlooked. For instance, the delay between a cause and effect is frequently
neglected — that is: how quickly does a participant’s behavior respond to an
treatment? The process of defining a more detailed a priori model itself can
lead to new insights and research questions by eliminating these oversights and
forcing critical thinking on the assumptions being made.
Intervention Design. When designing intervention options for a JiTAI appli-

cation, researchers will consider how a treatment influences the subject in the
context of the chosen user model. When using a CHBM, this means quantifying
the treatment’s effect on user context. For instance, consider an intervention
which provides information about the health repercussions of sedentary behav-
ior. Assuming our CHBM uses an adaptation of the Theory of Planned Behavior
[1], this intervention targets behavioral belief regarding sedentary behavior. Since
behavioral belief is part of the internal state and the treatment should be defined
as part of the user’s context, a context variable should be included in our model
to represent external influences on behavioral belief from the environment. After
defining the expected effect of a single treatment, the CHBM can then be used
to predict a detailed account of user response. The use of simulations such as this
in the process of designing controls is well-explored in many other areas, but is
nearly unheard of in behavioral science. This is in part due to the prevalence of
abstract causal descriptive models and the novelty of CHBMs, but there remain
several important issues highlighted below which have not yet been addressed
in this space.
Benefits of CHBMs in Persuasive Design:
1. By using a CHBM with dynamical equations, the dynamics of relationships

between variables can be explicitly described as a part of the model.
2. The use of an explicit a priori model for intervention design helps researchers
formulate testable research questions and experiment designs.
3. The additional pre-study detail removes post-study modeling assumptions
that can dilute the underlying behavioral theory or invalidate study results.
4. The process of defining a CHBM itself can lead to new insights and research
questions which are almost entirely unaddressed by existing theory.
Open Questions for CHBM-Empowered Persuasive Design:

1. The process of defining a CHBM requires detailed knowledge of both the
underlying behavioral theory and the mathematics. Relatively few researchers
today possess the necessary skillset.
2. Modeling software exists for other engineering domains, but is not directly
applicable to the problem of CHBM development.
3. Software for running simulations to test the function of an a priori CHBM is
non-existent.
4. Methodologies for creating an a priori CHBM are not fully established, and
mappings from existing causal descriptive models may be model-dependent.
5. The definition of a treatment’s effect on a user is a subjective process. That
is: how is one to know what amount of behavioral belief a specific “sedentary
activity fact treatment” imparts?
6. In order to get a more realistic look at user responses to an treatments,
many simulations with varying parameters set to match the expectations of
the researchers should be run and analyzed; this would require a CHBM
simulation software suite that does not yet exist.
4.2 CHBMs at Run-Time

In this section methods in which CHBMs may be used in the persuasive technol-
ogy itself are discussed. Options include model-based intervention optimization,
timing, and online ideographic modeling.
A crucial step in the development of a persuasive technology today is to
establish a set of decision rules based on behavioral theory which codify the cir-
cumstances in which a treatment should or should not be delivered. For instance,
a treatment might be delivered only during the daytime, right before a meal, only
in a particular location, or in response to a behavioral event such as cigarette use.
Establishing a set of decision rules for a small number of conditions is feasible for
a simple intervention, but as the number of conditions increases the number of
rules required increases combinatorially. Even worse, when making use of adap-
tive interventions this set of rules must be expanded even further to map between
all possible contexts and intervention permutations. Relying on simple decision
rules loosely guided by existing theory to define the optimization of interven-
tion delivery to control a complex system inevitably leads to under-optimized
interventions, over-simplified models, and weakened data. An additional prob-
lem with this approach is the use of a binary state (i.e. rule satisfied or not) to
optimize delivery over a continuous time. Because of this oversimplification, rules
which govern the behavior often become part of the theory underlying the appli-
cation and are clumsily expressed as decision rules. In contrast, optimization of
treatment delivery using a CHBM can be done algorithmically to minimize the
area between the desired and observed target behavior.
Because CHBMs are computational in nature, prediction of behavior is possi-
ble given information about the user’s present and future context. Furthermore,
because the behaviors in computational models are quantitative, an applica-
tion could search available treatment options to find one which produces the
ideal amount of a target behavior. Methods for model predictive control are a
well studied topic of control systems engineering, but many methods cannot be
applied to generic formulations. Without a constrained form to guide optimiza-
tion, all possible options must be explored with equal feasibility in a brute-force
search. With sufficient computational power this is effective for simple problems,
but this approach becomes increasingly infeasible as the number of options and
the number of future steps to be considered increase. If the functional form
describing variable relationships is constrained appropriately, however, mathe-
matical optimizations methods can greatly simplify this problem. Applications of
model-predictive control over intervention delivery have been explored for gesta-
tional weight gain [8], smoking cessation [16], and fibromyalgia treatment [6] by
limiting the functional form of the CHBM specification to a differential equation
based on a fluid-flow analogy. In this way, application creators can implement
software utilizing the advanced understanding of behavioral science described by
the CHBM, without direct knowledge of the underlying behavioral science.
Benefits of CHBMs for Persuasive Applications:
1. Using a CHBM enables the use of optimization algorithms instead of decision
rules. This change is needed to apply complex control over target behaviors.
2. CHBMs can be adapted to fit a user’s needs at run-time, establishing an
idiographic model of each subject from the generalized CHBM.
Open Questions for CHBM-enabled Persuasive Applications:
1. Optimization of intervention delivery can be computationally expensive unless
the functional form of modelling is restricted, and it is not yet clear what
formulations are most appropriate for behavioral construct relationships.
4.3 CHBMs Post-Study

Another rising challenge for persuasive technology researchers is the increasing
complexity of data analysis methods needed to handle large amounts of “in the
wild” data. Techniques designed to simplify construct relationships using statisti-
cal inferences between distinct groups of measurements cannot address emerging
research questions which span the full spectrum of subject demographics, situ-
ational context, and time-scale. Contemporary approaches apply data mining
and machine learning techniques to fit more advanced models to study data and
identify key factors, but findings revealed in these exercises can be difficult to
generalize and interpret. By using a model as the hypothesis of an experiment
rather than focusing solely on a particular relationship between two variables in
specific conditions, research findings can be generalized more easily to practical
persuasive applications. Methods for evaluating models, rather than evaluat-
ing correlation between two variables should be increasingly focused upon in
the analysis of behavioral data. While analysis of correlation between variables
looks at the statistical relationship between groups of data points, the evalua-
tion of a model involves comparing the experimental data to the predictions of
the model. CHBMs can be used with contextual data to produce a time series
of expected behavioral outcomes throughout the study. The simulated “theo-

retical data” can then be directly compared to the “observed data” to observe
how the theory differs from the reality. The process of comparing theoretical
predictions to empirical data can be repeated with simulations from alternate
theories and a goodness-of-fit metric can be used to evaluate the hypothesis
against alternatives. Additionally, unification of existing behavioral models into
this common paradigm would enable better collaboration between proponents
of different theories.
Benefits of CHBMs Post-Experiment:
1. Analysis of experimental data can shift focus from individual construct rela-
tionships to a larger view, evaluating the model as a hypothesis.
2. Comparison between different theories can be informed by a comparison of
their respective models using a goodness-of-fit metric against empirical data.
3. The use of CHBMs makes re-use of theory and therefore collaborative improve-
ment on existing theories easier — reversing the existing paradigm which has
lead to a dizzying multitude of fragmented theories and sub-theories.
Open Questions for CHBM Post-Experiment Methods:
1. Methods for fitting a model to experimental data require restrictions on the

functional form of the relationships between variables, and the optimum func-
tional form is not yet obvious.
2. Methods for evaluating the goodness-of-fit between empirical and simulated
data exist, but cutting-edge software for exploring the intricacies of data
mismatch may be difficult to apply to this use-case.
5 Conclusion
In this paper we have offered supporting terminology, the CHBM formalization,

and a set of open challenges to promote the interdisciplinary discussion needed
to push forward the emerging field of JiTAI engineering. The progression of
behavioral science towards computational modeling has progressed more slowly
than in other scientific domains because of the limited amount of detailed, time-
intensive contextual and behavioral measures available. This progression from
causal descriptive modeling to causal explanitory modeling and increased math-
ematical rigor is a natural progression which parallels historical trends in the
natural sciences. Now that behavioral and contextual data is becoming accessi-
ble, we should expect to see a similar paradigm shift in the behavioral sciences.
It is our hope that this formative work towards Computational Human Behavior
Modeling and the methods highlighted here act as a jumping-off point for others
on the forefront of this impending paradigm shift who can use these methods to
unlock the power of context-aware persuasive application driven by CHBMs.
References
1. Ajzen, I.: The theory of planned behavior. Organ. Behav. Hum. Decis. Process.
50(2), 179–211 (1991)
2. Beck, C., McSweeney, J.C., Richards, K.C., Roberson, P.K., Tsai, P.F., Souder, E.:
Challenges in tailored intervention research. Nurs. outlook 58(2), 104–110 (2010)
3. Brailsford, S.C., Desai, S.M., Viana, J.: Towards the holy grail: combining system
dynamics and discrete-event simulation in healthcare. In: Proceedings of the 2010
Winter Simulation Conference (WSC), pp. 2293–2303. IEEE (2010)
4. Collins, L.M., Murphy, S.A., Bierman, K.L.: A conceptual framework for adaptive
preventive interventions. Prev. Sci. 5(3), 185–196 (2004)
5. Dallery, J., Raiff, B.R.: Optimizing behavioral health interventions with single-case
designs: from development to dissemination. Transl. Behav. Med. 4(3), 290–303
(2014)
6. Deshpande, S., Nandola, N.N., Rivera, D.E., Younger, J.W.: Optimized treatment
of fibromyalgia using system identification and hybrid model predictive control.
Control Eng. Pract. 33, 161–173 (2014)
7. Dong, Y., Rivera, D.E., Thomas, D.M., Navarro-Barrientos, J.E., Downs, D.S.,
Savage, J.S., Collins, L.M.: A dynamical systems model for improving gesta-
tional weight gain behavioral interventions. In: 2012 American Control Conference
(ACC), pp. 4059–4064. IEEE (2012)
8. Dong, Y., Rivera, D.E., Downs, D.S., Savage, J.S., Thomas, D.M., Collins, L.M.:
Hybrid model predictive control for optimizing gestational weight gain behavioral
interventions. In: 2013 American Control Conference (ACC), pp. 1970–1975. IEEE
(2013)
9. Hekler, E.B., Klasnja, P., Traver, V., Hendriks, M.: Realizing effective behavioral
management of health: the metamorphosis of behavioral science methods. IEEE
Pulse 4(5), 29–34 (2013)
10. Kaptein, M.C.: Formalizing customization in persuasive technologies. In: Mac-
Tavish, T., Basapur, S. (eds.) PERSUASIVE 2015. LNCS, vol. 9072, pp. 27–38.
11. Nahum-Shani, I., Hekler, E.B., Spruijt-Metz, D.: Building health behavior mod-
els to guide the development of just-in-time adaptive interventions: a pragmatic
framework. Health Psychology (2015)
12. Nahum-Shani, I., Smith, S.N., Tewari, A., Witkiewitz, K., Collins, L.M., Spring, B.,
Murphy, S.: Just in time adaptive interventions (jitais): an organizing framework
for ongoing health behavior support. Methodology Center Technical report, pp.
14–126 (2014)
13. Prestwich, A., Sniehotta, F.F., Whittington, C., Dombrowski, S.U., Rogers, L.,
Michie, S.: Does theory influence the effectiveness of health behavior interventions?
Meta-analysis. Health Psychol. 33(5), 465 (2014)
14. Riley, W.T., Rivera, D.E., Atienza, A.A., Nilsen, W., Allison, S.M., Mermelstein,
R.: Health behavior models in the age of mobile interventions: are our theories up
to the task? Transl. Behav. Med. 1(1), 53–71 (2011)
15. Shiffman, S., Stone, A.A., Hufford, M.R.: Ecological momentary assessment. Annu.
Rev. Clin. Psychol. 4, 1–32 (2008)
16. Timms, K.P., Rivera, D.E., Piper, M.E., Collins, L.M.: A hybrid model predictive
control strategy for optimizing a smoking cessation intervention. In: 2014 American
Control Conference (ACC), pp. 2389–2394. IEEE. (2014)
The Persuasive Potential Questionnaire (PPQ):
Challenges, Drawbacks, and Lessons Learned
Alexander Meschtscherjakov(&), Magdalena Gärtner,

Alexander Mirnig, Christina Rödel, and Manfred Tscheligi
Christian-Doppler-Laboratory “Contextual Interfaces”, Center for HCI,

Department of Computer Sciences, University of Salzburg, Salzburg, Austria
{alexander.meschtscherjakov,magdalena.gaertner,
alexander.mirnig,christina.roedel,
manfred.tscheligi}@sbg.ac.at
Abstract. Measuring the potential persuasive effect of non-fully functional

prototypes is important in a user-centered design process. A tool for measuring
this persuasive potential should be deployable regardless of the persuasive goal,
be suited for a generic context, and be targeted at different user groups. In this
paper, we make a first step towards such an all-encompassing, quick and
easy-to-use tool to measure the potential of persuasive systems: the Persuasive
Potential Questionnaire (PPQ). We outline the development stages of the PPQ.
A literature analysis led to five dimensions characterizing the persuasive
potential of a system. We then formulated 50 items for the PPQ in an iterative
generation process and conducted an online survey with 94 participants. Based
on a statistical analysis, we propose a first version of the PPQ with 3 dimensions
and 15 items. We conclude with a reflection on the identified benefits and
drawbacks regarding the current iteration of the PPQ.
Keywords: Methods Persuasion Questionnaire
1 Introduction
Measuring the long-term persuasive effect of a system designed to change behaviors

can be difficult. It is even more difficult to evaluate the persuasive potential of a
persuasive technology, which is not fully developed, and, thus, cannot be incorporated
into everyday activities over time. Yet, in early phases of a user-centered design
process it may be beneficial to assess the potentials of a proposed persuasive tech-
nology even though the system itself is not or only partially implemented at that point.
At the moment, no universally applicable questionnaire to measure both the per-
suasive potential of not (yet) applicable systems and the persuasive effect of already
applicable systems is available. The persuasive research community identified a need
for such universal tools, especially to be able to measure and compare the effects of
persuasive systems in a generalized way [11]. Of course, some general assessment tools
and questionnaires already exist within the community, but overall, these tools either
focus on specific strategies (e.g., [8]) or only have validity for one specific system in
one specific study context (e.g., [2, 6, 16]). The potential of persuasive systems is often

DOI: 10.1007/978-3-319-31510-2_14
The Persuasive Potential Questionnaire (PPQ) 163
determined simply by behavior observation, user self-reporting (e.g., [8]), self-

reflection (e.g., [14]), as well as counting (e.g., [3]). Lehto et al. [10] have presented a
theoretical model of factors affecting perceived persuasiveness of behavior change
support systems based on Oinas-Kukkonen and Harjumaa’s Persuasive Systems
Design Model (PSD) [12]. They presented a model predicting perceived persuasiveness
that may be utilized in various of domains.
If we want to be able to not only measure the potential effect of one system, but also
compare different systems regarding their persuasive effects (and their success or lack
thereof), a standardized means is necessary. Such a measurement tool should ideally be
able to assess persuasive systems regardless of the targeted user group, utilized per-
suasive strategies, or context in which it is deployed and the complexity of the system
itself. A persuasive system might work differently for different user groups, and there
exist many persuasive systems to foster the same persuasive goal but employ different
strategies, as has been shown (e.g., [9]). Also, the context in which the persuasive
system is embedded plays a major role for the success of the system. All of these
aspects need to be taken into account when measuring and comparing persuasive
systems on a general level.
The overall goal we had in mind was to develop a Persuasive Potential Ques-
tionnaire (PPQ) to measure the potential persuasive effect of a system, which is quickly
applicable, easy to use for researchers and system users, and open enough regarding the
questions it contains to be suitable for all kinds of systems. The PPQ should be:
Item Conservative: The PPQ should be relatively quick and easy to use. A high
number of items increase study and evaluation times exponentially. We, therefore,
aimed at keeping the item count as low as possible, with a number of 20 items as our
first concrete target.
Context Agnostic: Tools and methods that are tailored towards a specific context are
often ill-suited to be applied to any other than their originally intended context.
The PPQ, therefore, had to be context-agnostic, i.e., either be powerful enough to
measure the persuasive effects of a system regardless of the context it is used in or be
able to accommodate relevant context variables.
User Agnostic: Similar to the previous goal, the PPQ should not depend on or target
only one or a number of specific user groups. It needs to be usable either regardless of
intended users or be able to accommodate any number of different user groups and their
most relevant influence factors.
With these goals in mind, we set out to develop the PPQ in a multi-stage iterative
process. We started with an extensive literature analysis and utilized various accep-
tance, adoption, and persuasion models creating five dimensions characterizing the
persuasive potential of a system. In an interdisciplinary item construction workshop,
we developed 161 items for these five dimensions. Through a rating process, we
selected 50 items to be evaluated in an online questionnaire. The analysis led us to a
3-dimension questionnaire with 15 items. Figure 1 illustrates this process.
In the following we will outline the development stages that led to the present
iteration of the PPQ (Sect. 2), present the analysis and results of a first evaluation round
164 A. Meschtscherjakov et al.
Fig. 1. Development process of the PPQ.
(Sect. 3), discuss challenges, drawbacks, and lessons learned from the development
process (Sect. 4), and conclude with an overall summary (Sect. 5).
2 Development of the Questionnaire
2.1 Development of Five Dimensions

Starting point for the questionnaire development was the definition of different
dimensions important for a successful persuasive system. As a base, we used the
Technology Acceptance model (TAM) by Davis’ et al. [5]. This model describes how
users come to accept and use a specific technology. In a nutshell, it states that the
perceived usefulness of a system and its perceived ease-of use influence the attitude
towards using a system, which then again leads to a behavioral intention to use the
system, as well as its actual usage. We also used the Unified Theory of Acceptance and
Use of Technology (UTAUT) - an extension of the TAM, which states that various
factors such as the user’s experience and voluntariness, as well as external factors [17],
influence the user acceptance and willingness to use a system.
Additionally, our approach was inspired by Lehto and colleagues’ [10] structural
model for behavior change support systems. Their model consists of eight constructs:
primary task support, dialogue support, perceived credibility, design aesthetics, perceived
persuasiveness, unobtrusiveness, intention to use, and actual use. They also present
measurement items for each of the constructs.
Based on these models three researchers from our department (one psychologist,
one communication scientist, and one computer scientist), all researching in the area of
persuasive technologies, conducted several informal workshops, which led to five
dimensions possibly capable of measuring the persuasive potential of a system.
These five dimensions are:
1. Persuasion Attitude (PA)
2. Perceived Persuasive System Potential (PPSP)
3. Perceived Persuasive Individual Potential (PPIP)
4. Intention to Use the System (IUS), and
5. Intention to Change Behavior (ICB).
The PA dimension is intended to measure someone’s susceptibility to, and attitude

on, persuasive efforts in general. The assumption behind this dimension is that the level
of persuadability is different for each individual (see e.g., [4]). We are interested in the
general susceptibility towards persuasion and not necessarily towards specific per-
suasive strategies as highlighted by Kaptein [9]. The PPSP dimension is based on the
assumption that users of persuasive systems experience the persuasive strategies and
efforts a persuasive system is build upon and, therefore, can rate the persuasive
potential of the system factually from a user’s perspective. This is motivated by Petty
and Cacioppo’s Elaboration likelihood model (ELM) [13]. Based on cognitive pro-
cesses via the central route of persuasion, users may assign a general persuasive
potential to the system not necessarily suited for them. The PPIP dimension, on the
other hand, is intended to measure the degree to which participants assess the system’s
persuasive potential from their personal, subjective perspective. This is also reflected in
Lehto et al. [10] construct of “perceived persuasiveness”. The IUS dimension is defined
to measure how willing and motivated users are to keep using the persuasive system
from a short- and long-term perspective. This is a crucial factor in many acceptance
models (e.g., [5, 17]). Finally, The ICB dimension is intended to capture the effec-
tiveness of the persuasive system by capturing the users’ true intention to change their
behavior as an effect of using the system. This dimension is based on Ajzen’s [1] theory
of planned behavior. In this model, behavioral intention is an indication of an indi-
vidual’s readiness to perform a certain behavior. Table 1 summarizes the five dimen-
sions and sources used to create them.
Table 1. Five dimensions characterizing the persuasive potential of a system and sources we
used to create them.
Dimensions Sources
D1: Persuasion Attitude (PA) Busch et al., 2013 [4], Kaptein, 2012 [9], Kaptein
et al., 2011 [7]
D2: Perceived Persuasive System Petty & Cacioppo, 1986 [13]
Potential (PPSP)
D3: Perceived Persuasive Individual Petty & Cacioppo, 1986 [13], Lehto et al., 2012 [10]
Potential (PPIP)
D4: Intention to Use the System Davis et al., 1989 [5], Venkatesh et al., 2003 [17],
(IUS) Lehto et al., 2012 [10]
D5: Intention to Change Behavior Ajzen, 1991 [1], Lehto et al., 2012 [10]
(ICB)
2.2 Creation of 50 Items

In a second step, a workshop was conducted in which we made use of Method 635 [15],
a more structured variation of the brainstorming technique, to create the items for the
pre-defined dimensions. Method 635 was applied as follows: Six human-computer-
interaction experts with differing scientific background were instructed to develop items
for each of the five aforementioned dimensions, which were to form the basis for the
persuasive potential questionnaire (PPQ). To do so, participants were introduced to each
dimension and its meaning and then asked to write down three items, i.e. statements,
they thought would fit with regard to capturing the meaning of the dimension. After
doing so, the three items were passed on to the next participant, who read them and then
wrote down another three items. After five rounds of passing along the written notes,
item-collection was finished for one dimension. We continued this procedure for all five
dimensions. Then, the collected items were presented to all participants to trigger a
debate to reconsider, revise or remove the items.
Overall, we collected 149 items; 27 items in the PA dimension, 21 items in the
PPSP dimension, 42 items in the PPIP dimension, 24 items in the IUS dimension, and
35 items in the ICB dimension. The number of items was then further reduced based on
the individual rating of three experts in the field of persuasion theory, who evaluated
the contribution of every single item to capture the meaning of the corresponding
dimension. First, the three experts rated each item individually, then they conducted a
separate workshop in which they discussed their ratings, reformulated items were
necessary and reduced the number of items to 10 items per dimension, which led to an
overall number of 50 items. Finally, the dimensions were renamed, to capture the
associated items in a more self-explanatory way.
The PA dimension was renamed to Susceptibility to Persuasion (SP), the PPSP
dimension was renamed to General Persuasive Potential of the system (GPP), the PPIP
dimension was renamed to Individual Persuasive Potential of the system (IPP), and the
ICB dimension was renamed to System Influence on Behavioral Change (SIBC). The
IUS dimension remained the same. These tentative items and corresponding dimen-
sions are shown in Table 3 in the Appendix.
2.3 Questionnaire Creation

In order to evaluate the proposed items and to analyze whether items load to the
proposed factors, we conducted an online analysis. Items were ordered in a random
order and entered into the online questionnaire tool lime survey (www.limesurvey.at).
In order to enable study participants to answer the proposed items of the questionnaire
we had to describe a potential persuasive system in a fictive scenario.
The scenario introduced the “Healthy Me” application for smartphones (see Fig. 2).
Study participants were encouraged to imagine that this application would actively
support them in leading a healthier lifestyle. To facilitate the participants’ imagination
process, the scenario was written in a lively way, so every participant could easily
connect with the scenario’s content. For example, the “Healthy Me” application was
not limited to supporting its users in losing weight, in case this was nothing the
participants were interested in, but it was presented as a universally applicable tool,
which could support its user in the improvement of every life circumstances, he or she
wished for (e.g., weight reduction, quitting smoking, doing more sports, balanced, and
healthier nutrition, etc.).
Fig. 2. The picture illustrates the “Healthy Me” Application, the fictional persuasive system,
which the scenario of the questionnaire was based on.
Participants were then provided with a concrete scenario. “Imagine you want to do
more sports to increase your personal fitness level. The “Healthy Me” App is sup-
porting you to reach this goal. For example, it suggests to go jogging two to three
times a week with a maximum pulse of 150 heartbeats a minute. Additionally, “Healthy
Me” recommends to do some strength exercises to buildup some muscles. As you are
especially fond of training the upper parts of your body, the app suggests some
exercises with dumbbells and floor exercises. After each training session, you record
data like duration of training, intensity of training, and frequency of repetitions with
the application. “Healthy Me” processes this data to analyze your training program
and adapt it to your personal goals. A curve diagram informs you about your progress
and steps back over time. Progress is rewarded with cheering up like “You’re pretty
quick, you did the 5 km in less than 45 min today!” If you are losing sight of your goal,
e.g., stopped training for some time, “Healthy Me” encourages you not to give up and
defines new interim goals to increase your motivation again.”
After reading the scenario, participants had to rate the questionnaire items on a
7-point Likert scale from completely disagree to completely agree. A first pre-test with
five participants proved the questionnaire to be deployable, so the study was announced
through our universities and other email lists. Overall, 94 people participated in the
online study. Among participants, five €10 Amazon vouchers were raffled.
3 Analysis
When we developed the questionnaire, we had defined five dimensions, which we con-
sidered to be crucially contributing to the persuasive potential of a system. Yet, the actual
factorial structure of the dimensions was still to be examined. Further, our aim was the
reduction of the 50 existing items to provide a quick and easy-to-use measurement tool.
Therefore, we explored the underlying factors of our items by calculating a common-
factor analysis. From the results of measures of sampling adequacy, we derived an
excellent appropriateness of our data for factor analyses. The Kaiser-Meyer-Olkin
measure suggested that the sample was very well factorable (KMO = .907) and the
Bartlett’s test of sphericity indicated that we could reject the null hypothesis of an identity
matrix (χ2(1225) = 4525.22, p < .001).
Initially, we conducted a common-factor-analysis without presetting a certain
number of factors to be extracted. In the settings, we suppressed those factor loadings
smaller than .3 and used varimax rotations of the factor-loading matrix. Based on the
Kaiser-Guttman-criterion, an 8-factor solution was suggested with eight factors having
eigen values above one. In sum, those 8 factors explained 73.85 % of the total variance.
The communalities of our items were all above .3, which, besides the high amount of
explained total variance, confirmed the factorability of them.
Nevertheless, we noted different problems in this factorial structure. First, the scree
plot indicated a maximum number of 3–4 factors. Second, most items loaded on the
first four factors, whereas the other factors consisted of only a few items. Third, several
items did not show clear primary factor loadings on one factor, indicating a weak and
ambiguous definition. Analyzing these problematic items, we detected that almost all of
them belonged to our predefined SP dimension. Considering these inconsistencies, we
examined a three, four, and five factors solution using again varimax rotations of the
factor-loading matrix. However, the explained total variance consequently decreased to
59,65 %, 63,66 % and 66,97 %. These solutions resulted in even more ambiguous
factor loadings of several items of the susceptibility dimension. We concluded that our
original dimension did not represent one personality factor but consisted of many
different dimensions, which themselves built separated factors.
Before excluding problematic and redundant items in a next step and conducting
factor analyses again, we wanted to investigate the consistency of our original
dimensions. We considered this to be an additional decision aid for the following
selection of items. Therefore, we conducted several item analyses. Since our ques-
tionnaire does not represent a power test, we tested the items only on their internal
consistency and not on their difficulty or discrimination power, in order to determine
whether the items measured one dimension. Using reliability analysis, we obtained a
Cronbach’s Alpha of .73 – .95 across the five dimensions, which indicates a very high
homogeneity for each dimension. However, the reliability of the SP dimension turned
out to be lowest in comparison to the others, probably being the consequence of some
very low item correlations with the dimension.
We excluded items stepwise, which showed the lowest item-dimension correla-
tions. This resulted in a dimension of four items and an increase of Cronbach’s Alpha
of the SP dimension from .73 to .80. The item reduction across the remaining
dimensions was based on content decisions on one hand and statistical decisions
derived from the previous factor analyses on the other. Thus, we selected those items of
which were, among the others, best verbalized and loaded clearly and highly on one
factor. Finally, we again conducted a common factor analysis with the selected items.
The loadings of these remaining 15 items showed clear simple-structure solutions with
substantial loadings above .5 on one primary factor and negligible secondary loadings
on the two other factors.
The first factor includes eight items, which were assigned beforehand to three
different dimensions, namely the SIBC, the IUS, and the IPP dimension. Due to the
factor analysis, SIBC and IUS dimensions are now summarized under the Individual
Persuasive Potential of the system (IPP) dimension, meaning the persuasive impact the
system has on the single user. The second factor includes three items and is summa-
rized by the General Persuasive Potential of the system (GPP) dimension, meaning the
potentially persuasive impact of the system on any user. Finally, the third factor
includes four items and is summarized by the Susceptibility to Persuasion by others
(SP) dimension. This factor represents one personality dimension of the user, which
reveals if he/she is easily or hardly influenced by other people which as a consequence
may allow to draw conclusions on whether he/she is also easily or hardly persuaded by
a system. To prove or falsify this correlation still remains an objective for future
iterations of the PPQ questionnaire. Table 2 shows the iterated version of the PPQ.
Table 2. Iterated PPQ questionnaire consisting of 15 items and the three dimensions
Susceptibility to Persuasion (SP), General Persuasive Potential of the System (GPP), Individual
Persuasive Potential of the System (IPP).
New Old Item
Dimension Dimension
SP_1 SP_1 What others say brings me to rethink my attitude towards
it.
SP_2 SP _2 I do not want to be influenced by others.
SP_3 SP _4 Even my friends have difficulties to influence me.
SP_4 SP _8 No one can tell me what to do.
GPP_1 GPP_1 The system makes people change their behavior.
GPP_2 GPP_2 The system has the potential to influence people.
GPP_3 GPP_5 The system gives the behavior of its users a new direction.
IPP_1 IPP_7 This system is exactly what I need to change my attitude.
IPP_2 IPP_8 Thanks to the system I reach my goals.
IPP_3 IUS_4 I will use this system as often as possible.
IPP_4 IUS_7 I think that I will also use such a system in the future.
IPP_5 IUS_10 I will use this system regularly.
IPP_6 SIBC_2 This system does not cause a change in behavior with me.
IPP_7 SIBC_6 This system causes me to do some things differently.
IPP_8 SIBC_10 With the help of the system, I will behave differently in the
future.
3.1 Results of Our Sample

Overall, 94 participants took part in our online-study (70 female, 24 male). They were
aged between 15 and 61 years (M = 30.05, SD = 10.60) with most of them (85) having
a secondary school qualification or even a university degree. Most of them have not
had experiences with a fitness application such as “Healthy Me” before. When ana-
lyzing the three dimensions of the PPQ, we found that people in our sample scored
rather above average with regard to their susceptibility to be persuaded by others
(M = 3.90, SD = 0.81). They stated a relatively high general persuasive potential of

“Healthy Me” (M = 4.48, SD = 1.28). However, the individual persuasive potential of
“Healthy Me” was considered to be only average. Thus, we assume that people can
distinguish between the general potential of a persuasive system and susceptible they
would be themselves. Further, they strengthen our pre-assumption that constructing
separate dimensions asking for the general potential of a system and for the actual effect
of the system on people’s own attitude and behavior is reasonable. In order to check for
requirements of the following statistical tests and to make a statement about the gen-
eralizability of our data we tested our sample on the assumption of normal distribution.
The Kolmogorov-Smirnov tests show that our data are normally distributed.
In a next step, we were interested in whether the individual level of persuadability
of people on our SP dimension influences their opinion of the general and individual
persuasive potential of the tested system. Therefore, we conducted t-tests for inde-
pendent samples and calculated group comparisons. We made a median split for the SP
dimension, which served as independent variable, thus comparing people scoring low
with people scoring high on that dimension. Dependent variables were the general
(GPP) and individual (IPP) persuasive potential of the system. We found a highly
significant difference concerning the individual persuasive potential of the system
between people who are sparsely susceptible of persuasion by others and people who
are highly susceptible to persuasion by others (t(92) = 4.11, p < .001). People, who are
sparsely susceptible to persuasion by others, show a higher individual persuasive
potential by the system than people who are highly susceptible to persuasion by others,
meaning that the system has more influence on people, who are less susceptible to
social persuasion. Although, this result might seem paradoxical at first sight, a logical
explanation for this effect can be that susceptibility of persuasion by others does not
necessarily include the same extent of susceptibility of persuasion by systems. Both
kinds of persuasion might rather be seen as competing constructs, which exclude each
other or are at least not activated at the same time to the same extent. As such, people
might either follow advice of other people, such as friends or they have more trust in
technological systems but not both.
In order to determine whether SP can even serve as a predictor for GPP and IPP, we
calculated linear regression analyses with SP as predictor and GPP and IPP as criterion
variables. Our results indicate that susceptibility of persuasion by others significantly
predicts the general (b = .54, t(92) = 6.20, p < .001) and individual persuasive potential
of a system, b = −.39, t(92) = −4.08, p < .001. SP also explained a significant
proportion of variance in GPP (R2 = .30, F(1,92) = 38.45, p < .001) and IPP (R2 = .15,
F(1,92) = 16.61, p < .001) scores. However, the explained proportion of variance,
especially concerning IPP, is comparatively small. This indicates that our
SP-dimension only covers a small range of factors that contribute to whether the
persuasive potential of a system is able to influence the single user or not. We further
suggest that even the user’s personality and its influence is, to such an extent, diver-
sified that such a narrow personality dimension like ours can impossibly capture the
dimensions which actually contribute to the persuasive potential of a system.
Developing a broad and extensive dimension including the different personality aspects
would, therefore, be of great relevance in a further iteration of the PPQ.
4 Discussion
This first iteration of the PPQ has confronted us with challenges, drawbacks, and
lessons learned. In the following we discuss concrete areas for improvement of the PPQ
and findings regarding the potential persuasive effect of a system in general.
4.1 It’s all About One’s Personality: Traits and Persuadability

The persuasive strength of a system strongly depends on how well the user responds to
the persuasive strategies employed. Not all strategies work equally well for everyone,
which is why persuasive strategies need to be chosen depending on the targeted
demographic and their expected personality spectrum as shown by Kaptein [9]. For the
PPQ, we tried to incorporate a person’s susceptibility to persuasive strategies in the
Susceptibility to Persuasion (SP) dimension. We are aware that this approach reduces
the complexity of personality traits in favor of a simple and short questionnaire. But as
it turned out, brevity is not quite such a virtue when it comes to assessing personalities.
Our personality dimension was too narrow and subsequently inadequate to capture all
or most relevant personality factors for the persuasive success of a system. We found
that a truly universal questionnaire requires a more broad and detailed personality
dimension, containing a much greater number of items than ours had. Of course, this is
contrary to the initial goal to keep the item count to a maximum of 20, but it is a
necessary modification if we want to achieve the goal of being user agnostic. One can
impossibly predict the whole range of potential users of a system, but what one can do
is providing a comprehensive personality spectrum, which ultimately serves the same
goal of clustering users into persuasion-relevant groups. The immediate next step in
this regard is, therefore, the development of a broad and extensive personality
dimension for the following iteration(s) of the PPQ.
4.2 Deployment and Validation of the PPQ

Regarding methodology, we have to mention that we used a fictive scenario in the
healthcare domain to evaluate the first version of our questionnaire in the online survey.
We deliberately formulated this scenario as general as possible. Nonetheless, the sce-
nario might have had an influence on the answers given and therefore the reliability of
the PPQ. The natural next step, after further iteration and refinements, is to deploy the
questionnaire to evaluate an actually existing system the users can really interact with,
but from the same domain. The PPQ is not, however, targeted at only healthcare-related
persuasive systems, but any system with a persuasive effect. Therefore, the PPQ also
needs to be evaluated in a wider number of different contexts in order to show whether it
produces adequate data in all of these contexts as well. This is a continuous process,
which will accompany the PPQ during its next development and iteration stages. It also
is strongly dependent on the community adapting, employing, and criticizing the PPQ in
different contexts and with different systems.
4.3 Human- vs. System-Induced Persuasion

Common sense would suggest that a person is either more susceptible to persuasion in
general or not. By delineating Individual Persuasive Potential (IPP) from General
Persuasive Potential (GPP) and assessing both individually, we found an interesting
difference regarding the IPP of a system that contradicts that initial intuition. It appears
that people who are highly susceptible to persuasion by others show less persuasive
potential by the system than people who are sparsely susceptible to persuasion. This
means that for people who are less likely to be influenced by other humans, the better a
persuasive technology should work for them. Accordingly, persuasive system strategies
cannot mimic human persuasion strategies on-to-one and hope for success, nor can they
be measured on the same level. Both system and human persuasion appear to operate
on different levels, with different activators and factors that determine success or
failure. It is an interesting insight on its own and also brings to light – by means of
practical example – the value of a universal assessment tool, which can accommodate
such system- and context-independent findings. Nonetheless, these considerations need
to be explored thoroughly in the future.
5 Conclusion
In this paper, we presented the development process and first evaluation round of the
PPQ, a universal tool to analyze and measure the persuasive potential of a system. The
first iteration presented in this paper is already capable of producing usable results but
still requires further testing in different contexts, with different user bases and a wide
range of systems. So far the PPQ was only applied in one context. Above that, there are
still issues that have to be targeted in further iteration such as the effect of the com-
plexity of the system on the persuasive effect, and how we can predict unconscious
persuasion [13].
As a next step it would be interesting to compare the results of the PPQ with the
actual persuasive effect of a system in order validate that the PPQ actually predicts the
persuasive potential. The PPQ will continue to undergo development and iteration, with
an appropriate personality and context dimension as the immediate next step. The
Persuasion community is encouraged to put the PPQ to the test for different systems, to
help speed along this process.

Research and Economy and the National Foundation for Research, Technology and Develop-
ment is gratefully acknowledged (Christian Doppler Laboratory for “Contextual Interfaces”).
Appendix
See Table 3
Table 3. The 50 items generated after the workshop. Each of the 5 dimensions is represent-ed
with 10 items: Susceptibility to Persuasion (SP), General Persuasive Potential of the System
(GPP), Individual Persuasive Potential of the System (IPP), Intention to Use the System (IUS),
System Influence on Behavioral Change (SIBC).
Dimension Item
SP1 What others say brings me to rethink my attitude towards it.
SP2 I do not want to be influenced by others.
SP3 I’m easily influenced.
SP4 Even my friends have difficulties to influence me.
SP5 I often change my attitude.
SP6 I am open to change.
SP7 I find it difficult to change.
SP8 No one can tell me what to do.
SP9 Changes are part of my everyday life
SP10 I have my principles, which are unchangeable.
GPP_1 The system makes people change their behavior.
GPP_2 The system has the potential to influence people.
GPP_3 The system can not change the attitudes of people.
GPP_4 The system is useless for changing my attitudes.
GPP_5 The system gives the behavior of its users a new direction.
GPP_6 The system has almost no influence on behavior.
GPP_7 If you want to change your own behavior, this system is useful.
GPP_8 This system highly encourages a change in behavior.
GPP_9 This system provides little support for a change in behavior.
GPP_10 Few people will behave differently when using the system.
IPP_1 I find it difficult to change my attitude with the help of this system.
IPP_2 My attitude is not affected by the system.
IPP_3 The system has no potential to change my attitude.
IPP_4 This system helps me to change myself.
IPP_5 I think that this system is well suited to make long-term changes to my attitude.
IPP_6 This system shows me the information that I need to change myself.
IPP_7 This system is exactly what I need to change my attitude.
IPP_8 Thanks to the system I reach my goals.
IPP_9 Even if I use the system, I do not change my attitude.
IPP_10 Without the system, I would not change my attitude.
IUS_1 I have no motivation to use this system.
IUS_2 I’ll continue to use this system.
IUS_3 I’ll never use this system.
(Continued)
Dimension Item
IUS_4 I will use this system as often as possible.
IUS_5 I see no point in using such a system again.
IUS_6 Even if I have this system, I will not use it.
IUS_7 I think that I will also use such a system in the future.
IUS_8 I will not utilize this system, although it could help me.
IUS_9 I hope I can possess and use this system.
IUS_10 I will use this system regularly.
SIBC_1 This system persuades me to change my behavior.
SIBC_2 This system does not cause a change in behavior with me.
SIBC_3 I will change my behavior because of this system.
SIBC_4 I think that I will have sustainable behavior change through use of this system.
SIBC_5 This system does not convince me to actively change my personal behavior.
SIBC_6 This system causes me to do some things differently.
SIBC_7 By using this system, I will not change my behavior.
SIBC_8 The functions of this system do not convince me to change my behavior.
SIBC_9 The system is designed so that I would like to change my behavior.
SIBC_10 With the help of the system, I will behave differently in the future.
References
1. Ajzen, I.: The theory of planned behavior. Organ. Behav. Hum. Decis. Process. 50(2), 179–
211 (1991)
2. Anderson, C., et al.: The persuasive power of oral health promotion messages: a theory of
planned behavior approach to dental checkups among young adults. Health Commun. 28(3),
304–313 (2013)
3. Bleich, S.N., et al.: Reduction in purchases of sugar-sweetened beverages among
low-income black adolescents after exposure to caloric information. Am. J. Public Health
102(2), 329–335 (2012)
4. Busch, M., Schrammel, J., Tscheligi, M.: Personalized persuasive technology – development
and validation of scales for measuring persuadability. In: Berkovsky, S., Freyne, J. (eds.)
5. Davis, F.D., Bagozzi, R.P., Warshaw, P.R.: User acceptance of computer technology: a
comparison of two theoretical models. Manage. Sci. 35, 982–1003 (1989)
6. Hendriks, B., et al.: Style congruency and persuasion: a cross-cultural study into the
influence of differences in style dimensions on the persuasiveness of business newsletters in
Great Britain and the Netherlands. IEEE Trans. Prof. Commun. 55(2), 122–141 (2012)
7. Kaptein, M., et al.: Means based adaptive persuasive systems. In: Proceedings of the
SIGCHI Conference on Human Factors in Computing Systems, pp. 335–344. ACM (2011)
8. Kaptein, M., et al.: Adaptive persuasive systems: a study of tailored persuasive text
messages to reduce snacking. ACM Trans. Interact. Intell. Syst. 2(2), 10 (2012)
9. Kaptein, M.: Personalized persuasion in ambient intelligence. J. Ambient Intell. Smart
Environ. 4(3), 279–280 (2012)
10. Lehto, T., Oinas-Kukkonen, H., Drozd, F.: Factors affecting perceived persuasiveness of a
behavior change support system. In: Proceedings of the 33rd International Conference on
Information Systems (2012)
11. Meschtscherjakov, A., et al.: Workshop on persuasive technologies in challenging contexts.
In: PT 2014: The 9th International Conference on Persuasive Technology (2014)
12. Oinas-Kukkonen, H., Harjumaa, M.: Persuasive systems design: Key issues, process model,
13. Petty, R.E., Cacioppo, J.T.: The elaboration likelihood model of persuasion, pp. 1–24.
Springer, New York (1986)
14. Ramachandran, D., et al.: Mobile-izing health workers in rural India. In: Proceedings of the
SIGCHI Conference on Human Factors in Computing Systems. ACM (2010)
15. Rohrbach, B.: Creative by rules—method 635, a new technique for solving problems.
Absatzwirtschaft 12, 73–75 (1969)
16. Thieme, A., et al.: We’ve bin watching you: designing for reflection and social persuasion to
promote sustainable lifestyles. In: Proceedings of the SIGCHI Conference on Human Factors
in Computing Systems. ACM (2012)
17. Venkatesh, V., Morris, M. G., Davis, G.B., and Davis, F.D.: User acceptance of information
technology: toward a unified view. MIS quarterly, pp. 425–478 (2003)
Persuasive Practices: Learning from Home
Security Advisory Services
Mateusz Dolata1(&), Tino Comes1, Birgit Schenk2,

and Gerhard Schwabe1
1
Department of Informatics, University of Zurich, Zurich, Switzerland
{dolata,comes,schwabe}@ifi.uzh.ch
2
University of Applied Science Ludwigsburg, Ludwigsburg, Germany
[email protected]
Abstract. Research on persuasive technologies (PT) focuses, primarily, on the

design and development of IT for inducing change of individual’s behavior and
attitude through computer-human and computer-mediated influence. The issue of
practices in co-located human-human persuasive encounters remained unat-
tended in the PT community. This study uses the notion of persuasive practices to
understand the course of events in face-to-face home security advisory sessions –
it specifies and illustrates such practices and discusses their impact on the per-
suasiveness of the encounter. Furthermore, it presents potential of IT to support
such persuasive practices thus opening new research possibilities of PT research.
Keywords: Advisory encounter Human-Human influence Practices F2F
1 Introduction
Persuasive technologies (PT) are engineered to reinforce, change or shape behaviors and
attitudes of individuals towards specific areas of their life [21]. Technology can either
directly influence an individual’s behavior or, alternatively, act as mediator or moderator
of social influence [26], i.e., it transfers information on other’s opinions or behaviors. In
this study, we explore yet another role that technology plays in the context of persua-
sion: it can facilitate persuasive practices in situations where human influence is
exhibited in a face-to-face encounter. There exists a category of such encounters where
considering solely the technology – a view that has dominated in PT community so far –
does not resemble the complexity of conducted activities and their effect on the per-
suasive effect. We postulate, that, in such situations, the IT should not be considered as a
standalone factor in the success or failure of persuasion efforts – it is not a machine that
produces persuasiveness. Instead, it becomes a tool, which – if embraced in specific
practices – can be very effective and support the change of persuadee’s attitude and
behaviors [3]. Such perspective on PT helps, in particular, in high-touch situations, i.e.,
where direct influence between humans through practices comes to the fore. Practices
are seeable, indigenous actions that participants directly engage in, but do not attend to
them in an analytic manner [7, 18]. Still, practices can be object of systematic analytic
approach in research [18] – their consideration in the area of PT is a response to the
recent call for the practice-turn in human-computer interaction [13].
DOI: 10.1007/978-3-319-31510-2_15
Persuasive Practices: Learning from Home Security Advisory Services 177
In the current study, we focus on the case of home security (HS) advisory services.
It is an encounter between (1) a help-seeking homeowner, who wants to make their
property more secure (persuadee), and (2) a professional HS advisor (persuader). Even
though persuadees often see the need for improving their home’s security, they lack
ability and motivation to tackle those issues. HS advisory service shall make it easier
for them to reach their goal: its goal is to identify most important flaws and pave the
way for improvement through mechanical upgrades and security-aware behavior.
However, according to a preliminary study, only 20 % of improvements suggested
during the advisory sessions are implemented by the homeowners [24]. Given the
reportedly successful application of PT in other difficult areas, e.g., preventive
healthcare, we propose to include its basic principles in a HS-dedicated socio-technical
persuasive system, which we define as an information system designed to reinforce,
change or shape persuadee’s attitudes or behaviors [20]. It consists of a human per-
suader who engages in persuasive practices with use of his tools such as: IT, brochures,
notebooks, and exhibits, as well as objects in their surroundings, e.g., windows and
doors at persuadee’s home. We understand persuasive practices as practices that exhibit
the desire to influence the behavior or attitude of conversation partner. We subscribe to
very local and timely-limited notion of practices like the one used in conversation
analysis [10] or multimodal analysis of encounters [12]. Such practices, normally,
involve use of tools and artifacts, which, in turn, shape the practices – the materialistic
and social perspective intertwine and form a socio-material view [27]. So far, the
socio-material character of IT in persuasive encounters did not attract much attention in
research. In particular, the relation between design of PT as a collaborative system and
the course of events in persuasive encounters remains unclear. The current study
addresses this gap while taking an exploratory mixed-method approach.
To frame the exploration we formulate our research questions as follows:
RQ 1: What persuasive practices emerge in persuasive encounters of HS advisors?
RQ 2: How can we support the persuasive practices of HS advisors by means of IT?
With these research questions in mind, we aim at presenting persuasive practices
employed by the advisor with and without an IT tool designed along the basic guidance
originating from the field of PT, and we want to show how the application of such
IT-enhanced practices improve the persuasive character of the encounters.
2 Related Work
The scenario we address in our research clearly relies on interpersonal, face-to-face

persuasion. This mode of influence so far remains outside the core focus of the dis-
course on PT. We propose to supplement the traditional conversation-based encounter
with IT designed, explicitly, to support the persuasion efforts. While it differs from the
core PT literature, reconsideration of it informs the design of the proposed tool.
Interpersonal Influence and Persuasive Technologies: Persuasion, being defined as
“human communication designed to influence the autonomous judgments and actions
of others” [25] as well as “a successful intentional effort at influencing another’s mental
178 M. Dolata et al.
state through communication in a circumstance in which the persuadee has some

measure of freedom” [22], relies on interaction between two actors. Conventionally, the
persuader provides arguments [22] and appeals to the deep human drives of the per-
suadee [4]. The persuadee does or does not undergo changes in their attitudes and
behaviors with regard to a topic [4]. This highly interactive nature of persuasion finds
acknowledgement in the area of PT where technology is postulated to substitute human
and establish a computer-human persuasion scenario [26].
Conventionally, two perspective emerged within PT research of how technology
influences the behavior and attitude of an individual: (1) computer-human influence and
(2) computer-mediated human-human influence [20]. Persuasive systems belonging to
the former category rely on the assumption that technology can act as social agent and,
thus, impact the behavior of an individual [26]. In cases where technology acts as
mediator (e.g., blogs, forums, and social networks), the individual’s behavior or attitude
underlies social influence in form of user-generated content mediated by dedicated
technology [8, 26]. Recently, a novel technology-dependent mode of persuasion was
proposed: computer-moderated influence [26]. Systems belonging to that category
transfer information on the behavior of others and influence an individual’s behavior or
attitude by promoting behavior-based and not content-based social influence [26].
According to the seminal paper of Fogg [6], persuasion is likely to be successful
when three interrelated factors are addressed: motivation related to the feeling of dis-
comfort and rejection of current state, ability describing how simple or difficult it is to
reach the target behavior, and trigger being a signal, facilitator, or spark that tells
people to perform the change at a particular moment [6]. Those factors were developed
in the context of computer-human influence, but were applied as design guidelines for
computer-mediated influence too [17]. We argue that software developed along those
lines, will contribute to the emergence of relevant persuasive practices and thus support
persuasion in face-to-face persuasive encounters.
Persuasive Practices: The topic of persuasive practices – as longitudinal gradual
changes – was addressed in relation to ubiquitous systems informing users about their
behaviors and bringing about change in their attitudes and behaviors [23]. Such sys-
tems do well in scenarios with clear goals and clear ways to reach them, such as in the
case of WaterBot [1] where the information on used water motivates the user to reduce
water wasting or in apps inducing change in the life style [23]. However, in the
scenario of HS, the ways of improving things are not straight forward and require
involvement of a human actor who can establish understanding for security issues and
related topics.
Persuasive practices employed in the traditional service encounters, relying on
interpersonal influence, aim at securing the attention of the persuadee and at ensuring the
right pre-condition for transferring the message [2]. As discussed in consulting litera-
ture, this involves directly addressing the persuadee, posing questions and provocative
statements, as well as using various encodings [2]. We expect that an IT tool equipped
with dedicated features will impact the way the advisor engages in such practices, thus
changing the general impression on the persuasiveness of the HS advisory service.
3 Methodology
Preliminary Studies: This study is a part of a research program on burglary-prevention

conducted in collaboration with the responsible authorities, i.e., state police departments
from Germany and Switzerland. The preliminary studies focused on shadowing the
advisor at real HS advisory sessions, interviews with the involved stakeholders, and
formative tests of the proposed technology. It enabled us to better understand how the
advisors see their main task and how they behave during advisory sessions. It pointed to
persuasion as a central issue in this context.
Technology Design: The tool was designed in a user-centered process according to
the requirements collected from stakeholders: advisors requested access to materials
they know (schemata, pictures) and wanted an easy-to-carry device; homeowners
requested a better understanding of the complex information and more personalization;
and authorities requested more standardization of the advisory service. The tool shall
improve the persuasive character of the encounter to make the persuadee tackle the
security issues. The design was inspired by the persuasion model by Fogg [6]: The tool
offers a list of standard needs and fears to give the advisor a possibility to address the
emotions and appeal to persuadee’s motivation (cf. Fig. 1a). It offers multiple ways of
visualizing important technical and behavioral information to address persuadee’s
ability (cf. Fig. 1c and d) [5]. A prioritization tool and means to email a PDF with the
priorities to the persuadee establishes a trigger (cf. Fig. 1b) [5]. We use the tool as a
vehicle to observe emerging practices and compare them to the non-IT condition.
Fig. 1. Exemplary screens included in the HS advisory service support tool (a) Homeowner’s
needs, (b) Prioritization, (c) Photo taking and annotation, (d) Schemata.
Data Collection: The setting we chose for data collection enables for a natural
interaction in a realistic setting. Collecting data from real advisory sessions is nearly
impossible because homeowners are reluctant to agree on recordings of their private
properties and, in particular, the security flaws thereof. Identification of low-level

interactional and conversational practices requires extensive and possibly multimodal
data set [10]. We therefore conducted a design experiments [16], thus presuming that
introduction of an IT has impact on conversation practices and aiming at their
identification.
The overall experiment followed the within-subject design with two conditions: IT
and non-IT. It was scenario-driven: each test person was asked to put oneself in a
position of a homebuyer who visits two different houses with a home security advisor
to receive advice on how to make their future property more secure. No advisor saw the
same property twice. We compensate for the order effects while alternating the con-
ditions order. Overall, 20 persuadees and 10 advisors participated in the experiment.
Whereas the advisors were policemen who conduct HS encounters on daily basis, the
homeowners were a convenience sample acquired through different channels including
social media – their age, status, and gender varied, but all of them knew the feelings
around buying a new house – they were in a similar situation before. They were not
paid for their participation – they were doing it out of interest and received inexpensive
gifts after the test. The test was conducted on five days in March 2015 in Mannheim
and Frankfurt, Germany, at a pre-fabricated houses fairs. Before the experiment, each
advisor participated in a day-long training on the features of the tool and could try it out
in role-play exercises. The trainings took part in the same week as the experiments.
Additionally, brush-up trainings were conducted on the day of the experiment.
After going through two advisory sessions (IT and non-IT), each persuadee attended
a survey and a semi-structured interview built around the topic of interaction quality
(e.g., mutual understanding) and persuasion (e.g., emotional response, motivation,
ability). In the survey, each participant answered questions related to perceived per-
suasiveness (PERS) and design aesthetics (DESA) of the encounter on a five-point Likert
scale adapted to reflect the HS advice [15]. Overall PERS score were computed based on
answers to the such questions as: (1) the advisory session (AS) influenced me, (2) the AS
was tailored to me personally, (3) the AS makes me rethink my security behavior. DESA
score uses on the following: (1) the AS uses attractive tools, (2) the general appearance of
the AS is appealing, (3) the AS provides nice visual experiences [15].
Data Analysis: The collected data in form of video and audio recordings was analyzed
by an experienced researcher. First, the audio recordings were transcribed. Second,
annotations regarding the ongoing actions of advisor and the persuadee were attached.
Third, a portion of data (10 random samples of fifteen minutes from various recordings)
were taken to identify criteria for interesting episodes – all episodes considered relevant
(i.e., including persuasive practices) by at least two researchers either of the following
occurred: directly addressing the persuadee with questions including second-person
personal pronoun (Engl. “you”, Germ. “Sie”), directly addressing the persuadee
through directive speech, or directly addressing the persuadee with statements
including modal auxiliary verbs (Engl. “you may…”, Germ. “Sie mögen…”). Fourth,
all episodes extracted from the videos based on the criteria (446 episodes) were
clustered based on their similarity in an interactive session involving two researchers.
Fourth, abstract descriptions of the clusters were generated based on the transcripts and
annotations and put into context of the ongoing action. Moreover, passages from the
interviews that related to the particular identified practices were extracted.
Additionally, we applied a mix of statistical tests to identify relation between DESA
or PERS values and observed practices. In those tests we treated the observed practices
as independent variables with value 1 if a practice was applied in an advisory session
and 0 otherwise. We identified practices, that have influence on the dependent vari-
ables: DESA and PERS. We run our tests separately for the IT and non-IT conditions –
the reported results can be treated as in a between subject study. Following the above,
we chose a set of statistical tests appropriate for identification of dependences between
nominal independent variables with two or more levels and continuous (interval)
dependent variables [11, 14]. We employed the following tests: two-independent-
sample t-test, Kruskal-Wallis test, and multiple regression. This approach allowed for
formulation of relevant hypotheses and should be treated as indication of possible
directions for future research, but not as ultimate evidence for the reported influence or
its direction. In particular, our experiment was not designed to detect them and the fact
that some of the practices were or were not applied in the particular sessions is more a
matter of chance than a consequence of deliberate experiment design. Here, we report
on observations with significance coefficient lower or equal to 0.1 (designated by p).
4 Results
In the following, we report on the observations we made across the very extensive data
set obtained in the described experiment. First, we show the practices employed by the
advisors in the course of their persuasive activities. In particular, we point to the
specific practices that are made available by the proposed technology, show how they
fit the advisory session, and compare them to corresponding practices in the non-IT
condition. Second, we discuss the collected opinions and statistical data that indicate
what practices are beneficial or destructive to the overall persuasive character of the
encounter.
4.1 Identifying Persuasive Practices in Home Security Advisory

Encounters
Thanks to the very extensive preliminary studies in the context of HS advisory services,
we could identify three particular goals that advisors follow in their daily work: First,
they want to offer help that fits homeowner’s needs and situation. Second, they want to
provide convincing explanations regarding complicated technical features and behav-
ioral issues. Third, they want to bring the persuadee to tackle the discussed issues –
they should contact, e.g., local providers of HS hardware, or change their behaviors.
During the analysis of the data, we were able to identify practices in each of the three
areas – in the following, we shortly characterize them and provide information on their
occurrence across our data set (Tables 1, 2, and 3).
Current Situation and Needs of the Homeowner: According to its definition, per-
suasion aims at changing, reinforcing or shaping new behaviors and attitudes. This may
be ineffective, if the current situation or needs of the persuadee remain hidden. Con-
sequently, that advisors try to approach this topic. If one considers the model proposed
by Fogg [6], the work that advisors do while learning to know the homeowner falls into
the area of motivation – discussing the needs and feelings regarding security makes
clear to the persuadee why the encounter takes place and why HS is important.
Table 1. Persuasive practices related to learning and addressing homeowner’s needs
Practice and its description

Practice 1 – asking: The advisors introduce the topic of HS mostly by simply asking for the reason of the
encounter – after a short introduction they simply pose a question. Often, the answer of the customer is not pro-
ceeded by any further discussion. Then, the advisor simply moves to the next point on their agenda, mostly re-
viewing the security features of doors. Practice 1 occurs with the following frequency:
(a) IT 5 times (out of 20) (b) non-IT 11 times (out of 20)
In 3 further non-IT cases, not even a single question was asked regarding the homeowner’s HS needs.
Practice 2 – discussing: After receiving an answer, the advisor continues on the topic while paraphrasing the
answer and, important, asking further questions to additional information instead of simply moving to the next
topic being mostly window’s or door’s hardware. Practice 2 occurs with the following frequency:
We speculate that this frequency results from the fact that the tool includes a screen (Figure 1a) where advisor can
choose from a set of standard needs to characterize the current situation of the homeowner. Even though the
suggestions in the tool are rather general, the accompanying discussions were more extensive than that.
Practice 3 – recording: While or after listening to the homeowner’s HS needs, the advisor takes notes of the
needs in the tool by choosing respective fields, but does not show to the client what he chooses or touches – the
advisor treats the tool as his private device. In the non-IT setting, the advisors did not make any effort to record
the information regarding the homeowner’s needs. Practice 3 occurs with the following frequency:
(a) IT 9 times (out of 20) (b) non-IT no comparable practice
Practice 4 – collaborative recording: As opposite to the previous situation, here the advisor takes care of involv-
ing the client into the recording. This happens by sharing the screen with them and by paraphrasing the needs
expressed by the client to fit the descriptions in the tool. Advisors leverage this situation to introduce the tool and
explain its role in the advisory session. Practice 4 occurs with the following frequency:
(a) IT 9 times (out of 20) (b) non-IT no comparable practice
Practice 5 – reviewing: The advisor gets back to a particular need or set of needs expressed earlier by the
persuadee. This often happens when he presents the final report of the encounter or, in fewer cases, when discuss-
es particular improvement of a security feature. Practice 5 occurs with the following frequency:
Missing Security Features: Persuasion is ineffective if the persuadee does not rec-
ognize the attitude or behavior she should change to reach her goals. In our particular
case, this includes assembling of new security elements for doors and windows, as well
as establishing new routines using those security elements (e.g., locking the windows
with a to-be-installed lock mechanism as opposed to simply closing them). The work
that the advisors do while teaching the new behaviors falls into the area of ability
according to Fogg’s model [6]. We observe a whole range of practices in this context.
Personalized Recommendation: Given the presented understanding of persuasion,
the practices presented above may be ineffective if the persuadee does not know how to
approach the set of proposed changes, i.e., what steps to take to reach her goal. In our
particular case, a persuadee needs to know what is obligatory and what is optional,
what can be done easily (e.g., on changing insecure behaviors), and what requires more
Table 2. Persuasive practices related to presenting security improvements and new routines

Practice 6 – illustrating through gesture: The advisors explain the flaws of windows and doors directly at the
object, while pointing to the particular features (e.g., lock mechanism) and explaining how it may be improved,
i.e., how it should look like and be used in the future. This is a practice that seems to be essential to all advisors
and is applied at least several times in the IT and non-IT condition.
Practice 7 – taking picture: The advisor takes picture of the particular object or feature via the camera included
in the tool. This practice has two different forms: (1) the advisor takes the picture alone while the homeowner
goes on the side; (2) the advisor tries to keep the homeowner involved by explaining what he does and maintain-
ing the conversation or by incorporating the persuadee in the process of photo taking (encouraging her to look at
the picture being taken). Practice 7 occurs with the following frequency:
(a) alone + IT 17 times (out of 20) (b) together + IT 3 times (out of 20)
Practice 8 – annotating picture: The advisor adds specific marking to the previously taken picture in form of
rough drawings done by touching the screen – such annotations depict particularly weak or strong points of the
object in the picture. Again, this practice is done only by the advisor or in collaborative manner with the
persuadee. It can only occur in IT condition. Practice 8 occurs with the following frequency:
Practice 9 – reviewing picture: Advisor reviews the picture and shows it to the homeowner to return to a particu-
lar topic from earlier part of the conversation. This typically happens when advisor recapitulates the advisory
session, returns to particular object or identifies appropriate solutions to address the issue discussed at the object.
This practice occurs only in IT condition; in 18 out of 20 cases.
Practice 10 – adding free text and notes: The advisor adds additional notes or chooses from predefined template
notes to denote the problem or the solution related to a particular object. This practice occurs in the IT condition,
but can be compared to Practice 14 from the non-IT case. Nevertheless, Practice 14 occurs in the late phase of the
advisory session, during recapitulation of particular problems and solution. Thanks to the IT, Practice 10 can
occur throughout the service provision – it occurs in 13 out of 20 cases.
Practice 11 – presenting a video: The advisor presents a video illustrating working methods of burglars and how
they deal with doors and windows. Advisors introduce the video shortly and then add further explanations or
clarifications to the presented material. This practice occurs in 14 out of 20 cases in IT condition.
Practice 12 – presenting a schema: The advisor presents a schema of a technical detail to the persuadee on the IT
tool. There are numerous schemas provided in there and they reflect material presented in brochures and other
printouts. This practice is the IT-based counterpart of Practice 13. It occurs in all 20 IT cases.
Practice 13 – presenting a brochure: The advisor presents a schema of a proposed solution to the persuadee in
the brochures and print outs he carries with him. The material includes mostly a technical drawing of the proposed
solution. This practice is the paper-based counterpart of Practice 12. It occurs in all 20 IT cases.
Practice 14 – annotating a brochure: The advisor adds additional notes and sketches to brochures and print outs.
The information he adds to the brochures includes, e.g., information on which window or door it belongs to or
what kind of materials can be used in the given context. This practice is the paper-based counterpart of Practice
11. It occurs in 10 out of 20 cases – four less than in the IT condition.
Practice 15 – presenting an exhibit: The advisor uses a mechanical example to illustrate how a specific locking
mechanism works. In most cases, advisors present difference between mushroom and roller cam in the window
fitting while presenting a piece of window hardware. Practice 15 occurs as follows:
effort (e.g., assembly to be done). If the persuadee is clear about all those points, it
becomes a trigger [6] to tackle the HS issues. The advisors apply particular practices
and employ specific materials to support the triggering effect of the advisory encounter.
The identified practices occur across the whole data set and do not show coinci-
dence with the particular advisors or do not result from order effects. However, they are
interrelated. Collaboratively visualizing things (like in Practice 19 and Practice 4)
imply more extensive discussions (Practice 17 and Practice 2). Some practices are
related to the material used and address the visualization of content (Practices 3, 4, 7–
15, 18, 19), others are conversational practices (Practices 1, 2, 5, 6, 16, 17). It is, thou,
obvious that this division is not binary – especially the collaborative practices, rely on
visualization as a common artifact as a basis for conversation (e.g., Practice 4, 5, 7, 8,
9, 19).
Table 3. Persuasive practices related to informing about the next necessary steps

Practice 16 – mentioning the priorities: In general, the advisors suggest upgrading the mechanical security
features (windows and doors) before going for electronics (e.g., alarm system). They provide a list of local crafts-
people who are certified to make specific improvements. After making a short utterance about the necessity to
contact a respective craftsman, they go over to the next topic. Practice 16 occurs as follows:
Practice 17 – discussing the priorities: After mentioning the general priorities, the advisor discusses them with
the homeowner to make sure that they fit her expectations. As opposite to Practice 16, here additional questions
are asked and the advisor makes sure that the homeowner understood the general tendency in this regard. Practice
17 occurs with the following frequency:
Practice 18 – listing things to be done: The advisor writes down and provides a list of things and issues to be
addressed. The list does not give any priority to one or the other problem or solution, but summarizes all topics
addressed throughout the provision of the service in a predefined order (door, windows, cellar, etc.). Alternative-
ly, the respective information is written on the brochures or print outs. Practice 18 is characteristic for the non-IT
condition and is applied there in 12 cases. In two cases (out of those 12), the advisor places the list between him
and the persuadee, such that the notes were made collaboratively.
Practice 19 – prioritizing things to be done: The advisor lists all issues to be addressed and orders them accord-
ing to the priorities and his personal assessment. This practice is supported by the provided IT and occurs only in
the IT condition: the advisor can sort all issues he addressed according to the dimensions of priority and time. In
most cases, he encourages the involvement of the persuadee such that the prioritizing has a collaborative charac-
ter. Practice 19 occurs with the following frequency:
4.2 Relating Persuasive Practices to the Perceptions of the Persuadee

The above analysis shows that supporting visualization of specific content enables for
occurrence of particular conversational and interactional practices. Nevertheless, the
effectiveness of the practices cannot be solely related to their occurrence in the advisory
encounter. In the following, we discuss the relation between PERS and DESA, as well
as the relation between the occurrence of particular practices and those two measures.
Relation DESA – PERS: DESA and PERS stand in relation to each other and both
define important aspects of persuadee experience in persuasion regarding health
behavior as measured by a general, online survey [15]. Our analysis confirms this for the
situation of HS advisory service, by using an onsite survey right after the these sessions.
There is a significant and moderate-to-strong correlation between DESA and PERS
across all our cases (two-tiled bivariate correlation: Pearson’s Coefficient
corr ¼ 0:5; p 0:001) and, especially, in the non-IT cases (corr ¼ 0:6; p 0:005). The
coefficient we measure is higher than reported in earlier research (0.43 [15]). Further-
more, we show that DESA is significantly higher in the IT condition than in the non-IT
with a large effect size (IT: x ¼ 4:72; non-IT: x ¼ 3:95; tð19Þ ¼ 3:29; p 0:005), and,
as consequence, PERS is higher in the IT than in the non-IT with a very small effect size
(IT: x ¼ 4:71; non-IT x ¼ 4:58; tð19Þ ¼ 1:05; p 0:1). We do not observe significant
results for other constructs from the model of Lehto et al. [15].
Influence of Practices on PERS and DESA: When asked about the most positive
episode across both conditions or about the what increased their understanding, many
persuadees point to the visualization potential of the IT – for video: “I valued the fact,
that one could directly show me how a potential burglary can look like” [H13] and for
pictures: “Schemata made it easy to understand the technical solution – they were
good in traditional advisory as well as in the modern one – just with the tablet you
could directly see it at the object” [H16]. The visualization potential of modern
technologies was emphasized in each interview. In particular, persuadees point to
videos and schemata as elements that leverage understanding (“It was a lot easier to
understand the one [advisory session] where I could see the video and pictures” [H8]),
and the practice of recording needs as a way to personalize and individualize the
advisory service (“It was a personalized experience – he addressed my personal
situation. One feels respected if their personal situation gets considered. One feels
proud” [H2]).
Asked about their motivation and ability to take next steps on HS, persuadees point
to the prioritization practices: “This was clearer in the IT, because one could see what
to do next and what can wait” [H16]. Some other account for the role of individualized
pictures: “I found the one with tablet more pleasant cause there was a through and
individualized discussion towards the end (…). I could take this PDF with the photos of
my doors and windows, and go to the craftsman, show this to him and ask for his help”
[H15]. Other persuadees emphasize the role of discussion in general: “The conversa-
tion at the closure, it encouraged me and acts as a reminder of most important things –
independent of whether with IT or not” [H8]. The collaborative character of activities
seems to be very central issue for many test participants: “It helped to understand when
we took the picture together. Clack… ‘Look, there is your door’.” [H9] and “So, my
advisor, she took the pictures, added markings, made notes with me. And I think, those
photos help to remind oneself of what needs to be done” [H18].
On the one hand, the persuadees refer to particular materials used throughout the
advisory sessions: schemata, videos, pictures, etc. On the other hand, they stress the
interactional and conversational character of advisory sessions. We explore this issue
while providing results of statistical dependence analysis of the PERS and DESA
measures and the observed practices. We did not identify any (nearly) significant
relation between practices and PERS or DESA for the non-IT condition. All the
indications presented below describe solely the IT condition in a between subject
mode.
The results of the Kruskal-Wallis test suggest a positive relation between Practice 2
(discussion of homeowner needs) and DESA (Hð1Þ ¼ 2:53; p 0:1), as well as
between Practice 19-b (prioritizing things to be done – together with IT) and DESA
(Hð1Þ ¼ 4:24; p 0:05). We also observe negative relation between Practice 7-a
(taking picture – alone with IT) and DESA (H ð1Þ ¼ 3:33; p 0:1) and between
Practice 7-a and PERS (H ð1Þ ¼ 2:46; p 0:1). Regression analysis, even if it does
not produce a general regression equation for DESA or PERS, it still provides indi-
cations that confirm two dependencies: the positive relation between Practice 19-b and
DESA (B ¼ 1:56; t ¼ 2:91; p ¼ 0:04), as well as the negative one between Practice 7-
a and PERS (B ¼ 0:8; t ¼ 1:8; p 0:1). Finally, the set of two-independent-sample
t-tests suggests the following positive relationships: (1) between presenting the video to
the persuadee (Practice 11) and PERS (tð17Þ ¼ 1:66; p 0:1), (2) between collabo-
rative annotation of the picture (Practice 8-b) and PERS (tð18Þ ¼ 1:64; p 0:1), and
(3) between Practice 19-b and DESA (tð17Þ ¼ 1:66; p 0:1). In summary, only the
relation between Practice 19-b and DESA was yielded in all tests, which seems an
intuitive and still valuable – collaborative work practices with a shared visualization
improve the perceived design aesthetics of the HS session. The other tendencies we
observe confirm that practices which have a collaborative character (discussion,
working together with the tool) may tend to improve perceptions on PERS and DESA,
while avoiding engagement in collaborative practices may lead to negative effects in
this regard.
Our analysis shows that advisors employ a whole range of persuasive practices. They
differ with regard to use of artifacts (IT, brochures, etc.) and with regard to their
collaborative character. Also, some practices address the homeowner’s HS needs and
the next steps, however the largest variety of practices is employed for addressing the
persuadee’s ability by discussing missing security features. The advisors put by far
most effort to support these activities by using brochures, exhibits, and referring to
windows and doors. With introduction of IT, new relevant practices emerge, such as:
collaborative picture taking or video watching. Practices that improve persuasive
character of the encounter involve collaboration and discussion around a shared artifact
– according to the interview data and the quantitative analysis of survey responses.
Importantly, IT improves design aesthetics (DESA), but has negligible effect on per-
suasiveness (PERS) as shown by comparison between conditions. Consequently, it is
not the technology itself that enhances persuasiveness in HS encounters, but the way it
is used makes the difference. In our opinion this is the key for further research in the
area of PT for interpersonal interaction where successful persuasion is essential.
Hitherto, the focus of research in PT was on the human-computer influence, as well
as computer-mediated and computer-moderated interpersonal influence [26]. The
results suggest, that technology designed along the same lines has potential to establish
effective practices for face-to-face interpersonal influence – in particular it shows that
specific practices can easily emerge and supplement existing practices if appropriate IT
is provided. We show that PT can be well applied in situations where a human per-
suader is needed due to the high complexity of decisions to be taken, as opposite to
more classical application scenarios with a clear target state or behavior [1, 23]. This
opens a new, fascinating area for PT researchers and shows first directions of research:
establishing persuasive practices as work practices between persuader and persuadee.
In the course of generalizing our observations, it is easy to imagine that systems
like the one used in current study, can be effortlessly extended in accordance with other
design principles borrowed from PT. In the case of HS, this includes information on the
improvements done by other people in similar situation and their attitude. We argue
that PT community needs to extend its research focus beyond this limit [23] – it is easy
to speculate about possible direction of research, e.g., health support apps where
computer-mediated or -moderated persuasive systems are linked with advisory services
at doctor’s office. Through the design and application of the tool presented above, we
show that the guidance developed for the computer-human influence [6] is applicable
for interpersonal encounters. Our research, also, contributes to the knowledge on IT
support in advisory services. So far, research in this area addresses concepts as
transparency [19] or education [9], thus addressing the objectivism of such encounters.
However, considering persuasion in this context stresses a different side of those

encounters being a meeting of two socially and organizationally linked actors following
their goals. The identification of relevant persuasive practices helps the designers to
engineer systems inducing those practices and the researcher to identify design ele-
ments linked with persuasion and differentiate them from features for facilitation or
moderation.
Finally, the set of collaborative practices identified via statistical tests as having
potentially strong influence on persuasiveness confirms the importance of practice-
based studies for the PT. While following the general description of practices from the
consulting literature [2], we are able to identify particular practices specific in the given
context and provide a zoom-in analysis [18]. Thereby we confirm the value of such
perspective in the PT research and claim, that it is the appropriate way to study effects,
especially in human-human influence scenarios. The results do not come without limi-
tations: practices perspective focuses on local, internal validity over the external one.
Nevertheless, we show its potential for research in PT to explore human-human influence
in IT-supported encounters and claim that the results can be adopted in other scenarios,
such as medical advice – doctor who fails to convince a patient via words, could, e.g.,
employ videos showing negative impact of particular factor on the patient’s body. We
call for intensification of research oriented at practices resulting from use of PT in real
situations and for deepening the understanding of persuasive practices. Alike in general
HCI, we call for practice-turn in PT research [13]: What other persuasive practices
emerge? Which of them can be supported by PT? How to design appropriate PT?
References
1. Arroyo, E., et al.: Waterbot: exploring feedback and persuasive techniques at the sink. In:
Proceedings of International Conference on Human Factors in Computing Systems,
pp. 631–639. ACM (2005)
2. Blundel, R.: Effective organisational communication: perspectives, principles, and practices.
Financial Times Prentice Hall, Harlow, England; New York (2004)
3. Budde, R., Züllighoven, H.: Software tools in a programming workshop. In: Floyd, C.,
Züllighoven, H., Budde, R., Keil-Slawik, R. (eds.) Software Development and Reality
Construction, pp. 252–268. Springer, Heidelberg (1992)
4. Cialdini, R.B.: Influence: the psychology of persuasion. Collins, New York (2007)
5. Comes, T., Schwabe, G.: From fuzzy exploration to transparent advice: insights into mobile
advisory services. In: Hawaii International Conference on System Sciences. Los Alamitos,
USA (2016)
6. Fogg, B.J.: A behavior model for persuasive design. In: Proceedings of the 4rd International
Conference on Persuasive Technology. ACM (2009)
7. Garfinkel, H.: Studies in Ethnomethodology. Prentice-Hall, Englewood Cliffs (1967)
8. Harjumaa, M., Oinas-Kukkonen, H.: Persuasion theories and IT design. In: de Kort, Y.A.,
IJsselsteijn, W.A., Midden, C., Eggen, B., Fogg, B.J. (eds.) PERSUASIVE 2007. LNCS,
9. Heinrich, P., et al.: Microworlds as the locus of consumer education in financial advisory
services. In: Proceedings of the International Conference on Information Systems (2014)
10. Hutchby, I., Wooffitt, R.: Conversation Analysis: Principles, Practices, and Applications.
Polity Press, Cambridge (1998)
11. Kanji, G.K.: 100 Statistical Tests. SAGE Pub., London, Thousand Oaks (2006)
12. Kress, G.: Multimodality: A Social Semiotic Approach to Contemporary Communication.
Routledge, Abingdon, New York (2009)
13. Kuutti, K., Bannon, L.J.: The turn to practice in HCI: towards a research agenda. In:
Proceedings of the Conference Human Factors in Computing Systems, pp. 3543–3552.
ACM Press (2014)
14. Leeper, J.D., Hartman, J.: Choosing the Correct Statistical Test (CHS 627: University of
Alabama), http://bama.ua.edu/*jleeper/627/choosestat.html
15. Lehto, T. et al.: Factors affecting perceived persuasiveness of a behavior change support
system. In: Proceedings of the International Conference on Information Systems (2012)
16. Mettler, T., et al.: On the use of experiments in design science research: a proposition of an
evaluation framework. Commun. AIS. 34(1), 223–240 (2014)
17. Muntean, C.I.: Raising engagement in e-learning through gamification. In: Proceedings of
the International Conference on Virtual Learning, pp. 323–329 (2011)
18. Nicolini, D.: Practice Theory, Work, and Organization: An Introduction. Oxford University
Press, Oxfor (2012)
19. Nussbaumer, P., et al.: “Enforced” vs. “Casual” transparency – findings from IT-supported
financial advisory encounters. ACM Trans. Manag. Inf. Syst. 3(2), 11:1–11:19 (2012)
20. Oinas-Kukkonen, H., Harjumaa, M.: Persuasive systems design: Key issues, process model,
21. Oinas-Kukkonen, H., Harjumaa, M.: Towards Deeper Understanding of Persuasion in
Software and Information Systems. In: Proceedings of the International Conference on
Advances in Human-Computer Interaction. IEEE (2008)
22. O’Keefe, D.J.: Persuasion: Theory & Research. Sage Pub, Thousand Oaks, CA (2002)
23. Rogers, Y.: Moving on from weiser’s vision of calm computing: engaging UbiComp
experiences. In: Dourish, P., Friday, A. (eds.) UbiComp 2006. LNCS, vol. 4206,
pp. 404–421. Springer, Heidelberg (2006)
24. Schwabe, G., et al.: Advancing collaboration engineering: new ThinkLets for dyadic
problem solving and an application for mobile advisory services. In: Hawaii International
Conference on System Sciences, Los Alamitos, USA (2016)
25. Simons, H.W., Jones, J.: Persuasion in society. Routledge, New York (2011)
26. Stibe, A.: Advancing typology of computer-supported influence: moderation effects in
socially influencing systems. In: MacTavish, T., Basapur, S. (eds.) PERSUASIVE 2015.
27. Suchman, L.A.: Human-machine reconfigurations. Cambridge University Press, Cambridge
(2007)
Persuasive Patterns in Q&A Social Networks
Ifeoma Adaji ✉ and Julita Vassileva

( )
University of Saskatchewan, Saskatchewan, Canada

[email protected],[email protected]
Abstract. Social networks differ in their structure and objectives, hence, the
effectiveness of persuasion principles or patterns may vary from one type of
network to another. This study aims to identify the persuasive principles that make
a typical Q&A network successful. Using Stack Overflow as a case study, we
applied the PSD model to evaluate the persuasive principles of the network. Our
results show that all but four principles of the PSD model were implemented in
Stack Overflow. This study can help social network developers build persuasive
Q&A social networks and improve on existing social platforms.
Keywords: Persuasive technology · Social networks · Persuasive systems design
1 Introduction
Online social networks are peculiar in that they have different objectives and operate in
different ways, thus making a “one size fits all” model difficult to apply to the various
social networks. For instance, the objective of users on Facebook and Stack Overflow
differ, and so does the structure of both networks. Hence, the persuasive strategies that
will work in Facebook might not work in Stack Overflow. This study aims at identifying
the persuasive patterns that make a typical Q&A social network successful. These
patterns can help social network developers and stake holders build or improve existing
social platforms, and serve as a guide to building a persuasive framework for Q&A social
networks.
2 Related Work
2.1 Stack Overflow
Stack Overflow is a Q&A network where users can ask and answer specific IT related
questions to gain points and build their reputation. Users do not need to sign up in order
to view existing answers to already asked questions. However, in order to ask or answer
a question, a user has to sign up. Through active participation, asking good questions
and providing good answers, users can gain incentives, such as higher reputation score,
badges and privileges. While users can up vote or down vote other users’ questions and
answers, only the user who asked a question can select the best answer to his/her

DOI: 10.1007/978-3-319-31510-2_16
190 I. Adaji and J. Vassileva
question. Stack Overflow currently has over 4 million registered users and over 9 million
questions1.
2.2 Persuasive Systems Design (PSD) Model

There are currently several frameworks and strategies for designing persuasive systems
in different domains. A review carried out by Wiafe and Nakata [1] shows that the
frequently applied persuasive frameworks or models include Fogg’s functional triad [2]
and the transtheoretical model [3]. Despite the popularity of these frameworks, in this
study we applied the persuasive systems design model (PSD) [4]. We used this model
for two reasons. First, the Fogg’s functional triad has been studied extensively over the
years and new frameworks have been developed based on his model, the PSD being one
of them. Second, as noted by Oinas-Kukkonen and Harjumaa [4], Fogg’s framework
and principles are too general in terms of designing and evaluating persuasive systems.
Though the transtheoretical model is a common framework, it is however mainly used
to model health behavior change [3].
PSD is a framework for designing and evaluating persuasive systems. It categorizes
and maps the elements of persuasion in a system and also describes the software func‐
tionality expected in the end product [4]. The PSD model recommends three phases of
development and evaluation: understanding the key issues behind persuasive systems,
analyzing the persuasion context and designing of system qualities. In this study, we
only evaluated the third stage, designing of system qualities, as we are interested in
identifying the persuasive patterns and principles adopted in the design of a system.
3 Research Method and Results
In this study, our focus was on the third phase of the PSD model: design of system
qualities. This stage is important as it focuses on the persuasive principles that should
be adopted in designing a system to make it more engaging. The principles are catego‐
rized into four groups: providing primary task support, dialogue support, system cred‐
ibility support and social support [4]. We analyzed Stack Overflow and identified the
persuasive principles that exist and how they were implemented.
This project is still work in progress. In the next phase of this study, we will validate
the different principles identified below, by carrying out a user study with Stack Over‐
flow users to verity if these principles work in practice.
3.1 Primary Task Support
The principles of persuasion in this category assist users of a system in achieving their
primary objective. These principles include the following:
1
https://data.stackexchange.com/StackOverflow/revision/325050/420211/count-of-all-users-
all-questions-and-all-answers.
Persuasive Patterns in Q&A Social Networks 191
Reduction: With the principle of reduction, a system should make complex tasks as
easy as possible by reducing them into simpler ones to persuade users to carry out such
tasks. For example, a very short registration process for users makes it more likely that
users will register since the task of signing up has been reduced to something simpler
[5]. Stack Overflow reduces the task of registration by allowing users sign up with their
Facebook or Google account, hence eliminating the need for users to fill out a registration
form which some could regard as time consuming.
Tunneling: The tunneling principle states that a system should guide users through a
process and provide means of persuading them along the way. We could not identify
any application of tunneling in Stack Overflow.
Tailoring: A system that allows users to tailor the information or service to their inter‐
ests or preferences could be more persuasive than one that does not. With the tailoring
principle, different user groups should be provided with tailored information for that
group. Stack Overflow applies the tailoring principle by letting users specify what
subject areas they are interested in and receive notifications when questions are asked
in these areas. This is done with the use of tags. Tags are labels or keywords used to
classify a question2. Users can subscribe to as many tags as they have interest in, and
are notified of questions that have those tags. This prevents users from receiving noti‐
fications when questions they have no interest in are asked.
Personalization: The principle of personalization states that a system that offers more
personalized content is more likely to be persuasive compared to one that does not. In
Stack Overflow, when users register for the first time, a profile is created for them. The
system uses this profile to personalize their experience. On the profile page, among
others, users can change their display name, location, edit what information that other
users can see about them, edit the tags they have interest in, and edit what notifications
they want to receive. All this information is to ensure the system personalizes the content
the user receives. This principle is very similar to tailoring.
Self-monitoring. According to the self-monitoring principle, for a system to be persua‐

sive, users should be able to keep track of their performance or status, as this will support
users in attaining their goals. Stack Overflow allows self-monitoring of users’ perform‐
ance. On each user’s profile page, they can see a summary of their activities on the site.
The details of a user’s performance include their reputation score, badges they have
received, the number of questions they have answered, the number of questions they
have asked, any points earned from the questions and answers, number of profile views,
how long the user has been a member and the last time the user visited the network.
Going by the self-monitoring profile, having access to this information could persuade
users to use the system.
Simulation: The simulation principle states that persuasive systems should enable
users to see the link between cause and effect. Stack Overflow allows users to observe
2
Tags in Stack Overflow, http://stackoverflow.com/tags.
the connection between posts and reputation score. A user’s profile page states the
number of questions and answers the user has asked and answered and the reputation
score earned from those posts. Hence users can see that there is a link between posting
questions or answers and their reputation score. This, according to the simulation prin‐
ciple, could persuade users to answer more questions.
Rehearsal: For a system to be persuasive using this principle, the system should
provide a means for users to rehearse a target behavior. A typical example is a flying
simulator, which student pilots can use to rehearse being certified pilots. We could not
identify the use of this principle in Stack Overflow. What we found to be similar to this
however, is that unregistered users can view existing questions and answers. They
however cannot ask or answer questions. They also do not have a profile so cannot
benefit from personalization. Viewing existing questions and answers could give
unregistered users a feel of the site and hence persuade them to register.
3.2 Dialogue Support
This category of influence principles uses a feedback system to provide support to users
with the aim of bringing them closer to their goal. In Stack Overflow, this goal could
include asking a question, answering one, improving one’s reputation score or finding
questions that could be useful to a user. The dialogue support category includes the
following persuasive principles:
Praise: According to the PSD model, a persuasive system should use praise as a means
of feedback to users, as praise can make a user more open to persuasion. Praise could
be in the form of words, images, symbols or sounds. Stack Overflow uses praise in the
form of up votes. When a user asks a question and other users find the question valuable,
they can vote it up. The number of up votes serve as a feedback to the user who asked
the question. The up votes also provide a feedback on the quality of the answers provided
to a question. Users can up vote the answers they feel provide the best solution to the
question asked. In Stack Overflow, praise is also offered in the form of favorite question.
Users who think a question is relevant or useful can make it one of their favorite ques‐
tions. Accepted answer is another form of praise in Stack Overflow. The user who asks
a question is the one to pick the best answer from the various answers provided. All
these forms of praise are displayed beside the question or answer so are visible to all
users. They also contribute to the final reputation score of the users who earned them.
Rewards: The rewards principle states that systems that offer rewards to their users
could be more persuasive than those that do not. Stack Overflow offers rewards to users
in form of privileges and badges. Privileges control what users can do in the network.
The higher the privilege, the more actions a user can perform. Privileges are gained by
increasing one’s reputation score, and reputation score is increased by posting helpful
questions and answers3. Users with reputation score of over 25,000 for example, are
3
Privileges in Stack Overflow, http://stackoverflow.com/help/privileges.
rewarded with the privilege of access to site analytics which includes Stack Overflow’s
internal and Google site analytics.
Badges is another reward offered to users in Stack Overflow. Users receive badges
when they are especially helpful4. Badges earned by users can be seen on their profile
page and on their posts. An example of a badge in Stack Overflow is the Great Question
badge. It is rewarded to users who score 100 or more on a question. Scores are computed
by the number of up votes a question receives. There are over 40 badges in Stack
Overflow and are classified into various categories.
Reminders: The rewards principle asserts that users are expected to achieve their target
objective if a system reminds them to do so. Hence, a system that implements a reminder
technique could be more persuasive. We could not identify the use of reminders in Stack
Overflow.
Suggestion: According to the suggestion principle of the PSD model, systems that
offer suggestions to users are more persuasive than systems that do not. Stack Overflow
offers suggestions to users in several scenarios, for example, when a user is typing a
question. When a user asks a question, the system searches for similar questions that
have been asked and answered in the past and suggests such questions to the asker. The
asker then has the opportunity to review the suggested questions and decide if they are
useful, in which case the user does not go ahead with asking a new question. Another
use of suggestion in Stack Overflow is with the use of tags. Tags are labels or keywords
used to classify a question. Not all users can create a new tag. Create tag is a privilege
rewarded only to users when they gain a reputation score of 1,500. However, all users
can use existing tags. At the point of typing a new question, the system suggests possible
tags that a user can choose for the question.
Similarity: The similarity principle states that systems that remind users of themselves
are more likely to be persuasive. In Stack Overflow, we could not identify the use of
similarity principle in this context. However, in our future study, one of the questions
we will try to answer is “Do users have friends or people they admire who use Stack
Overflow, if yes, do they form special networks”. This will help us identify how persua‐
sive the similarity principle is.
Liking: This principle asserts that a visually attractive system is likely to be more
persuasive. Rating the attractiveness of Stack Overflow is subjective and can only be
determined by the users of the system, hence we did not give an opinion of this principle.
Validating the persuasiveness of this principle is one of the objectives of the user study
that we intend carrying out in the future.
Social Role: The social role principle states that systems that embrace the use of social
roles are likely to be more persuasive. Stack Overflow has several social roles which are
rewarded to users as they achieve various milestones. For example, when a user attains
a reputation score of 10,000, they are awarded the access to moderator tools privilege
4
Badges in Stack Overflow, http://stackoverflow.com/help/badges.
which enables them access reports, delete inappropriate or duplicate questions and
answers, add new tags, create and moderate chat rooms among other functions. Knowing
that there are users in such social roles could persuade users to use the system.
3.3 Social Support

This category of principles leverages on social influence to persuade users to use a
system. The principles in this category include:
Social Learning: According to the PSD model, a system using the social learning
principle should allow users to observe other users performing their target behavior in
order to see the outcome of such behavior. In Stack Overflow, users can observe other
users performing their target behavior. Since the questions asked by users are public,
and so are their comments, votes, favorites, as well as the answers to the questions, their
votes and the best answer award, the users can learn from these community actions, both
the social norms of posting, answering and providing feedback, and about the subject
of the posts.
Social Comparison: This principle asserts that users are more persuaded to perform
their target behavior if they can compare their performance to that of others within the
system. According to Stack Overflow, reputation is a measure of the confidence the
community has in a user5. The better the performance of a user, the more reputation
score he earns. Users can view the reputation score and other privileges earned by other
users within the network by viewing their profile. They can compare their performance
with those of others by looking at their profile.
Normative Influence: The principle of normative influence states that a system should
take advantage of peer pressure to increase the possibility of persuading a user to adopt
a target behavior. According to this principle, a system that can bring users who share
the same goals together, is more likely to persuade its users to use the system. In Stack
Overflow, users are brought together in chat rooms where they can share knowledge and
information. A user with the communication privilege can create a chat room and grant
talking privileges to other users who can then join in the conversation. Another example
of normative influence is the use of the reward system. Users could feel persuaded when
their friends earn high reputation or badges in the system.
Social Facilitation: This principle states that a system’s users are more likely to be
persuaded to use the system if they can discern from the system that other users are
carrying out a similar task along with them. In Stack Overflow, an example of the
implementation of this principle is the chat room. From Stack Overflow’s chat portal6,
users have an overview of active chat rooms including a description of the chat room
and the number of active users. If a user clicks on any of the active sessions, he or she
5
Reputation in Stack Overflow, http://stackoverflow.com/help/whats-reputation.
6
Stack Overflow, http://chat.stackoverflow.com/?tab=all&sort=people.
can view the details of the chat including the names of the users, the details of their
conversation and time stamp. Going by the social facilitation principle, users in the chat
room are persuaded to participate because other users are active.
Cooperation: The cooperation principle of the PSD model suggests that for a system
to be persuasive, it should take advantage of human beings’ natural desire to co-operate.
In other words, a system should provide a method of co-operation among users. In Stack
Overflow, users are able to co-operate with other users when asking questions. For
instance, users can comment on a question in order to improve on it. This kind of co-
operation among users is also evident with answers provided on Stack Overflow. It is
common for users to comment on answers provided by other users to ask for clarification
or to improve an existing answer.
Competition: The competition principle states that for a system to be persuasive, it

should take advantage of human beings’ natural desire to compete. In other words, a
system should provide a method of competition among users. Stack Overflow rewards
users who are first to answer a question with a score of ten or more. This type of reward
could lead to competition among users who want to receive the first accepted answer
reward.
Recognition: According to the PSD model, offering public recognition to an individual

can increase the possibility of that user being persuaded. In Stack Overflow, a user’s
reputation score and badges earned form part of the user’s display image and are found
beside every question, answer or comment posted by the user. Hence, it is easy to see
the user with the highest reputation score and rewards from a list of users who answered
a question or who posted comments to an answer. This makes recognition of users
possible.
This paper identified the persuasion principles adopted in a typical Q&A social network,
Stack Overflow. Using the PSD model, we were able to identify all but four principles
of the model namely tunneling, rehearsal, reminders and similarity. Each principle
identified promotes engagement of users. This research shows that persuasion in a
typical Q&A social network adopts regular observable patterns. Software developers
can adopt these principles to develop applications that will possibly succeed.
This study is still work in progress. We were unable to evaluate the system credibility
support category of the PSD model. This category is based majorly on feedback from
users. Evaluation of this category will be carried out in the next phase of the project. In
that phase, we will also validate the different principles listed above, by carrying out a
user study with Stack Overflow users’ data, to show that these principles actually work
in practice. For instance, one of the research questions we will answer in the user study
is if users are persuaded to answer more questions when they receive a hard to earn
badge. The study will show the impact and effect of these persuasive principles on user
participation.
References
1. Wiafe, I., Nakata, K.: Bibliographic analysis of persuasive systems: techniques, methods and
domains of application. In: Persuasive Technology, vol. 61 (2012)
2. Fogg, B.J.: Persuasive technology: using computers to change what we think and do. Ubiquity
2002, 5 (2002)
3. Prochaska, J.O., Velicer, W.F.: The transtheoretical model of health behavior change. Am. J.
Health Promot. 12, 38–48 (1997)
Øhrstrøm, P. (eds.) PERSUASIVE 2008. LNCS, vol. 5033, pp. 164–176. Springer, Heidelberg
(2008)
5. Fogg, B.J., Eckles, D.: The behavior chain for online participation: how successful web services
structure persuasion. In: de Kort, Y.A., IJsselsteijn, W.A., Midden, C., Eggen, B., Fogg, B.J.
(eds.) PERSUASIVE 2007. LNCS, vol. 4744, pp. 199–209. Springer, Heidelberg (2007)
Games and Gamification
Utilizing a Digital Game as a Mediatory Artifact
for Social Persuasion to Prevent Speeding
Bernhard Maurer ✉ , Magdalena Gärtner, Martin Wuchse,

( )
Alexander Meschtscherjakov, and Manfred Tscheligi

Center for Human-Computer Interaction, Department of Computer Sciences,
{bernhard.maurer,magdalena.gaertner,martin.wuchse,
alexander.meschtscherjakov,manfred.tscheligi}@sbg.ac.at
Abstract. In this paper we present a game-based approach to stop a driver from

speeding by means of social persuasion. The approach utilizes a digital game
played by a passenger inside the car. The game serves as a mediatory artifact,
which translates the speed of the car into in-game events, thus, nudging the
passenger to communicate with the driver about his/her driving behavior. As a
game we used Tetris, which was coupled to the speed of a virtual vehicle in our
driving simulator. We designed four different in-game representations of the real
car data and examined, which of these designs is most suitable to trigger an intui‐
tive, understandable linkage between the speeding behavior and the corre‐
sponding in-game events in order to enable a prompt intervention of the
passenger. We evaluated the four designs in an exploratory user study. Our find‐
ings highlight the feasibility of our approach, as even passengers, who were rather
uninvolved in the driving task, were successfully encouraged to slow down the
driver. Based on our study results, we recommend a hybrid design strategy for
the game, between designing for a dynamically increasing in-game challenge to
foster passenger engagement based on fun, and simultaneously intervening
dynamically in the playability of the game to foster communication with the driver
to pave the way for social persuasion in the car.
Keywords: Persuasive game design · Social persuasion · Automotive domain
1 Introduction
Understanding and designing games as socially influencing systems can be an insightful

perspective and fruitful opportunity for researchers and designers of persuasive tech‐
nology, in particular in the automotive domain, because the car is an inherently collab‐
orative and social space that creates social interdependencies between the co-located
persons in the car [7]. Yet, the car is also a safety critical environment where incautious
driving behaviors such as speeding can result in dangerous, or even fatal incidents, not
only affecting the driver but all people in the car. Hence, having a means of persuasion
in order to prevent drivers from speeding is an important design challenge for persuasive
technologies in the automotive domain.

DOI: 10.1007/978-3-319-31510-2_17
200 B. Maurer et al.
Games and play (both digital and physical) have the ability to create a social setting
and connect people within the boundaries of this setting. Through collaborative game‐
play social couplings are formed by which people communicate and interact with each
other for an underlying, common goal [9]. In many games, gameplay is inherently based
on combining various social influence principles such as competition, cooperation,
recognition, and learning. Thus, interweaving game design and persuasive system design
by incorporating these principles into specific game mechanics offers great potential for
designers of persuasive technology to generate impact and induce behavior change.
Following this assumption we present an approach that incorporates the real speed
of a car in a digital game. With this linkage we are deliberately intervening in the social
setting between the driver and a passenger with the aim to utilize the passenger as a
resource for social persuasion in order to keep the driver from speeding. Our approach
is based on the following scenario: A person is driving a car accompanied by a passenger.
The passenger is playing the tile-matching, digital puzzle game Tetris during the car
trip. Some features of the game are linked to the actual speed of the car in real time.
Thus, when the driver starts speeding, his/her driving style is immediately incorporated
in the game. As a consequence the game gets harder to master or even unplayable, and
the passenger is alerted that the driver is exceeding the speed limit. By communicating
the status of the game to the driver, the passenger consequently persuades the driver to
reduce the driving speed – first and foremost, for the safety of the two, but also for
keeping up a positive gaming experience.
In terms of interaction design, there are several ways of how to link the car’s driving
speed to in-game changes. This inlcudes, e.g., directly coupling the game speed with
the driving speed (i.e. the faster the driving speed is, the faster the Tetris tiles fall down)
or even pausing the game when the car exceeds a certain speed limit. We implemented
four different game designs and evaluated their effect in a driving simulator study. We
aimed at answering two main research questions:
1. Does a digital game, whose game mechanics are based on real car-data (i.e., speed),
serve as a useful mediatory artifact to foster communication as a means of social
persuasion in the car to prevent speeding?
2. Which of the four game designs are the most useful and promising ones in order to
foster social persuasion in the car?
i. Which game designs create the most intuitive and easy to understand linkage
between the real-world misbehavior of speeding and the in-game consequences/
events?
ii. Which game designs do the passengers prefer and for which reasons?
In the following, we present the theorectical background of our research and describe
the development process of the four different game designs we used for our study. To
examine our approach and identify the most promising game design(s), we evaluated
the designs with 16 participants in an exploratory user study in a car-simulator. We
report on the study procedure, the data analysis and discuss the findings of the study
with regard to the research questions and future work.
Utilizing a Digital Game as a Mediatory Artifact 201
2 Related Work
2.1 Games as Socially Influencing Systems
Digital games are information systems that are highly social and collaborative often
incorporating a variety of social influence principles such as social learning, cooperation,
or competition. Oinas-Kukkonen et al. [10] highlight the role of social influence as a
persuasive strategy by providing a framework with a variety of principles also prevalent
in most games (i.e., social learning, social comparison, normative influence, social
facilitation, cooperation, competition, recognition). In a similar direction, Bogost argues
that the persuasive effect that games create can be understood by seeing them as proce‐
dural rhetorics, i.e., “the art of persuasion through rule-based representations and inter‐
actions” [2].
In our approach, the game uses and embodies the real car speed, thus forming a
procedural rhetoric around translating this data related to driving misbehavior into
changes within the game (i.e., a rule-based representation of real world data).
2.2 In-Car Technology and Collaboration
Today, cars are pervaded with technology like entertainment systems, or navigation
devices. These technologies are often not designed around the social nature and collab‐
orative mechanisms within the car, though. Perterer et al. [7] highlight how the sharing
of information (e.g., making it visible for both driver and passenger) can have large
impact on the collaborative nature of driving. Previous research has in particular focused
on investigating concepts around driver-passenger cooperation e.g., for navigational
tasks [3, 4, 7], while Shepherd et al. [8] investigated the influence of the social setting
within the car on the driver’s behavior itself. They examined a scenario with peers posing
as passengers who provide verbal feedback in order to persuade the driver to change
his/her driving behavior. Their findings show how effective the influence of peers as
passengers can be in risk-related driving scenarios.
Our approach differs from these approaches, though, as we design for the passengers
as a means for social persuasion. In providing the passenger with a digital game, which
incorporates real-car data, we assign him/her to take the role of a social persuader, who
influences the driver. With the game serving as a mediatory artefact in this process, we
connect the persuasive powers of digital gaming and social influence in order to change
the driver’s driving style and keep him/her from the gas pedal.
3 Approach
The motivation for our approach originated from the results of a probing study we
conducted to investigate the persuasive potentials of in-car interfaces [5]. This study
identified the potential of creating technology driven social spaces within the car for
persuasive purposes. Inspired by the design ideas presented in this paper, we developed
our approach to use a game to create a social coupling between driver and passenger.
We use and intervene in this coupling based on different embodiments of real car data
that are translated into different in-game interventions. The game serves as a mediatory
artifact the social coupling is formed around and facilitates social communication as a
persuasive means. When the driver starts speeding the game adapts to this problematic
behavior, and as a consequence becomes a lot harder to master or even unplayable. This
urges the player of the game (i.e., the passenger) to actively communicate with the driver,
which potentially results in persuading the driver to stop speeding. In order to make that
happen, the passenger, who is playing the game during the ride, is supposed to link the
in-game changes to the actual driving behavior. Therefore, one major challenge in the
design process was to find a suitable way of translating the problematic, real-world
driving behavior of speeding into in-game events and changes in gameplay. This linkage
not only had to work on a technical but also on a semantic level for the passenger in
order to enable him/her to persuade the driver effectively. Another challenge was to find
an accurate balance between maintaining the safety of the driver and passenger, as well
as creating a positive gaming experience, which keeps the passenger engaged with the
game, hence aware of the driver’s behavior.
3.1 Game Design
As a starting point for the different game designs we wanted to explore, we chose the
tile-matching game Tetris as it is well known, skill-based and relatively easy to play.
As it is also a rather abstract game with little theming, we were able to pursue our
approach and adapt the game mechanics according to our design goals (i.e., creating
different levels of game interventions) without breaking the aesthetics and recognition
value of the game. We used an open-source version of Tetris (www.openprocessing.org/
sketch/34481) to develop the prototype and adapt the game towards the different designs.
As it would have been an interfering factor for two of our designs (acceleration and
deceleration), we decided to remove the possibility for the player to drop the tiles imme‐
diately, in contrast to the implementation of the original game. Thus, we ensured that
players could not bypass the speed dependent game mechanics, because that would have
otherwise made it difficult for them to recognize the linkage between the real-car speed
and the in-game changes. In the following the four different game designs we created
are described in detail. In each design the car’s speed has a different effect on the game‐
play. These effects vary in the strength of their consequences on the gameplay. More
precisely, the designs are ranked in descending order, from the most severe to the least
severe interference regarding the game’s playability (see also Fig. 1):
• Freeze: The game freezes, when the driver is speeding (i.e., the playing area turns
black, tiles are not visible anymore and the game is paused). Playing is no longer
possible for the passenger, until the driver reaches accurate driving speed again. The
goal of this design was to completely interupt the player in the playing experience.
• Blocking Input: The passenger cannot move the tiles to the left or right, or both, or
rotate them anymore on a gradual level depending on the severity of the speeding
violation (i.e., the faster the driver is driving, the more control elements are blocked).
This design aims at removing the player’s input capabilities stepwise in order to create
a notion of loss of control and simultaneously increase the difficulty level.
• Acceleration: The speed of the falling tiles is changing according to the actual speed
of the car. The faster the driver gets, the faster the tiles fall and the harder the game
gets to master. The goal was to gradually increase the challenge for the player up to
a point where the game actually becomes unplayable, depending on the severity of
the speeding.
• Deceleration: This design is similar to the aforementioned design, but this time in
reverse manner. The faster the car is driving, the slower the tiles are falling. Our goal
with this design was to explore how we could make the game “less fun” for the
passenger if the driver is speeding (i.e., making it extremely slow in order to
completely remove the challenge aspect of the game as well as the typical expected
game progress of tiles falling faster and faster).
Car Speed 50 60 >70

Game Designs
X
Blocking input
X X X X X
Freeze
Acceleration
Deceleration
Fig. 1. Schematic of the four different game designs (blocking input, freeze, acceleration,
deceleration): Starting with a speed limit of 50 km/h, different changes in the game mechanics
and behavior are induced depending on how much the driver exceeds the speed limit.
4 Exploratory User Study
In order to investigate the general potential of our approach to act as a trigger for
communication as means of social persuasion, as well as to explore the individual
designs, we conducted a user study in the driving simulator (see Fig. 2). We focused on
the passenger’s role of being a successful persuader by means of the game in order to
keep the driver off the gas pedal. Thus, the role of the driver was executed by one of our
colleagues.
We conducted our study with 16 participants (9f, 7 m), aged between 22 and 35
years, with a mean age of 30 years, who all took on the role of the passenger in the study.
All participants were rather experienced drivers, who held their driving licenses for an
average of 11.5 years. The annual mileage travelled by car of all participants averaged
out at 7.170 km. Eleven participants declared themselves to be actively playing digital
games, whereas five participants don’t play digital games at all. The gamers were rather
moderate ones regarding the frequency and duration of playing, as the mean gaming
frequency was once a week, with one gaming session lasting about 65.5 min on average.
All participants were familiar with the game Tetris, and 15 out of 16 participants had
already played the game at least once before.
4.1 Study Procedure

At the beginning of the study each participant was asked to sign an informed consent.
Then participants were seated next to the driver and introduced to the game and its
controls, as well as, the study procedure. To allow for consistent study conditions, we
briefed one of our colleagues to take on the role of the driver. He was told to generally
stick to the speed limit (50 km/h) but to exceed this limit at certain spots on the track in
order to provoke the desired in-game changes. He was allowed to communicate with
the passenger, but only if the communication (i.e., asking the driver to stop speeding)
was first initiated by the passenger.
The study was realized as within-design, with each participant testing all four
designs. To prevent aftereffects, the sequence in which the game designs were tested
was permutated. The participants were informed beforehand that the driving style of the
driver would somehow influence the game, but without going into details on the exact
in-game changes or the influencing car-parameters. Furthermore, participants were
encouraged to think aloud of their experiences and impressions during the ride, as well
as, to make assumptions about the underlying linkage of the real-car data and the in-
game changes. After testing a design, participants were asked to outline their assump‐
tions about how the driver behavior and the game design were linked with each other,
how much the game design encouraged communication with the driver and how much
fun it was to play. One researcher was present on the backseat of the car to lead through
the study and to ask questions after every condition. Another researcher took notes on
observations during the ride, as well as on user comments and feedback after the ride.
After testing all conditions, participants were asked in a final interview on their opinion
on the general approach, as well as, how they would rank the different designs regarding
their ability to foster driver-passenger communication as a means of social persuasion.
Fig. 2. Picture of the study setup within the car simulator. Passenger is playing Tetris on a laptop
while the briefed driver is varying the speed of the car to trigger in-game changes.
4.2 Technical Setup

Our prototype setup used VDrift (vdrift.net) as a simulation software and combined it
with Spacebrew (docs.spacebrew.cc) for communication matters between the game
(running on a network connected laptop) and the car simulation. Spacebrew, is a
websocket-based prototyping framework that was used to connect the game running on
a laptop with the car simulation software. The simulation software sent the current speed
of the car to the game, which in turn translated the speed into the different in-game
changes (based on the aforementioned designs).
5 Results and Findings
5.1 Analysis and Results of Quantitative Data
The quantitative data consisted of participants’ answers on which designs were the most
intutitive ones regarding the linkage between car speed and in-game changes (multiple
answers possible) as well as, the final rankings on which designs were the most fun to
play and also which design fostered communication with the driver the most. For a first
comparative analysis every rank was associated with a predefined number of points.
A first rank was worth 20 points, a second rank 15 points, a third rank 10 points, and a
fourth rank 5 points. Every game design received points from 16 participants, which
were totaled for the final fun and communication score. Furthermore, the number of
mentions (i.e., how often the specific designs were mentioned to be the most intuitive
ones) was counted for each design, and multiplied by 20 for the final linkage score.
Finally, each design was assigned an overall score in points, which totaled the scores
for fun, communication, and linkage (see Table 1).
Table 1. Rankings of the designs, rated in terms of fun, communication, and best linkage. The
overall score combines all three rankings, with a higher score meaning a higher rating.
Condition Fun Communication Best linkage (=nr of Overall score

(=based on points) mentions)
Acceleration 285 205 160 650
Blocking input 150 235 120 505
Freeze 155 230 60 445
Deceleration 215 160 40 415
In terms of fun, condition acceleration was ranked as being the most fun design by
participants, followed by condition deceleration. We think that this is due to the fact
that the design of condition acceleration was based on increasing the challenge for the
players, and for many players in-game challenge is directly related to the fun-level of a
game. Further, we assume that condition deceleration was ranked second highest,
because it represented the least severe interventions in terms of playability of the game
(i.e., the game only got slower). The conditions blocking input and freeze were clearly
ranked last in terms of fun. A reason for this could be, that both of them strongly interfere
with the playability of the game. Additionally, the blocking input design, which was
ranked last, was experienced to be the most frustrating one.
In turn, the conditions blocking input and freeze triggered driver-passenger commu‐
nication the most, whereas condition acceleration and deceleration were ranked as being
the last in terms of communication. We conclude that the more severe the interference
in the playability and the higher the experienced frustration level were the more the
driver and passenger communicated with each other.
Regarding the linkage of the car speed and the in-game changes, design acceleration
was rated best, which was presumably a result of the direct mapping between the in-
game speed and car speed and corresponded to the general expectations of Tetris tiles
falling faster as the game progresses. That is also, why we reason, that the reverse design
implementation deceleration was experienced to be the least intuitive one. We assume,
that blocking input was ranked second, due to the stepwise removal of input controls
which was linked to a stepwise rise in car speed. Condition freeze on the other hand was
designed to be binary (i.e., the game was either on or off), which could have made it
slightly more difficult for participants to establish a reasonable connection between the
speeding and the in-game changes.
Condition accelerate and blocking input received the best overall scores with all
three rankings combined. Based on this result, we argue that a potential real world
implementation of our approach should follow a hybrid strategy between designing
around increasing the challenge for the player (to foster engagement based on fun) and
intervening drastically in the players’ game experience e.g., by stepwise removing the
player’s input capabilities (to foster driver-passenger communication). Such a hybrid
design could dynamically adapt the in-game changes based on the severity of real world
driving misbehavior and change in-game embodiments accordingly (i.e., the more the
driver is speeding the more drastic the in-game interventions could get).
5.2 Analysis and Results of Qualitative Data
The qualitative data consisted of data and notes based on the observations during the
study and questions asked at the end of the game sessions. The data was analyzed
according to the basics of qualitative content analysis [6].
In general, participants thought our approach was fun and that it was feasible in terms
of providing them with a means to stop the driver from speeding. One participant stated:
“The game is an indirect way of keeping more eyes on the road, and it encouraged me
to tell the driver to slow down”. A point of critique was that the mappings of car data
and in-game changes were not dynamic, meaning that the game did not adapt to differ‐
ently skilled players, or players that became better at the game over time. Thus, the
balance between creating enough challenge for the player to have an engaging gaming
experience but at the same time creating in-game situations that to some extent force
the player to give verbal feedback to the driver, is a design challenge for future work.
Participants liked the designs that increased the challenge better than those that punished
the player. Participants also suggested to create game designs around rewarding the
passenger-driver couple for positive driving behavior, rather than punishing the
passenger for the driver’s neagtive behavior.
In the following, the “common themes” and similarities among the different designs
that emerged during the study are described:
Challenge (condition accelerate and decelerate): At a certain point, the game became
boring or unplayable for the participants in these conditions. Thus, the initiated commu‐
nication with the driver was based on the game getting too easy or too difficult. Some
players also felt the need to get better at the game. However, perceived challenge of the
game was a matter of individual player skills. This shows, that for a real world imple‐
mentation of our approach, the mappings between real world misbehavior and in-game
changes would need to be dynamic to the current driving scenario and the player’s skills
and/or game progression in order to enable more long term engagement in the game,
hence awareness for the driving style.
Frustration/Punishment/Loss of control (condition blocking input and freeze): The

communication between driver and player was often a direct result of the player being
interrupted in the flow of gameplay. The discrepancy between seeing the visuals of the
game, but not having adequate control (condition blocking tiles) over the game, felt
punishing and frustrating for the players. This frustration and loss of control was reported
to lead to an increased amount of communication and feedback from player to driver.
Attempts to slow down the driver like “Your speeding is breaking my game!” or “I can’t
play this way, drive slower!” illustrate that the interventions of condition blocking input
and condition freeze were not perceived as being fun for the player, although on the
other hand fostered immediate intervention of the passenger.
Commands/Agreements/Common Goal: The commands and feedback from the

players were very precise and clearly targeted at slowing down the driver in order to
make the game playable and fun to play again. This, in turn, also led to a negotiation
process between the driver and passenger regarding the “ideal speed” as a common goal
in order to have the best gaming experience but also stick to the speed limit. One partic‐
ipant stated “I achieve more within the game if the driver is sticking to the speedlimit”
which illustrates the interconnectedness and common goal between the two of them.
Changing Visual Attention: Some participants were very aware of the speedometer in
the central dashboard of the car simulator. They constantly switched their visual atten‐
tion between the game and the speedometer to compare the car speed and the in-game
changes, whereas others were completely focused on the game itself without actually
looking at the surroundings. Nevertheless, all types of passengers – the active and the
rather passive ones – were aware of the driver’s behavior and interfered, when the driver
was speeding, which proves our approach to be a viable means of involving even an
inattentive passenger in the driving situation.
Social connection: Regardless of the game design, the better the passengers understood
how and why the in-game changes were related to the driving behavior, the more inter‐
action between them and the driver emerged. Some participants stated that they felt like
“playing together co-operatively” and said that the more dependent they were towards
the driver, the more they communicated and wanted to persuade him to stick to the speed
limit. One participant concluded: “I like the idea, of influencing someone’s behavior,
like the driving style, by communicating with each other, which in turn influences the
game, and hence, my gaming experience, and so on.” This mutual influence loop
described by the participant exemplarily illustrates the potential of our approach to
deploy gaming technologies in the car, which unfold their persuasive effects based on
social influence.
6 Discussion
Our approach proved to be feasible in presenting a game that acts as a mediatory artifact
to trigger social persuasion. The different game designs involved the front-seat passenger
into the driving experience even when focused on playing the game. This resulted in
increased amounts of communication between the front-seat passenger and the driver.
Hence, our approach allows to involve co-located passengers in the driving task based
on creating a social coupling between them and fostering embodied communication
through a shared resource and task (i.e., the game).
Dynamic Mappings: In a real world implementation of our prototype, the mappings

between in-game changes and driving behavior would have to be dynamic in terms of
presenting the right challenge to the player at the right time to be applicable in a real car
setting. As perceived challenge is a matter of individual player skills and preferences, a
future iteration of our game prototype could start at a level that suits these factors (i.e.,
being aware of the player’s skill level and game progression).
A major challenge will be to find a balance between safety (i.e., preventing unsafe
driving behavior) and a positive gaming experience. For instance, if the players are
getting better, they can handle the speed of the falling tiles better but should nevertheless
encourage the driver to stop speeding. Hence, the better the player gets, the more the
driver could speed, which would be an undesired side effect. As real world settings are
much more complex, a future implementation of a game in the real car would need more
“intelligence” to react to the ever changing surroundings and traffic situations as well
as the player’s skill level and game progression, in order to present the right in-game
consequence at the right time. If in a future iteration of our approach this level of intel‐
ligence in the game is reached, the game itself as well as the mappings between in-game
changes and real world driving behavior could also be much more complex thus, being
closer to real world scenarios. In that sense, new designs around multiple disadvanta‐
geous driving behavior (e.g., tailgating, insufficient lane keeping, etc.) could be incor‐
porated in one game.
Balance of In-game Interventions (Frustration vs. Persuasion): Regarding the in-

game interventions, the biggest design challenge would be to find the right level between
presenting an appropriate challenge to the player, but also encouraging him/her to
actually influence the driver. In our study, we found that intervening in the input and
control capabilities of the player, as well as pulling the player out of the game by pausing
and interrupting the experience (conditions blocking input and freeze), create frustration
and punishment for the player. These designs provoked more and immediate
communication between driver and passenger, though. On the other hand the designs
that worked with continous effects (condition accelerate and decelerate) were ranked
as being more fun. However, these designs were not ranked as high in terms of fostering
driver-passenger communication, thus, proved to be harder to achieve the desired
persuasive effect of keeping the driver from the gas pedal through social influence of
the passenger. Finding the right balance between creating an understandable link
between in-game and real world, while presenting the right level of challenge and at the
same time also increasing communication based on in-game consequences, is an enour‐
mous design challenge. Based on our study results, we argue for a hybrid design strategy
based on creating a game that, depending on the severity of speeding, also adapts the
severity of in-game interventions as needed. This would in turn be an ideal compromise
between passenger-game engagement and increased communication between driver and
passenger. Based on the results and observations, we conclude that intervening in a
player’s input capabilities or taking her/him out of the game itself (condition freeze),
can result in immediate frustration, however also fostering communication rapidly.
Thus, we argue that designs based on such input interventions are more likely to be
applicable for short term settings where a player is only playing over a short period of
time. On the other hand, designs around the mentioned challenge aspect have the poten‐
tial to create long term engagement, but are a bigger design challenge.
Our approach presents an opportunity to let a person currently focused on gameplay,
simultaneously take part in the driving situation. Creating a shared embodied resource,
i.e., the game itself, in order to foster communication and social interdependence as a
persuasive strategy, proofed to be feasible. This notion is supported by the theory of
embodied simulation [1] which describes how humans create meaning by continuously
constructing mental models of the currently perceived scene, which in our case is
mirroring the experience of the driver based on the current game scene. Different in-
game embodiments can facilitate this mirroring process and create a “window” for the
player to mirror the behavior of the driver to interfere if necessary.
Limitations and Future Work. A persuasive approach in a safety critical environment

like the car that focuses on verbal feedback and increased communication, also has some
conflict potential for the passengers (e.g., increased distraction for the driver). In future
iterations of our prototype this challenge of potentially increased driver distraction
through increased verbal activities has to be mitigated (e.g., by alowing the driver to
take over control and disable the connection of the real-car data and the in-game events
at some point). In our future work we plan to iterate our designs based on the study
findings and create a game that incorporates the mentioned dynamic mappings in order
to investigate our approach in a real car scenario.
7 Conclusion
We presented an approach that utilizes a digital in-car game, which translates real-world
misbehavior (i.e., speeding) into in-game interventions. We designed different levels of
in-game interventions and investigated their potentials of acting as a communication
trigger for social persuasion, as well as their impact on fun and engagement. Our approach
fostered communication between driver and front-seat passenger by using different in-
game embodiments of car data to create a connection between the real- and the game
world. Thus, the approach proved to be effective in terms of positive short-term effects
on the speeding behavior of car drivers.
Our study findings illustrate how using a game as a facilitator for social persuasion
and mediatory artifact in a safety critical scenario can be implemented. The respective
game should be dynamic to player skills and preferences as well as dynamic to the current
driving situation and environment (i.e., a hybrid design strategy with dynamic mappings
between in-game changes and driving behavior based on the severity of speeding). Based
on our findings we argue, that there is great potential in using contextual data for creating
new interactive persuasive systems that are designed around social couplings in the car.

Research and Economy and the National Foundation for Research, Technology and Development
is gratefully acknowledged (Christian Doppler Laboratory for Contextual Interfaces).
References
1. Bergen, B.K.: Louder than Words: the New Science of How the Mind Makes Meaning. Basic
Books, New York (2012)
2. Bogost, I.: Persuasive Games: the Expressive Power of Videogames. MIT Press, Cambridge
(2007)
3. Brunnberg, L., Juhlin, O.: Motion and spatiality in a gaming situation–enhancing mobile
computer games with the highway experience. In: Proceedings of Interact 2003 (2003)
4. Forlizzi, J., Barley, W.C., Seder, T.: Where should i turn: moving from individual to
collaborative navigation strategies to inform the interaction design of future navigation
systems. In: Proceedings of CHI 2010. ACM (2010)
5. Gärtner, M., Meschtscherjakov, A., Maurer, B., Wilfinger, D., Tscheligi, M.: Dad, stop
crashing my car!: Making use of probing to inspire the design of future in-car interfaces. In
Proceedings of AUI 2014. ACM (2014)
6. Mayring, P.: Qualitative content analysis. In: Flick, U., von Kardoff, E., Steinke, I. (eds.) A
Companion to Qualitative Research. Sage, London (2004)
7. Perterer, N., Sundström, P., Meschtscherjakov, A., Wilfinger, D., Tscheligi, M.: Come drive
with me: an ethnographic study of driver-passenger pairs to inform future in-car assistance.
In: Proceedings of the CSCW. ACM (2013)
8. Shepherd, J.L., Lane, D.J., Tapscott, R.L., Gentile, D.A.: Susceptible to social influence: risky
“Driving” in response to peer pressure. J. Appl. Soc. Psychol. 41(4), 773–797 (2011)
9. Sjöblom, B.: Gaming as a situated collaborative practice. Human IT: J. Inf. Technol. Stud.
Hum. Sci. 9(3), 128–165 (2013)
Smile Catcher: Can Game Design Lead to Positive
Social Interactions?
Niaja Farve ✉ and Pattie Maes

( )
MIT Media Lab, Massachusetts Institute of Technology, Cambridge, MA 02139, USA

Abstract. Our hectic and increasingly digital lives can have a negative effect on
our health and well-being. Some authors have argued that we socialize less
frequently with other people in person and that people feel increasingly lonely [1].
Loneliness has been shown to significantly affect health and wellbeing in a negative
way. To combat this, we designed a game, Smile Catcher, which encourages
players to engage in in-person, social interactions and get others to smile. Partici‐
pants wear a device that takes regular pictures of what is in front of them and the
system analyzes the pictures captured to detect the number of smiles. The game
was evaluated with a two-week study that sought to determine if game design can
increase positive social interactions and thereby people’s well-being. Our results
show that users increased their frequency of positive interactions as a result of
playing our game.
Contribution Statement: This paper describes a unique game that improves
well-being by encouraging positive social interactions.
Keywords: Happiness · Well-being · Gamification
1 Introduction
Research has shown for years that having a healthy social life is good for well-being
and can double life expectancy [2]. Smile Catcher is a multi-user real world game that
encourages players to catch as many smiles as possible in a single day or session. The
motivation for the game is to have players interact with other people in a positive way,
which hopefully results in an overall increase in their well-being and happiness. To play,
users only need to wear a Narrative Clip [3] or Google Glass that can take pictures at
regular intervals of whatever is in front of the user. At the end of the day, the players
upload their images to be analyzed. The simple implementation of the game allows users
to play with only minor changes in their apparel and daily behavior. While other projects
have tried to log smiles or even force smiles, not many projects have tried to combine
the two in a mobile game for players to enjoy.

DOI: 10.1007/978-3-319-31510-2_18
212 N. Farve and P. Maes
2 Previous Work
Previous research has shown that smiles directly correlate to happiness and can even
incite happiness in a person [4]. This has resulted in several projects that attempt to
measure or force users to smile. All of the projects discussed below primarily focus on
inducing smiles in the user. In contrast, the Smile Catcher project encourages the user
to induce a smile in another person, which means that they benefit not only from
producing a positive affect, but also possibly from emotional contagion [5] and a
strengthening of social connections.
2.1 Mood Meter
The mood meter [6] was a MIT project that displayed mood that was measured in terms
of the number of people smiling in a particular location. Camera systems were placed
in several areas around MIT’s campus. These systems were then used to detect smiles
in community members as they frequented these areas. The data collected was used to
try to raise awareness of how smiles can positively effect an environment. In contrast
with our system, the mood meter captured smiles but did not actively encourage users
to smile.
2.2 HappinessCounter
The HappinessCounter [7] was a project that attempted to force smiles through the use
of an electronic magnet system attached to a fridge. To open the fridge, the user had to
smile. This project was intended for users that lived alone and had limited human inter‐
action. While the HappinessCounter forced smiles and logged smiles, the applications
were limited to static locations. Smiles occurring in areas away from the system were
not logged. In contrast with our system, the HappinessCounter did not involve any
human interactions and dealt with contrived smiles.
2.3 Emotional Flowers

Emotional Flowers [8] is a game that also utilizes the player’s emotion. A user can play
with several people in their social circle. A webcam on each player’s computer monitors
his or her facial expression. Facial expressions classified as happy or surprised allow
their personal flower to grow and change its color. Negative emotions result in their
flowers shrinking. Players attempt to competitively grow the most flowers. Emotional
Flowers also utilizes gamification to encourage happiness, however it does not
encourage positive interactions among players in the game. Besides competing against
each other, there was no advantage to interacting socially. The short interval at which a
smile would be captured resulted in players forcing smiles several times an hour. This
resulted in players quickly becoming tired of having to smile at such frequent time
periods.
Smile Catcher: Can Game Design Lead 213
3 Game Implementation
The hypothesis of our study is that we can improve people’s mood and increase their
number of positive social interactions by making a game of catching smiles they
encounter in their day. For our study, players utilized Narrative Clips to play Smile
Catcher. Game implementation consisted of three parts: image capture, image
processing and feedback. A session is typically played over the course of the day;
however the game can last indefinitely.
On the Narrative Clip, a game or session does not need to be started as images are
taken automatically every 30 s. Proper capture depended on correct placement of the
camera and the moment being captured by the camera. Placement of the camera differed
per person and took some trial and error.
To process the images captured by the Narrative Clip, players use an online appli‐
cation to upload images. Each image is processed using the Face++ [9] API. The images
were analyzed for faces and subsequently smiles. If a smile is present, the image is saved
for the player to review. Other images are not saved as the player already has the image
saved on their personal hard drive and on the Narrative Clip server. The web application
saves their daily counter of smiles and shows trends.
At the end of the day, players are given a score calculated from the day or game
counter (if multiple days). The score is used to help the player gauge the happiness they
have created over a period of time. This score can be sent to others to incite a long-term
competition. Hopefully, this score will become as important as how many steps were
taken or how many calories were consumed during a day.
4 Methodology
4.1 Participants and Methods
Our game was evaluated using a two-week study with thirty-three participants. Partic‐
ipant’s ages ranged from 19 to 67 (14 males and 19 females). The participants were
recruited from a database of willing study participants. Each were told they would be
required to wear the Narrative Clip for two weeks and would upload the images at the
end of each day. So that their first week could be used as a baseline, they were not told
the intention of the study. The only criterion for participation was daily access to a
computer with reliable Internet.
A participant was considered to have successfully completed the study if they fully
participated throughout the two weeks. Twenty participants successfully completed the
study. Full participation consisted of uploading all of the images captured at the end of
each day. Successful participants uploaded at least 100 images each day; ten participants
did not upload a sufficient number of images (less than 100 over the two week period).
Three participants lost their tracker and therefore were unable to participate for the full
two weeks.
4.2 Study Procedures

Prior to starting the study, each participant attended a thirty minute orientation session
where they were given the Narrative Clip, shown how it worked and given an overview
of the website they would be using throughout the duration of the study. Each participant
was given suggestions for how to avoid privacy issues. They were permitted to remove
the device if an individual felt uncomfortable or they were in a private environment.
Each night, participants visited the website to upload their images. The first week of
the study the website was called Life Logger and did not mention anything about smiles.
Participants were asked several questions in regards to their mood (see Fig. 1). Once
they completed the quick survey, they would gain access to a form to upload their images.
During the first week, this was the final step needed to be successfully complete the daily
requirements.
Fig. 1. Baseline week upload page
At the start of the second week of the study the website name changed to be called
Smile Catcher. A smile goal was visible at the top of the website as well as a tracker
Fig. 2. Week 2 upload page

that reflected any past smiles caught. This goal was the same for all participants and set
by averaging daily smile totals from test participants. Participants once again had to
report on their day before uploading their images. After uploading, the users were able
to review five randomly selected smiles caught in previous days and were once more
asked about their mood. After completing the survey, they were shown graphs illus‐
trating any trends in the number of smiles they caught. Participants were not rewarded
for reaching their smile goals (Figs. 2 and 3).
Fig. 3. Smiles caught by participants
At the conclusion of the study each participant attended a thirty-minute closing

session. During this time they returned the Narrative Clips, completed a post survey and
were asked about their experience playing the game.
5 Results
We analyzed the total count of smiles caught to determine if participants had more
positive social interactions during the study. The first week of the study was treated as
baseline. When introducing a new device to users, a novelty effect may occur [10]. Our
participants were not initially aware of the purpose of the study. If a novelty effect were
present, any peak over the baseline week would further confirm that we were able to
influence social behavior.
Evaluating the smiles caught by the twenty successful participants, we were able to
find, with significance (p < .05), that participants caught more smiles during the second
week of the study. Users caught an average of 6.25 smiles during the baseline week and
10.85 smiles during the following week. Those that caught more smiles in the second
week increased the number of smiles caught by 9 images on average, while other partic‐
ipants caught an average of 3.57 less smiles in the second week.
During the closing session most participants were able to report on smiles that were
caught or missed. Participants developed tactics to increase their chances of catching
smiles. Several participants noted making efforts to engage with strangers in an attempt
to gain more smiles. Other participants chose to smile more at individuals hoping they
would reciprocate the gesture.
While users were able to mentally recall smiles in their day, this did not have any
significant effect on their self-reported mood. This may be due to participants only
submitting their mood at the end of the day. Pervious work has shown that recall
of mood can be inaccurate [11]. Our participants may have more accurately submitted
their current mood and not an accurate measure of the day’s mood. While there were no
apparent trends between number of smiles and the mood data collected at the end of the
day, 13 participants commented on how the number of smiles they caught affected them
either positively or negatively.
Not catching smiles was disappointing to many participants; this may be due to
limitations of the system. Due to the nature of the Clip, it was easy for smiles to be
missed. Participants were very aware when a smile was not caught by the camera and
expressed frustration. Misses were caused by error in detection software and simply not
capturing the moment due to the capture rate of the Narrative Clip. This demonstrated
that users were independently searching their images for smiles and that they were also
able to mentally log smiles during the day.
Users found other trends in the images they captured during the day. Several users
commented on the majority of their images being of their computer screens. They were
shocked by the amount of time they spent in front of a computer and the lack of time
spent having human interactions. They also expressed disappointment in the lack of
smiles they caught during the study.
Three participants enjoyed the process of capturing their day and the introspection
it provided that they decided to purchase their own devices so they could continue after
the study ended (Fig. 4).
Fig. 4. Total smiles caught by week
6 Discussion
6.1 User Feedback
Through user feedback we were able to identify several areas where the game could be
improved. It was clear that for long-term adoption accuracy of the game and ease of use
would need to be improved. Accuracy can be improved in further iterations by utilizing

a more accurate API to analyze images. Newer versions of the narrative clip have the
ability to change the image capture frequency. Increasing the capture rate decreases the
chance that a smile would be missed. Outside of the study we would like participants to
be able to play without having to upload their images. However, there is currently no
open API to develop directly with the Narrative Clip.
Women also expressed frustration in trying to find an effective location to wear the
clip. Most examples given by the company highlight men and clothing typically worn
by men on a regular basis. Women tend to not have a flat surface on their chest that the
clip could be easily pinned.
6.2 System Limitations

Using a computer vision algorithm to determine smiles allows for errors in detection
and false positives. Several participants used public transportation daily. Advertising
used on public transportation often uses images of happy people. If this were in the
participant’s field of view, it would result in several “catches”. However these catches
accounted for less than 10 % of all smiles caught and only three participants caught
smiles of this nature. We allowed these to be counted toward the overall total, as
emotional contagion is effective in print materials [12].
6.3 Future Work
While we were able to influence the number of smiles caught by participants, it would
also be interesting to see how their interactions have changed. Do they have more social
interactions overall? Do they target the same people when they try to get their smiles?
Other goals could be provided to users to challenge who they target to get smiles.
7 Conclusion
Through a two-week study we were able to show that gamification could be used to
persuade users to attempt to have more positive social interactions in their day-to-day
lives. By simply tracking the smiles users witnessed during the day and giving a goal,
users significantly changed their behavior. We expected users to attempt to target people
in their social circles when attempting to catch smiles. However, many users reported
engaging with people outside their social circles. We also expected users to be aware of
the positive moments in their day. This proved to be the case, but users were also made
aware of their lack of social interactions. By simply asking users to catch smiles, they
became more cognizant of the smiles in their day-to-day actions. This ability to posi‐
tively recall memories can be linked to an increase in overall mood and gratitude [13]
and therefore an improvement in well-being.
Overall, we were able to positively persuade study participants of all ages to change
how they will interact with individuals inside and outside of their social networks
through the use of very simple mechanisms. While game design was used to encourage
behavior change, participants were not rewarded for completing their goals. We hope
this self-motivation will lend to long-term change. As there is no end to the game, users
simply monitor their progress.
References
1. Coget, J.-F., Yamauchi, Y., Suman, M.: The internet, social networks and loneliness. It Soc.
1(1), 180 (2002)
2. Holt-Lunstad, J., Smith, T.B., Bradley Layton, J.: Social relationships and mortality risk: a
meta-analytic review. PLoS Med. 7(7), 859 (2010)
3. Narrative Clip - The Wearable Camera For Moments That Matter. Narrative. Web. 23 March
2015. http://getnarrative.com
4. James, W.: The principles of psychology, vol. 2
5. Hatfield, E., Cacioppo, J.T., Rapson, R.L.: Emotional Contagion. Cambridge University
Press, Cambridge (1994)
6. Hernandez, J., Hoque, M., Drevo, W., Picard, R.: Mood meter: counting smiles in the wild.
In: UbiComp 2012, pp. 301–310. ACM Press (2012). Dover Publications, New York (1950)
7. Tsujita, H., Rekimoto, J.: HappinessCounter: smile-encouraging appliance to increase
positive mood. In: Alt-Chi: Emotions, Ethics and Civics, CHI 2011 (2011)
8. Using emotion in games: Emotional flowers. In: Proceedings of the International Conference
on Advances in Computer Entertainment Technology
9. Face++- Leading Face Recognition on Cloud. Face++. Web, 23 November 2015. http://
www.faceplusplus.com
10. Kormi-Nouri, R., Nilsson, L.-G., Ohta, N.: The novelty effect: support for the novelty-
encoding hypothesis. Scand. J. Psychol. 46(2), 133–143 (2005)
11. Thomas, D.L., Diener, E.: Memory accuracy in the recall of emotions. J. Pers. Soc. Psychol.
59(2), 291–297 (1990). doi:10.1037/0022-3514.59.2.291
12. Small, D.A., Verrochi, N.M.: The face of need: facial emotion expression on charity
advertisements. J. Mark. Res. 46(6), 777–787 (2009)
13. Watkins, P.C., Grimm, D.L., Kolts, R.: Counting your blessings: positive memories among
grateful persons. Curr. Psychol. 23(1), 52–67 (2004)
More than Sex: The Role of Femininity and Masculinity
in the Design of Personalized Persuasive Games
Marc Busch1 ✉ , Elke Mattheiss1, Michaela Reisinger1, Rita Orji2,

( )
Peter Fröhlich1, and Manfred Tscheligi1,3

1
AIT Austrian Institute of Technology, Vienna, Austria
{Marc.Busch,Elke.Mattheiss,Michaela.Reisinger,
Peter.Froehlich,Manfred.Tscheligi}@ait.ac.at
2
McGill University, Montreal, Canada
[email protected]
3
[email protected]
Abstract. The goal of persuasive games is to change behavior and attitudes in

a desirable manner, e.g., to promote physical activity. Research has shown that
personalized persuasive approaches are more successful than one-size-fits-all
approaches. As a means for personalization, sex has been investigated with results
showing that women are overall more persuadable than men. We argue that
considering only a dichotomous sex-type categorization may not be able to fully
capture the differences in the persuasiveness of persuasion strategies. To that end
we apply a dimensional approach of capturing gender identity – femininity and
masculinity. We investigate the relationship between masculinity, femininity, sex
and the persuasiveness of ten persuasion strategies in an online study (n = 592).
Results show that femininity is significantly associated with seven of the ten
strategies, while sex does only show differences for two strategies, suggesting
gender identity could be a reliable variable for personalizing persuasive games.
Keywords: Persuasive games · Personalization · Gender identity · Sex ·

Femininity · Masculinity
1 Introduction
Persuasive games (PGs) are games that aim at changing human behavior and underlying
attitudes [1]. PGs have been used for achieving several behavioral outcomes, such as
promoting healthy eating [2] and physical activity [3]. PGs designed in a way that take
individual differences into account are more effective than using a “one-size-fits-all”
approach [4]. One way of tailoring persuasive technologies (PTs) to individual differences
is to design for individual susceptibility to certain persuasion strategies, which is called
persuadability [5]. Another popular approach is to tailor PTs based on gender or sex [6].
However, research and design based on gender or sex differences are inherently
biased, as recent studies [7] show that sex/gender differences (e.g., in spatial abilities)
may be more a product of society and education than of biological differences. Although

DOI: 10.1007/978-3-319-31510-2_19
220 M. Busch et al.
sex and gender are often used interchangeably, the world health organization differen‐
tiates the two terms1. While sex refers to “biological and physiological characteristics
that define men and women”, gender refers to “socially constructed roles, behaviors,
activities, and attributes that a given society considers appropriate for men and women”.
This differentiation shows a clear need to depart from a dichotomous classification of
women and men and to establish more enduring differentiable characteristics. In this
paper, we will use the term sex to refer to the dichotomous categorization of men and
women, and the term gender identity, when we refer to a multi-dimensional construct
that includes fine-grained notions on masculinity and femininity.
While sex has been investigated in the area of persuasive games [6, 8, 9], there has
been – to the best of our knowledge – no previous work investigating whether there are
significant relationships between factors of gender identity (such as the degree of femi‐
ninity and masculinity of people) and the persuasiveness of selected strategies. If such
relationships exist, gender identity could meaningfully contribute to the understanding
and the personalization of persuasive games.
We selected the same persuasion strategies that have been investigated by Orji et al. in
the context of a persuasive game to foster healthy eating [8, 9] and let 592 participants
evaluate the persuasiveness of these strategies in an online study. We assessed partici‐
pants’ sex as well as femininity and masculinity by using a short version of these scales
that are included in Bem’s Sex Role Inventory [10], a self-assessment questionnaire
designed to assess the degree of femininity and masculinity of a person. We hypothesized
that the more fine-grained notion of femininity and masculinity can provide a better under‐
standing of the persuasiveness of persuasion strategies than biological sex recorded as
female or male alone. Our results show that femininity and masculinity affect the persua‐
siveness of more strategies than sex does and that sex differences might not be a reliable
predictor of persuasiveness of persuasion strategies or are strongly depending on the target
behavior. We investigated very similar relationships as Orji et al. [8, 9], but have different
results.
Our contribution is twofold: First, in opposite to self-reported biological sex, we
introduce femininity and masculinity into the domain of persuasive games, and second,
we show that accounting for femininity and masculinity can meaningfully inform
personalized persuasive game design. The rest of the paper is organized as follows: We
start with a review of related work in the area of sex and gender studies and how they
account for differences in the effectiveness of PTs and PGs. We describe our online study
procedure and the results. We contrast our results to the results of Orji et al. [8, 9] and
give an outlook into the future of personalized persuasive game design based on gender
identity.
2 Previous Work
Sex and gender are complex constructs that encompass a number of sub dimensions:
measurements of sex can include assessment of bodily characteristics, fetal testosterone
1
http://apps.who.int/gender/whatisgender/en/index.html.
More than Sex: The Role of Femininity and Masculinity 221
markers, chromosomal or hormonal status. Many studies have investigated how and
when to measure sex, and what levels of assessment could support defining differences
between participants and groups of participants. Gender identity refers to the personal
conception of oneself – a conception that is culturally derived. This often refers to
masculine and feminine personality traits [e.g., 11, 12], though some employ a broader
definition of gender identity, which includes those gendered personality traits as one
factor (multifactorial gender identity theory).
Only few authors explicitly discuss routine sex/gender items and different opera‐
tionalizations of sex and gender in questionnaires [e.g., 13]. Sex and gender are usually
assessed through single item measures, within which participants are asked to indicate
their sex/gender by checking “male” or “female” (or, sometimes used for the latter,
“masculine” or “feminine”) and which results in a dichotomous variable. A strong
connection between sex and gender is often presumed – so extensively, that they are
used synonymously, or one is consciously used to account for the other.
Since items like “are you male/female?” can be associated with “biological sex” (i.e.
concepts the individual holds thereof) as well as gender (identity), they are hardly unam‐
biguous. It has also been noted that male/female are neither exclusive nor exhaustive
categories, which would also be necessary to use them as nominally scaled data [13].
Using psychometric scales to assess gender identity on the other hand is much less
common. Of these scales, the Bem Sex Role Inventory (BSRI) [10] and its shortened
versions, e.g., the Gender Traits Index (GTI) [14] are most frequently used. For an
overview of other scales for the assessment of gender see Döring [13]. Originally, the
BSRI has three subscales (femininity, masculinity and androgyny) and consists of 60
items. Identification with these attributes categorizes individuals across four categories:
feminine, masculine, androgynous and undifferentiated. In this research we did not want
to assign people to categories (similar to assigning them to sex categories), instead we
wanted to understand the finer grained notion of gender identity regarding masculinity
and femininity in relation to persuasion strategies. To achieve this, we adopted a shorter
version of the BSRI with ten items of the femininity and masculinity scales each that had
the highest item-scale correlations [15].
Sex/Gender have been shown to influence interests and behavior [2, 16]. Both
constructs also play a role within persuasion pathways – persuasiveness can be moder‐
ated by gender similarity of human as well as non-human interaction partners [17, 18],
and message pathway preferences differ according to sex/gender [19–21]. Gender (no
details on assessment given) and age can impact the effectiveness of gain- or loss-framed
messages promoting physical activity, with gender furthermore acting as moderating
factor in promoting physical activity [22].
In the last years, personalization of PTs and PGs has been in the focus of research
[23]: For example Kaptein et al. [24] showed that in the health promotion domain,
tailored messages were more successful in persuading people to perform more phys‐
ical activity or to eat more healthy. Investigating the persuasiveness of selected
persuasion strategies from Fogg [25] and Oinas-Kukkonen and Harjumaa [26], Orji
et al. [8] found that females evaluate the strategies cooperation, customization,
personalization, praise and simulation as significantly more persuasive. Investi‐
gating sex differences in the strategies by Cialdini [27], Orji et al. [6] found that
222 M. Busch et al.
females evaluate the strategies reciprocity, commitment and consensus as signifi‐

cantly more persuasive. Regarding a model of persuasiveness, Drozd [28] found that
the relationships between several variables in that model (e.g., unobtrusiveness of
technology and intention to use) was the same for women and men.
We aim to combine and advance these different research streams: we apply an
existing method from gender studies (scales from BSRI) to explore gender identity as a
possible new meaningful factor for the personalization of PGs.
3 Method
To answer the research question whether femininity and masculinity as measured

through a short version of the BSRI can contribute more to the understanding of persua‐
siveness of persuasion strategies than sex alone, we conducted an online study with 592
participants. A link to the online study was distributed to a database consisting of people
volunteering to participate in research studies.
3.1 Procedure of the Online Study

The procedure of the online study was as follows: After a short introduction into the
goal of the study (the design of videogames for promoting physical activity), participants
saw the persuasion strategies, which were presented as storyboards. Storyboards are
considered to be a useful tool in product design to communicate concepts in a visually
appealing way [29] and they have also been successfully used in prior work in the area
of persuasive technology [8, 9]. In this regard, storyboards have been used to opera‐
tionalize the persuasion strategies and to empirically compare the persuasiveness of
these strategies to ultimately decide which persuasion strategies should be implemented
in real systems. We chose the same persuasion strategies as Orji et al. [8, 9], which were
taken from Fogg [25] and Oinas-Kukkonen and Harjumaa [26], but investigated all of
them separately (e.g., to be able to have results for these strategies separately, we did
not investigate competition and comparison in one storyboard). Each of the ten persua‐
sion strategies that we investigated in this study was operationalized in one storyboard.
Every participant saw each persuasion strategy one by one in a random order.
After each storyboard, participants had to answer the following three questions on
a 7-step Likert-type scale ranging from Totally Disagree to Totally Agree:
• This approach is fun.
• This approach would help me to be more aware of my physical activity.
• This approach would motivate me to do more physical activity.
From the three items, one persuasiveness score per persuasion strategy was calcu‐
lated. Next, gender identity was assessed with a short version of the BSRI containing
20 adjectives (e.g., “strong”). There are ten adjectives that are more associated with
women and ten adjectives that are more associated with men. On a 7-step Likert-type
scale, participants were asked to decide how much each adjective applied to them. From
these ratings, we calculated an overall score for masculinity and one for femininity for
each participant.
We report Cronbach’s Alpha as a measure for internal consistency of the scales used
in this study: Femininity has an Alpha of .83, masculinity has .86 and the mean Alpha
of the persuasiveness scales is .93. These values are satisfying for internal consistency,
which allows us to calculate and interpret overall scores from the single items.
At the end, participants were asked basic demographic variables such as biological
sex and age as well as interest in video games.
3.2 Storyboard Development
In the following section, we describe how we designed the storyboards used in this study.
Although the design of the storyboards was inspired by Orji et al. [8, 9], the storyboards
in this study were modified as follows: The gamer shown in the storyboards of Orji
et al. [8, 9] was always female. Instead of showing a person which can be identified as
either female or male, we used a sketch/outline of a hand to account for potential biases.
To ensure internal validity of this study, we checked the following quality criteria in the
(message) design of the storyboards:
• The storyboards should only vary in the persuasion strategies, not in the level of
detail, design fidelity or other factors.
• The storyboards should clearly represent one and only one strategy.
• The storyboards should comprehensively relay the persuasion message.
We evaluated and iteratively refined the persuasion strategies in two subsequent
expert interviews with HCI experts in a research organization. Initial interviews
conducted with five experts led to a first design iteration, further interviews with two
different experts led to a second design iteration. An example for a storyboard repre‐
senting the persuasion strategy reward is depicted in Fig. 1.
Fig. 1. Storyboard representing the persuasion strategy Reward
Graphically, all ten storyboards looked the same. What differed was the text on the
left and on the right picture. This is a list with all texts of all ten storyboards:
• Customization: Indicate, how your results regarding running should be displayed in
the game – Settings: o Speed/o Distance/o Burned calories.
• Personalization: Enter some information about you: Age, Weight – Your training
plan will be personalized accordingly.
224 M. Busch et al.
• Praise: You managed to run more than 20 km in the last week – Very good! That is
quite a lot!
• Self-Monitoring: You ran 21 km in the last week – You improved: the week before
it was 19 km.
• Simulation: The number of km that you ran has increased from week to week – A
sneak peek into the future: if you continue with your progress, you will run 30 km
per week in two months.
• Suggestion: To improve your running performance, here is a tip – Try to walk some
distances occasionally during running. Doing that, you manage longer distances.
• Comparison: You ran 21 km in the last week – Alex ran 23 km.
• Competition: In the next week you can enter a competition against Alex – Will you
manage to run more km than Alex?
• Cooperation: In the next week Alex’ and your results will be added – Can you
manage to run over 75 km together?
4 Sample and Results
Sample. The sample (n = 592) indicated a medium interest in video games (M = 3.77;
SD = 2.17). Participants’ age ranged from 14 to 80 (M = 35.14; SD = 14.06) and 272
(46 %) assessed themselves as female.
Distribution of Persuasion Strategies. To get an overall view on how persuasive the

single persuasion strategies were rated, Fig. 2 presents notched boxplots. The notch
represents the 95 % confidence interval of the median and allows estimating significant
differences between the strategies regarding their persuasiveness. Customization,
personalization, reward, self-monitoring and simulation have significant higher persua‐
siveness values than the neutral mid-point of the persuasiveness scale. Comparison has
a significant lower value than the neutral mid-point.
Fig. 2. Evaluation of the persuasiveness (y-axis) of the ten persuasion strategies (x-axis) on a
Likert-type scale ranging from 1 to 7 (higher numbers indicate a higher persuasiveness; the black
line at 4 indicates the neutral mid-point).
Distribution of Masculinity and Femininity and Influence of Sex. Masculinity

(M = 4.92; SD = .91) and femininity (M = 4.89; SD = .85) are similarly distributed.
Femininity and masculinity show a medium correlation of .16. According to two-

sample t-tests, self-reported sex has a significant influence on femininity,
t(585) = 5.81, p < .001, d = .46, with women describing themselves as more femi‐
nine (M = 5.13; SD = .77) as men (M = 4.75; SD = .87). Self-reported sex has not a
significant influence on masculinity, t(567) = −1.58, p = .12, d = .13, however men
described themselves slightly more masculine (M = 4.94; SD = .89) than women
(M = 4.82; SD = .92).
Table 1. Influence of femininity, masculinity and sex on persuasion strategies (it is indicated
in bold when a gender identity factor has a significant influence on persuasiveness of a strategy).
Strategy Regression Statistic Influence of feminini-

ty/masculinity
Customization R2=.01a , F(3,584)=3.27, p<.05 Femininity (β=.19, p<.05)
Masculinity (β=-.05, p=.53)
Sex (β=-.19, p=.16)
Personalization R2=.01, F(3,584)=3.34, p<.05 Femininity (β=.22, p<.01)
Masculinity (β=.00, p=.97)
Sex (β=-.15, p=.28)
Praise R2=.01, F(3,584)=5.52, p<.05 Femininity (β=.24, p<.01)
Sex (β=-.12, p=.41)
Reward R2=.01, F(3,584)=3.47, p<.05 Femininity (β=.26, p<.01)
Sex (β=-.27, p=.07)
Self-Monitoring R2=.00, F(3,584)=2.08, p=.10 Femininity (β=.20, p<.05)
Sex (β=-.05, p=.74)
Simulation R2=.01, F(3,584)=2.67, p<.05 Femininity (β=.22, p<.01)
Sex (β=-.05, p=.70)
Suggestion R2=.00, F(3,584)=1.91, p=.13 Femininity (β=.18, p<.05)
Sex (β=-.04, p=.76)
Comparison R2=.02, F(3,584)=6.774, p=.00 Femininity (β=.15, p=.07)
Sex (β=.60, p=<.001)
Competition R2=.03, F(3,584)=7.101, p=.00 Femininity (β=.20, p<.001)
Sex (β=.59, p<.001)
Cooperation R2=.00, F(3,584)=1.607, p=.19 Femininity (β=.15, p=.07)
Sex (β=-.24, p=.10)
a
Adjusted R2
226 M. Busch et al.
Influence of Femininity, Masculinity and Sex on Persuasion Strategies. To answer

our main research question – whether femininity and masculinity can contribute more
to the understanding of the persuasiveness of persuasion strategies than sex– we calcu‐
lated multiple regressions. We chose not to perform multivariate analyses, as we had no
hypotheses about the interdependencies of the persuasion strategies, but only about the
influence of the predictor variables (femininity, masculinity, sex) on the outcome vari‐
able (persuasiveness of a persuasion strategy). The results are displayed in Table 1. In
eight out of ten strategies, femininity has a significant positive influence on the subjective
persuasiveness. The more feminine a person describes him-/herself, the more persuasive
he/she rates the persuasion strategies customization, personalization, praise, reward,
self-monitoring, simulation and suggestion and competition. Regarding the influence of
sex: Men rate the strategies comparison (men: M = 3.99, SD = 1.64; women: M = 3.44,
SD = 1.71) and competition (Men: M = 4.09, SD = 1.65; Women: M = 3.57, SD = 1.77)
as significantly more persuasive than women.
5 Summary and Discussion
Persuasive games (PGs) can be successful tools in changing behavior and attitudes
towards specific desirable goals. For several years now, researchers and designers have
aimed at defining and discovering methods for personalizing PGs, as personalized PGs
has been shown to be more effective in motivating behavior change than one-size-fits-
all approaches. To add to and extend existing work on tailoring persuasive technology
to individual characteristics including sex and gender, we explored whether femininity
and masculinity (as a dimensional construct) can be a useful and more predictive char‐
acteristic for tailoring persuasion strategies than sex alone.
To that end, we conducted a large-scale study with 592 participants to evaluate the
persuasiveness of 10 commonly employed strategies in PG design. Participants did a
self-assessment on both femininity and masculinity scales (short version of BSRI). Our
results show that femininity and masculinity can indeed provide more insight into the
persuasiveness of persuasion strategies than sex alone: In our study, men find compar‐
ison and competition significantly more persuasive than women. This finding reflects a
common male tendency in our society, which is reproduced in our results. However,
considering the gender identity factors masculinity and femininity, our results show that
participants (regardless of sex) who self-assessed as more feminine find the strategies
customization, personalization, praise, reward, self-monitoring, simulation, suggestion
and competition more persuasive. Masculinity does not have a significant effect on the
perceived persuasiveness. Overall, gender identity factors femininity and masculinity
mediates the perceived persuasiveness of more strategies than sex.
We have a finding that might appear counterintuitive: Men find competition signif‐
icantly more persuasive than women while femininity (but not masculinity) is positively
associated with the persuasiveness of competition. However, we argue that this finding
shows that sex and gender identity – although related – are not the same construct, they
reflect different factors and cannot be used interchangeably.
Since the study by Orji et al. [8, 9] investigated the same persuasion principles as
we did, we expected similar results. However, our results stand partly in contrast to their
study, where women rated the persuasiveness of cooperation, customization, personal‐

ization, praise and simulation significantly higher than men (which is reflected in our
finding on the effect of femininity), while for comparison and competition no sex effects
were shown. Also, Orji et al. [6] showed that for the Cialdini persuasion principles [27],
women are in general more susceptible to persuasion than men. There are several
explanations for such different results: Their operationalization of the persuasion strat‐
egies differed slightly (e.g., they investigated competition and comparison in one story‐
board). Furthermore, while Orji’s et al. study [8, 9] was targeted at promoting healthy
eating, the current study was targeted at promoting physical activity. Although both
target behaviors referred to healthy living and lifestyle, the nutrition topic could be more
appealing to women, while the physical activity topic could be more appealing to men,
which could in turn have influenced the perceived persuasiveness of persuasion strat‐
egies in these topics. This leads to an important conclusion: The application domain in
which the technology is applied is an important determining factor of its effectiveness.
Furthermore, also potential predictors of the effectiveness of persuasive technology
(e.g., sex, gender identity) are mediated by the behavior that should be changed.
In a broader sense, the influence of sex (and gender) on the persuasiveness of persua‐
sion strategies could depend on contextual and situational factors. Under this consider‐
ation it is not only necessary to personalize PTs and PGs – assuming that a strategy that
persuades a person might not persuade another person - but also to contextualize them,
assuming that a strategy that persuades in a specific context or situation might not
persuade in another specific context or situation.
Surprisingly, sex was only significantly associated with femininity, not with mascu‐
linity. This could be attributed to the instrument with which femininity and masculinity
was measured: although the Bem Sex Role Inventory (BSRI) [10] is one of the most
widely used inventories to assess gender identity, it is possible that other (psychometric)
operationalization methods of gender identity might produce other results, which is a
limitation of this study.
This paper presents the first exploratory study on gender identity, sex and persuasion
strategies for persuasive games (PGs). Our results have the following implications for
the personalization of PGs: The sex of the target audience, but also how feminine they
describe themselves, can influence the effectiveness of persuasion strategies. Taking
into consideration that different sexes and gender identities might perceive persuasion
strategies differently depending on target behavior (e.g., healthy eating, physical
activity), strategies should be selected with care and on the basis of empirical data: When
promoting physical activity through PGs, strategies comparison and competition are
more persuasive for men than women; strategies customization, personalization, praise,
reward, self-monitoring, simulation, suggestion and competition are more persuasive
the more the target audience describes themselves as feminine.
To understand the underlying mechanisms of these effects future work has to
critically explore the interactions between societal images of men and women and
228 M. Busch et al.
how this affects their perception of PGs. We have to investigate if and to which
extent the findings in this and related studies [9] can be generalized to other domains
(e.g., personal transportation, security) and target behaviors. Additionally, the
stability of the findings needs investigation: Long-term studies under realistic condi‐
tions will shed more light on the interrelations between gender identity, sex and
persuasiveness of PGs.
Acknowledgements. This research has partly been funded by the Austrian Research Promotion
Agency under contract no. 844845 (GEMPLAY).
References
1. Bogost, I.: Persuasive Games: The Expressive Power of Videogames. MIT Press, Cambridge
(2007)
2. Orji, R., Vassileva, J., Mandryk, R.L.: Modeling gender differences in healthy eating
determinants for persuasive intervention design. In: Berkovsky, S., Freyne, J. (eds.)
Persuasive 2013. LNCS, vol. 4128, pp. 161–173. Springer, Heidelberg (2013)
3. Berkovsky, S., Freyne, J., Coombe, M.: Physical activity motivating games. ACM Trans.
Comput. Interact. 19, 1–41 (2012)
4. Orji, R., Mandryk, R.L., Vassileva, J., Gerling, K.M.: Tailoring persuasive health games to
gamer type. In: Proceedings of the SIGCHI Conference on Human Factors in Computing
Systems - CHI 2013, pp. 2467–2476. ACM Press, New York (2013)
5. Kaptein, M.C.: Personalized persuasion in ambient intelligence. J. Ambient Intell. Smart
Environ. 4, 279–280 (2012)
6. Orji, R., Mandryk, R.L., Vassileva, J.: Gender, age, and responsiveness to Cialdini’s
persuasion strategies. In: MacTavish, T., Basapur, S. (eds.) PERSUASIVE 2015. LNCS, vol.
7. Hoffman, M., Gneezy, U., List, J.A.: Nurture affects gender differences in spatial abilities.
Proc. Natl. Acad. Sci. 108, 14786–14788 (2011)
8. Orji, R.: Exploring the persuasiveness of behavior change support strategies and possible
gender differences. In: Öörni, A., Kelders, S., van Gemert-Pijnen, L., Oinas-Kukkonen, H.
(eds.) Second International Workshop on Behavior Change Support Systems, CEUR,
vol. 1153, pp. 41–57 (2014)
9. Orji, R., Mandryk, R.L., Vassileva, J.: Gender and persuasive technology: examining the
persuasiveness of persuasive strategies by gender groups. In: Gamberini, L., Spagnolli, A.,
Chittaro, L., Zamboni, L. (eds.) Persuasive 2014, Adjunct Proceedings, pp. 48–52. University
of Padova, Padova (2014)
10. Bem, S.L.: Bem Sex-Role Inventory: Professional Manual. Consulting Psychologists Press,
Palo Alto (1981)
11. Palan, K.M.: Gender identity in consumer behavior research: a literature review and research
agenda. Acad. Mark. Sci. Rev. 2001, 1–24 (2001)
12. Tate, C.: The “problem of number” revisited: the relative contributions of psychosocial,
experiential, and evolutionary factors to the desired number of sexual partners. Sex Roles
64, 644–657 (2011)
13. Döring, N.: Zur Operationalisierung von Geschlecht im Fragebogen: Probleme und
Lösungsansätze aus Sicht von Mess-, Umfrage-, Gender- und Queer-Theorie. Gend. –
Zeitschrift für Geschlecht, Kult. und Gesellschaft. vol. 5, pp. 94–113 (2013)
14. Barak, B., Stern, B.: Sex-linked trait indexes among baby-boomers and pre-boomers: a
research note. Adv. Consum. Res. 13, 204–210 (1986)
15. Schneider-Düker, M., Kohler, A.: Die Erfassung von Geschlechtsrollen: Ergebnisse zur
deutschen Neukonstruktion des Bem Sex-Role Inventory. Diagnostica 34, 256–270 (1988)
16. Vilela, A.M., Nelson, M.R.: Testing the selectivity hypothesis in cause-related marketing
among generation Y: [when] does gender matter for short- and long-term persuasion? J. Mark.
Commun. 1–18 (2013)
17. Siegel, M., Breazeal, C., Norton, M.I.: Persuasive robotics: the influence of robot gender on
human behavior. In: IROS 2009, IEEE/RSJ International Conference on Intelligent Robots
and Systems, pp. 2563–2568. IEEE Press, New York (2009)
18. Mellor, S., Barclay, L.A., Bulger, C.A., Kath, L.M.: Augmenting the effect of verbal
persuasion on self-efficacy to serve as a steward: gender similarity in a union environment.
J. Occup. Organ. Psychol. 79, 121–129 (2006)
19. Guadagno, R.E., Cialdini, R.B.: Persuade him by email, but see her in person: online
persuasion revisited. Comput. Human Behav. 23, 999–1015 (2007)
20. Vossen, S., Ham, J., Midden, C.: Social influence of a persuasive agent: the role of agent
embodiment and evaluative feedback. In: Persuasive 2009, pp. 46–52. ACM Press, New York
(2009)
21. Ruijten, P.A.M., Midden, C.J.H., Ham, J.: Lonely and susceptible: the influence of social
exclusion and gender on persuasion by an artificial agent. Int. J. Hum. Comput. Interact. 31,
832–842 (2015)
22. Li, K.-K., Cheng, S.-T., Fung, H.H.: Effects of message framing on self-report and
accelerometer-assessed physical activity across age and gender groups. J. Sport Exerc.
Psychol. 36, 40–51 (2014)
23. Kaptein, M., Markopoulos, P., de Ruyter, B., Aarts, E.: Personalizing persuasive
technologies: explicit and implicit personalization using persuasion profiles. Int. J. Hum.
Comput. Stud. 77, 38–51 (2015)
24. Kaptein, M., Lacroix, J., Saini, P.: Individual differences in persuadability in the health
promotion domain. In: Ploug, T., Hasle, P., Oinas-Kukkonen, H. (eds.) PERSUASIVE 2010.
25. Fogg, B.J.: Persuasive Technology - Using Computers to Change What We Think and Do.
Morgan Kaufmann Publishers, San Francisco (2002)
Øhrstrøm, P. (eds.) PERSUASIVE 2008. LNCS, vol. 5033, pp. 164–176. Springer,
Heidelberg (2008)
27. Cialdini, R.B.: Harnessing the science of persuasion. Harv. Bus. Rev. 79, 72–81 (2001)
28. Drozd, F., Lehto, T., Oinas-Kukkonen, H.: Exploring perceived persuasiveness of a behavior
change support system: a structural model. In: Bang, M., Ragnemalm, E.L. (eds.)
29. van der Lelie, C.: The value of storyboards in the product design process. Pers. Ubiquit.
Comput. 10, 159–162 (2006)
A Gamified Solution to Brief Interventions
for Nightlife Well-Being
L. Gamberini1(&), A. Spagnolli1, M. Nucci1, G. DeGiuli2, C. Villa2,

V. Monarca2, A. Privitera1, L. Zamboni2, and S. Leclerq2,3
1
HtLab, Department of General Psychology, University of Padua, Padova, Italy
[email protected]
2
Psicologi Senza Frontiere Onlus, Padova, Italy
3
ABD - Energy Control, Barcelona, Spain
Abstract. This paper addresses a specific type of intervention for health and
wellbeing, i.e., brief interventions during nightlife events, aimed at improving
partygoers’ awareness of the risks related to the consumption of psychoactive
substances. It is argued that a gamified modality would help overcome some
limiting contextual constraints while preserving the seriousness of the inter-
vention purpose. To support this last point, the paper describes two studies
(N = 227, N = 81 respectively). The first study was conducted in the field; it was
found that after a game session, users’ self-reported awareness of nightlife risks
increased significantly from pre-session scores. In addition, users evaluated the
intervention as credible in terms of its accuracy, novelty, and contextual
appropriateness. The second study showed, under controlled laboratory condi-
tions, that the number of correct answers about substance consumption conse-
quences, substance characteristics, and risk prevention choices improved after
the game session. Together, these studies suggest that using gamification as an
attractor would not compromise the serious goals of this intervention modality.
Keywords: Well-being intervention Risk awareness Nightlife Serious

games
1 Introduction
Substance consumption is so common in recreational venues such as clubs, festivals,

raves, or parties [9] that the terms “club drugs” has emerged to identify a set of
substances commonly associated with raves and parties. A recently survey in nine cities
across Europe reported that only 9.8 % of the interviewees were nonusers [6]. Practices
such as binge drinking [17] or polydrug abuse [31] are not restricted to deviant groups
but are diffuse in the young population, supported by specific cultures and lifestyles
[18]. The abuse or misuse of recreational drugs can generate serious health problems,
jeopardize relationships, and lower performance in activities such as driving [5].
Interventions range from prevention programs addressing the community, law
enforcement, and alcohol sellers to interventions carried out in nightlife settings. The
latter are brief [23] and devoted to sensitizing to the need for acquiring better infor-
mation or changing habits. They generally employ leaflets and infocards [3]. On the

DOI: 10.1007/978-3-319-31510-2_20
A Gamified Solution to Brief Interventions 231
one hand, organizing brief interventions in nightlife settings allows people to reach
consumers of recreational drugs where they are likely to perform risk behaviors. On the
other hand, this context makes the intervention difficult because it might seem
inconsistent with the situational cultures and values and because the hectic environment
hampers sustained attention. We built on the opportunities offered by digital technol-
ogy to overcome some of the contextual constraints and developed a multiplayer digital
quiz projected on a large screen to deliver prevention information during such inter-
ventions. Whereas noise, crowd, and dominant mood during these events can deter
serious information-seeking about prevention behavior, these features might turn into
assets if the intervention takes place via a game.
Digital environments for health treatment designed as full-fledged games or that
include game elements such as scores, competition among players, roleplaying, chance,
rules, goals, levels, rewards, and narratives [8] have added persuasive value [15]. They
exploit fun and entertainment to increase the users’ motivation to undertake and
commit with the intervention, and emphasize the progression of challenges and
achievements that boosts the users’ self-efficacy by overcoming a progressive set of
challenges [e.g., 20]. In addition, they can appeal to classes of users that are particularly
attracted or accustomed to digital technology. The World Health Organization
acknowledges that the “widespread appeal of computer and videogame playing among
children and adults creates a unique opportunity to deliver health education during
leisure time” [21, p. 27].
A few examples of interventions directed to young adults to prevent drug abuse and
supported by serious games are in the literature. “Reach Out Central” [4] and “Climate”
[35] are school-based interventions, which require a different stylistic register and level
of explicitness than interventions at parties, are often longer than one session, and do
not focus on nightlife and its specificities. “Spot the Triggers Challenge,” [24, North]
targets alcohol and drug abuse and aims at addiction treatment through a structured
program requiring multiple sessions. Closer to our purpose are Happy Farm [13] and
VideoDope [12]; the first relies on an interactive storytelling paradigm, which is too
complex for nightlife interventions; the latter is limited in scope, being focused on
substance effects on driving.
Compared with the existing computer-based solutions to deliver prevention infor-
mation in this domain, our gamified intervention has a unique combination of char-
acteristics: it is a brief intervention specifically targeting nightlife events, targets a large
set of nightlife behaviors, and is especially designed for use in nightlife settings (e.g.,
multiplayer interface, easy rationale, party aesthetics). Although serious games [2] can
be used for health information in general [7, 27, 32, 33], we believe a crucial point in
gamified applications is proving that they do not spoil the serious purpose of the
intervention, in each context in which they are applied. In this paper, we report two
studies with the purpose of assessing whether this was the case with our solution. We
will start by describing the game and then report the two validation studies.
232 L. Gamberini et al.
2 Game
The intervention presented here represents the gamification of a typical nightlife

intervention, which is usually brief [23], with an information area in a relaxed zone of
the party, and expert operators—most often peer operators—available to hand out
prevention material and provide information. In our intervention, information and
advice reside in a digital database and their provision occurs through quiz questions.
This is then a serious game [2], in that it exploits for serious purposes the motivational
characteristics of a game to provide serious prevention information. The full version of
the game contains 4,032 questions in five different languages (English, Italian, German,
French, and Spanish), including 1,193 multiple choice and 2,839 true/false questions.
We categorized questions according to the 11 risk areas1 and 13 topics2.
The game logic includes a set of steps: a training video showing the use of the
remote control and the basic rationale of the game (Fig. 1); avatar selection; the set of
quiz questions and related explanations; and a player’s ranking screen. Questions are
randomly selected from the database. Players can select their answer via the controller;
after a predefined time, the correct answer is displayed with an explanation and the
players’ scores are updated and displayed. Players can modify and set some features
before each intervention: the language, difficulty level, number of participants, and time
to answer each question.
Fig. 1. (a, b). A screenshot from the game interface, where the scores of the players are shown
along with the correctness of the last response provided and the difficulty level of the last quiz
question (a, left); an image from one field intervention (b, right).
The base system is composed of several modules, each dealing with a different
aspect of the game.
1
Alcohol, Cannabis, Ecstasy and Methamphetamines, Hallucinogenic Substances, Ketamine,
GHB-GBL, Popper and other Solvents, Cocaine, Heroin, Friendship & Sexuality-Mix.
2
Desired Effects, Not Desirable Acute Psycho Effects, Not Desirable Long-Term Psycho Effects, Not
Desirable Physical/Medical Effects, Legal Issues, Historical-Political-Geographical Issues,
Curiosities-Myths and Legends, Woman Specificities & Pregnancy, Driving, Sex & Sexually
Transmitted Diseases, Other Indirect Risks, Violence-Bullies-Micro Interethnic Conflicts, First Aid
and Precautions.
The data management module contains the quiz questions and the data collected by
players’ data (ID, date, avatar, scores, and response time). The scoring system module
is responsible for storing and correctly calculating the scores depending on the answers
selected by the player and on the question category. The event triggering module
receives messages from the controller’s handler and interacts with the game logic. The
controller handler collects inputs from up to six players’ controllers and converts them
into messages to the base system. It also contains a console for starting the game and an
anti-piracy system for preventing the improper use of the software.
Technically, the base system was developed with Web application technology,
using the Ruby on Rails framework for development and SQLite as a database for
storing all data involved (questions, scoring system, user data collection, and level
data). This technology permits easy delivery of the game to the most used platforms
and the Web. The controller handler is written in is written in C#, using a .NET
framework. The data management module is a SQLite Database. Game controller
support in Web platforms is at an early stage of development and not yet implemented
in all browsers. For this reason, we designed and developed this module as a separate
piece of software, in C# using Microsoft DirectInput Library. It runs on Microsoft
Windows. We created audio content using Ableton Live8. We tested the game for
usability before the studies described here.
2.1 Additional Design Strategies to Improve Persuasion

We benefited from the specific persuasive characteristic of a game to attract users, but
put great emphasis during design into ensuring that the game was credible and con-
textually appropriate, which are preconditions to persuasion [11, 26].
Credibility required accurate information and institutional or scientific sources for
the questions. We built the database by collecting and examining scientific publications
and prevention material produced by reliable sources. Experts double-checked each
piece of information and turned the pieces into questions in various formats (e.g.,
multiple choice, true/false, and numeric entry). A component of credibility is the
perception that the information is unbiased [e.g., 14]. Indeed, in the specific context of
nightlife interventions and health communication, International guidelines [9, 30, 35]
stress the need that the recipient perceives the intervention as consonant with his/her
prior experience. Thus, our conceptual approach is not to question users’ repertoire of
values framing certain risky behaviors, which are well rooted in their culture [16, 19],
but to appeal to alternative values still consistent with their goals during their nights
and parties: well-being and friendship. To be contextually appropriate, the game
appearance was curated so as to fit the aesthetics of nightlife events. Two professional
designers developed graphics, avatars, and music especially for the game, close to
those from nightlife parties. The soundtrack changed according to the category of the
question; for instance, the music had a fast pace for exciting substances, a reggae
rhythm for marijuana, and a soft melody for sexuality. For outdoor interventions (such
as during summer music festivals), a gazebo was equipped with an information desk
and projection wall and decorated according to party-like aesthetics.
In addition, the game used illustrated narratives (Fig. 2), which increase recall in
health communication [22, 29] and allowed some tailoring (selection of topic cate-
gories, kind of substance addressed in the questions, questions’ difficulty, and time to
respond before each intervention), which increases persuasiveness [11].
Fig. 2. An instance of an illustrated narrative accompanying a question, here in the form of a

statement, which players had to determine was true or false.
The whole design process was participatory [1], which maximizes the probability
that the design fit the context of use and the target users’ perspective and was in line
with the recommendation for persuasive technologies to build upon users’ own moti-
vation or goal [25]. This entailed a brainstorming session with 21 European peer
operators and experienced nightlife operators; five workshops with peer operators,
experienced operators and minors with legal issues related to drug consumption. Each
workshop involved 8–10 participants to produce narratives about the risky situations,
from initial brainstorming to shooting the scenes; experts participating in the European
network Newnet (www.safernightlife.org) to localize the questions.
3 Study 1: Risk Awareness and User Experience
The first validation study reported here was conducted in the field. This forced us to
keep the validation instrument as short as possible to avoid dropouts. We hypothesize
that after using the game, users’ awareness of nightlife risks increased significantly
compared with the pre-session scores (H1). We also hypothesized that the scores of the
user experience items were positive and significantly higher than the central value of
the response scale (H2). We ran analyses with SPSS.
3.1 Method
Setting. The intervention was carried out within two events in Italy. The former took
place during the summer; our intervention took place along three days on 4–7
July 2013 and was located beside a park where a dance party was organized with
famous radio deejays targeting young adults. Since the event was outdoors, we pro-
jected game images onto a wall of a specially designed gazebo. Players could sit in a
row of chairs in front of it. The second event was a techno music event with a program
consisting of several concerts. We carried out the intervention during three concerts in
2013, on October 5, October 31, and November 23. In this case, the event was indoors
and our intervention took place in a space (the “chill-out” zone), where attendees could
find less noise, seating, and information. All interventions took place in agreement with
the event organizers.
Questionnaire. This intervention used a brief questionnaire, since long, articulated
questionnaires were not possible in these conditions. It was a paper-and-pencil ques-
tionnaire with pre- and post-session parts. The pre-session questionnaire contained
demographic items asking the participants’ age, gender, education, and nationality, as
well as two statements assessing risk awareness, which could be answered on a 6-degree
Likert scale where 6 meant “full agreement” (Table 1). The post-session questionnaire
included the same two statements as the pre-session questionnaire in addition to 10
statements (Table 2) investigating whether the intervention was appreciated and was
considered appropriate, relevant, and useful. In particular, items 1, 2, 5, and 6
(Cronbach’s α = 0.62) addressed enjoyment, namely the emotional engagement with the
game [28]; items 3 and 4 (Cronbach’s α = 0.77) take inspiration from the Technology
Acceptance Model, where acceptance is measured in terms of intention of using a
certain technology and is affected by the extent to which important others would think
that the system is worth being used [34]. Items 8, 10, 11, and 12 (Cronbach’s α = 0.73)
measured distinct components of credibility [10, 14]: Items 8 and 10 measured the
perceived novelty of the information given in the game; item 12 checked the perceived
accuracy of the information; and item 11 investigated the contextual appropriateness of
the intervention to the context of use, an important dimension for nightlife interventions
in general [9, 30, 35]. All questions were in Italian.
Participants and procedure. The sample consisted of 227 participants (77 women,
150 men, age: M = 24.69, SD = 6.78). Participants were event attendees who volun-
tarily approached the intervention area. After signing an informed consent form, par-
ticipants were shown how to interact with the game. Then, the sessions started. The
game included 9 questions and up to 6 players per session. After completing the
session, they were administered the post-session questionnaire.
3.2 Results
Let’s consider first the items administered twice, the pre- and post-sessions (Table 1).
We compared the scores obtained before and after the game with a Wilcoxon test for
paired samples, finding the difference significant in both cases (Z = −3.16; p < 0.05;
r = −0.210 and Z = −2.103; p < 0.05; r = −0.14).
Table 1. Median, mean value, and standard deviation of the awareness items in the pre- and
post-session questionnaires
Pre Mdn M SD Post Mdn M SD
7. I am aware that with the 5 4.75 1.46 5 5.05 1.28
proper company I can
avoid some risks.
9. I am aware of the risks 5 4.65 1.58 5 4.87 1.45
one can run during a
night out.
For the other user experience items (Table 2), we compared the scores obtained
with the central value of the scale (3.5) with the nonparametric signed rank Wilcoxon
test with a single sample. The underlying dimension is assumed to vary on a contin-
uum. This justifies comparison with a central point even if this point is not provided in
the response scale. The results are shown in Table 2, which indicates that all scores are
positive and significantly different from the central value of the scale.
Table 2. Mean value, standard deviation, and result of Wilcoxon signed rank test for each item
of the questionnaire. * = P < 0.001
Item M SD V
1. I liked playing the game 4.98 1.25 23936.5*
2. Had I played alone, I’d have less fun 4.40 1.89 18676*
3. I’d like to use this game again in the future 4.33 1.49 20122.5*
4. I would recommend my friends to play 4.65 1.38 22148*
5. I liked the remote control used in the game 4.20 1.56 18679.5*
6. The game is attractive 4.61 1.34 22157.5*
7. I am aware that with the proper company I can avoid some 4.87 1.45 22659*
risks
8. While playing, I felt the wish to get better informed about 3.94 1.60 16701*
substances
9. I am aware of the risks one can run during a night out 5.10 1.19 23865.5*
10. By playing I learned something new about substances 4.57 1.56 21039.5*
11. I think that a videogame is a good way to provide 4.93 1.28 23733*
information during a night out
12. Although brief, the information provided by the game is 4.85 1.26 23513*
accurate
4 Study 2: Knowledge Increase
The goal of the second study was to test whether a game session was able to increase
users’ short-term knowledge about substance consumption risks and coping skills. This
is in line with the goals of a so-called “brief intervention” in general [23] and in this
domain in particular, which does not aim at long-term effects or providing articulated
information. We therefore assessed the players’ knowledge before and after the game
session. We hypothesized that the number of correct answers to test questions increased
after the game session (H1), whereas no variation was observed in the control questions
unrelated to the game material (H2). This study does not test long-term knowledge
improvements.
4.1 Method
From the database of the game, we selected 40 quiz items for the study. They were items
that, during previous interventions, received the highest percentage of wrong answers
and varied to cover different kinds of substances (alcohol, amphetamines, cannabis,
ketamine, cocaine and other stimulants, ecstasy, heroin and methamphetamines,
GHB-GBL, popper and others solvents, and hallucinogens) and effects (first aid, sex-
ually transmitted diseases, urban legends, medical effects, pregnancy, history, and short-
and long-terms effects). We used these 40 items as follows (Table 3):
• 20 items made up the pre- and post-test questions, assessing users’ knowledge, and
also appeared in the game, where the user could see the correct answer and learn its
explanation;
• 5 items informed pre- and post-session questions but not the game, so players could
not learn the correct answer from the game;
• 15 were distracting questions with no related pre- and post-test.
Table 3. Study 2 design

Pre-game Game Post-game
20 test questions x x x
5 control questions x - x
15 distracting questions - x -
A game session consisted of 35 questions, 20 test, and 15 distracting; their order

was randomized automatically by the game system. Before and after each session the
participant was asked 25 questions, i.e., 20 test questions and 5 control. Each test and
control question was asked twice, once before the game session, and once after the
game session. They were multiple choice, with one correct response option only. The
order of response options changed between pre- and post-game session.
Participants and procedure. The sample consisted of 81 participants (51 men,
30 women), aged 23.36 on average (SD = 2.02). We recruited university students by
asking people encountered on campus, and offered no compensation for participating.
The game sessions were multiplayer, as this is the modality in the field interventions. Four
participated in each. A session started by welcoming the participants and having them
read and sign the informed consent form. Afterwards, we collected background data
(gender, age, nationality, mother tongue, education, and familiarity with videogames),
the 25-item, we administered the pre-session questionnaire, followed by the game ses-
sions. After the game session, we collected the 25-item post-session questionnaire.
During the game, participants could talk but not make shared decisions or affect each
other’s responses in other ways. They were reminded that game ranked their individual
performance so they were actually competing with each other. The whole procedure took
45–50 min to complete.
4.2 Results
We have run analyses with SPSS. Regarding the percentage of correct answers pre- and
post-game session, the Wilcoxon test shows that the latter is higher than the former
(Mdn = 90 vs. 60), Z = −7.785, p < 0.05, r = −0.865 (Fig. 3).
Fig. 3. The median percentage of correct answers before and after the game session.
Regarding the five control questions, we analyzed their variation before and after
the session and found it very low: only 18.52 % of them varied after the session,
suggesting that it was the specific content acquired during the game to modify the
answers in the test questions. Finally, actual performance in the game is not related to
the percentage of correct answers before or after the session, F (3, 77) = 0.62, p = .60.
showing that test performance at the test questions improved also if the performance at
the game was bad, thanks to the explanation provided during the game (and offering a
learning opportunity).
5 General Discussion and Conclusions
The two validation studies gave encouraging results. The strength of the first study was
to test the game reception in the field. The reasons for carrying out the second study in
the lab were twofold. First, its controlled design needed a limited number of questions;
had the test taken place in the field, the large screen of the game would have shown the
same quiz questions over and over, spoiling the sample after just a few sessions.
Second, the scientific purposes of the study demanded it follow a pre-defined, rigid
structure, difficult to maintain under the shifting circumstances of a field intervention.
Nonetheless, none of these requirements are necessary for field intervention or an
effective learning experience. In addition, during field interventions, participants vol-
unteer because of their interest in the topic and the information acquired has immediate
relevance, both preconditions for effective learning. A higher motivation would
increase their ability to retain the information learnt in a field intervention, compen-
sating for the lower ability to sustain attention due to noise and physical state.
Together, the first and second studies prove that gamified intervention is able to
reach its goals, in terms of user experience and learning, in addition to its powerful
attraction by default. These results encourage more testing and adoption of (carefully
designed) serious games in brief field interventions.
Acknowledgments. The European Commission, via the Nightlife, Empowerment and

Well-being Implementation (NEW-IP, n. 29299) project, supported this work. The authors would
also like to thank Leonardo Montecchi, the Rimini Municipality, Giuseppe di Pino (Municipality
of Venice) and S.E.R.T in Faenza for their support. The funding body did not affect the authors’
decisions about the study design, collection, analysis and interpretation of data, the writing of the
report, or the decision to submit the article for publication.
References
1. Bødker, S., Grønbæk, K.: Cooperative prototyping: users and designers in mutual activity.
Int. J. Man Mach. Stud. 34(3), 453–478 (1991)
2. Bogost, I.: Persuasive Games: The Expressive Power of Videogames. MIT Press, Cambridge
(2007)
3. Bolier, L., Voorham, L., Monshouwer, K., Hasselt, N.V., Bellis, M.: Alcohol and drug
prevention in nightlife settings: a review of experimental studies. Subst. Use Misuse 46(13),
1569–1591 (2011)
4. Burns, J.M., Webb, M., Durkin, L.A., Hickie, I.B.: Reach out central: a serious game
designed to engage young men to improve mental health and wellbeing. Med. J. Aust. 192
(11), S27 (2010)
5. Calafat, A., Blay, N., Juan, M., Adrover, D., Bellis, M.A., Hughes, K., et al.: Traffic risk
behaviors at nightlife: drinking, taking drugs, driving, and use of public transport by young
people. J. Traffic Inj. Prev. 10, 162–169 (2009)
6. Calafat, A., Fernàndez, C., Juan, M., Becona, E.: Weekend nightlife recreational habits:
prominent intrapersonal “risk factors” for drug use? Subst. Use Misuse 42, 1443–1454
(2007)
7. Connolly, T., Boyle, E., MacArthur, E., Hainey, T., Boyle, J.: A systematic literature review
of empirical evidence on computer games and serious games. Comput. Educ. 59(2), 661–
686 (2012)
8. Deterding, S., Dixon, D., Khaled, R., Nacke, L.: From game design elements to
gamefulness: defining “gamification”. In: Proceedings of the 15th International Academic
MindTrek Conference: Envisioning Future Media Environments, pp. 9–15. ACM,
New York, NY (2011)
9. European Monitoring Centre for Drugs and Drug Addiction: Responding to drug use and
related problems in recreational settings. Document n. TDXA12003ENN (2012). http://
www.emcdda.europa.eu/attachements.cfm/att_184673_EN_Recreational_settings_WEB.pdf
10. Flanagin, A.J., Metzger, M.J.: The role of site features, user attributes, and information
verification behaviors on the perceived credibility of web-based information. New Media
Soc. 9(2), 319–342 (2007)
12. Gamberini, L., Breda, L., Grassi, A.: VIDEODOPE: applying persuasive technology to
improve awareness of drugs abuse effects. In: Shumaker, R. (ed.) HCII 2007 and ICVR
2007. LNCS, vol. 4563, pp. 633–641. Springer, Heidelberg (2007)
13. Gamberini, L., Marchetti, F., Martino, F., Spagnolli, A.: Designing a serious game for young
users: the case of happy farm. Annu. Rev. Cybertherapy Telemedicine 7, 77–81 (2009)
14. Gaziano, C., McGrath, K.: Measuring the concept of credibility. J. Q. 63, 451 (1986)
15. Hamari, J., Koivisto, J.: “Working out for likes”: an empirical study on social influence in
exercise gamification. Comput. Hum. Behav. 50, 333–347 (2015)
16. Hunt, G., Moloney, M., Evans, K.: Youth, Drugs, and Nightlife: Pleasures, Risks, and
Identity. Routledge, London (2010)
17. Hunt, G., Moloney, M., Fazio, A.: “A cool little buzz”: alcohol intoxication in the dance
club scene. Subst. Use Misuse 49(8), 968–981 (2014)
18. Kelly, B.C.: Conceptions of risk in the lives of ectasy-using youth. In: Sanders, B. (ed.)
Drug, Clubs and Young People—Sociological and Public Health Perspectives, pp. 50–66.
Ashgate Publishing, Hampshire (2006)
19. Kelly, B.C., LeClair, A., Parsons, J.T.: Methamphetamine use in club subcultures. Subst.
Use Misuse 48(14), 1541–1552 (2013)
20. Kraft, P., Drozd, F., Olsen, E.: ePsychology: designing theory-based health promotion
interventions. Commun. Assoc. Inf. Syst. 24, 399–426 (2009)
21. Lieberman, D.: Management of chronic pediatric diseases with interactive health games:
theory and research findings. J. Ambul. Care Manag. 24(1), 26–38 (2001)
22. Miller-Day, M., Hecht, M.L.: Narrative means to preventative ends: a narrative engagement
framework for designing prevention interventions. Health Commun. 28, 657–670 (2013)
23. Moyer, A., Finney, J.W.: Brief interventions for alcohol problems: factors that facilitate
implementation. Alcohol Res. Health 28(1), 44–50 (2004)
24. North, L., Robinson, C., Haffegee, A., Sharkey, P.M., Hwang, F.: Using virtual
environments for trigger identification in addiction treatment. In: Proceedings of the 9th
International Conference on Disability, Virtual Reality & Associated Technologies, pp. 345–
353. Laval, France, 10–12 September 2012
25. Oinas-Kukkonen, H.: A foundation for the study of behavior change support systems. Pers.
Ubiquit. Comput. 17(6), 1223–1235 (2013)
27. Papastergiou, M.: Exploring the potential of computer and video games for health and
physical education: a literature review. Comput. Educ. 53(3), 603–622 (2009)
28. Phillips, R.S., Horstman, T., Vye, N., Bransford, J.: Engagement and games for learning
expanding definitions and methodologies. Simul. Gaming 45(4–5), 548–568 (2014)
29. Quintero, G., Peterson, J., Young, B.: An exploratory study of socio-cultural factors
contributing to prescription drug misuse among college students. J. Drug Issues 22, 903–926
(2006)
30. Roberts, G., McCall, D., Stevens, Lavigne, A., Anderson, J., Paglia, A., Bollenbach. S.,
et al.: Preventing substance use problems among young people: a compendium of best
practices. Report n. Cat. No. H39-580/2001E, Produced by Canadian Centre on Substance
Abuse (2001). http://www.hc-sc.gc.ca/hc-ps/alt_formats/hecs-sesc/pdf/pubs/adp-apd/
prevent/young-jeune-eng.pdf
31. Schensul, J.J., Convey, M., Burkholder, G.: Challenges in measuring concurrency, agency
and intentionality in polydrug research. Addict. Behav. 30(3), 571–574 (2005)
32. Seaborn, K., Fels, D.J.: Gamification in theory and action: a survey. Int. J. Hum.-Comput.
Stud. 74(2015), 14–31 (2015)
33. Stewart, J., Bleumers, L., Van Looy, J., Mariln, I., Schurmans, D., Willaert, K., De Grove,
F., Jacobs, A., Misuraca, G.: The potential of digital games for empowerment and social
inclusion of groups at risk of social and economic exclusion: evidence and opportunity for
policy. EU JRC Institute for Prospective Technological Studies, Scientific and Policy
Reports, 2013 (2013)
34. Venkatesh, V., Davis, F.D.: A theoretical extension of the technology acceptance model:
four longitudinal field studies. Manag. Sci. 46(2), 186–204 (2000)
35. Vogl, L.E., Teesson, M., Newton, N.C., Andrews, G.: Developing a school-based drug
prevention program to overcome barriers to effective program implementation: the
CLIMATE schools: alcohol module. Open J. Prev. Med. 2(3), 410–422 (2012)
Long-Term Effects of Computerized Simulations
in Protracted Conflicts: The Case of Global Conflicts
Ronit Kampf ✉
( )
Tel Aviv University, Tel Aviv, Israel

[email protected]
Abstract. This article presents an experimental study examining the short term
and long term effects of Global Conflicts on attitude change towards the Israeli-
Palestinian conflict. Global Conflicts is a role-playing computer game simulating
this conflict. 180 undergraduate students from Israel and Palestine participating
in the study were divided into game intervention and no-game controls. The
participants were required to fill in questionnaires measuring attitudes regarding
the conflict immediately before and after the game intervention and 12 months
following this intervention. Results suggested that participants who played the
game, unlike those who did not play it, shifted towards a more impartial perspec‐
tive, being able to look at the conflict from both Israeli and Palestinian points of
view immediately after the game intervention, and retained this perspective even
one year after participation in this intervention, despite the serious clashes
between Israel and the Palestinians that occurred during this time.
Keywords: Games for change · Israeli-Palestinian conflict · Attitude change ·

Conflict resolution · Long term effects · Computerized simulations
1 Introduction
This article investigates the effectiveness of a computer based peace game called Global
Conflicts that simulates the Israeli-Palestinian situation, in order to see whether this
game enhances the taking of an impartial perspective regarding the conflict (i.e., being
able to look at the conflict from both Israeli and Palestinian points of view) and if this
learning outcome is retained 12 months following the game intervention. The study
compares the short term and long term learning outcomes of Israeli-Jewish and Pales‐
tinian undergraduate students playing the game with those not playing it in order to
examine whether this game-based intervention actually works.
Global Conflicts is an award-winning educational game developed in 2010 by
Serious Games Interactive in Denmark (http://globalconflicts.eu/). The game illustrating
the tensions between Israeli and Palestinian sides in a checkpoint scenario. The player
is represented by the avatar of a Western reporter who arrives in Jerusalem. Her task is
to write for one of the following newspapers: Israeli, Palestinian, or Western. The player
is expected to produce a news report geared to the audience of one of these newspapers,
based on the interviews she conducts with various Israeli and Palestinian characters at
the checkpoint in the Palestinian territories.

DOI: 10.1007/978-3-319-31510-2_21
Long-Term Effects of Computerized Simulations 243
Previous studies have already shown that Israeli and Palestinian young people (like
those participating in this study) know almost nothing about what transpires on the other
side of the Israeli-Palestinian divide, except for the stereotypic and ethnocentric images
constructed by key socialization agents (e.g., the media) and daily incidents [e.g., 1].
Moreover, since these young people have never actually experienced a state of peace
they may not regard it as a significant value for which a price should be paid [e.g., 2].
Interestingly, studies have indicated the positive short term learning effects of
computerized simulations like Global Conflicts [e.g., 3, 4], but no empirical research,
to my knowledge, have been conducted to evaluate the long term effects of game-based
interventions in intractable conflicts like the Israeli-Palestinian situation. In general, the
number of studies on the long-term effects of peace workshops in such protracted
conflicts is extremely limited [e.g., 2, 5].
Previous studies have suggested that peace games like Global Conflicts may be
successful as tools for learning about the “other” for youth on the Israeli-Palestinian
divide, for a few key considerations. First, Global Conflicts may induce more immersion
and can therefore be more enjoyable, boosting players’ engagement and motivation
[e.g., 6, 7]. Second, Global Conflicts focuses on the hardships of both Israeli soldiers
and Palestinian civilians at a checkpoint in the Palestinian territories, an experience
which may produce empathy and identification with both sides [e.g., 8]. Finally, in
Global Conflicts, a player assumes the role of a more distant party to the conflict who
is obliged to look at the situation from both perspectives and act according to norms of
balance and objectivity (i.e., a Western journalist). Therefore, the player may be less
prone towards the tendency to interpret, favor, and recall information in a way that
confirms her own beliefs while giving disproportionately less attention to information
that contradicts it [e.g., 9]. As a result, the player may find it easier to look at the conflict
through the lenses of both sides [e.g., 10, 11].
2 Research Hypotheses
H1: Participants will hold ethnocentric attitudes before the game intervention.
H2: Participants playing the game will become more impartial toward the conflict
immediately after the game intervention, while those who do not play it will retain
ethnocentric attitudes toward the conflict during this time.
H3: Participants playing the game will retain impartial attitudes toward the conflict
even one year after participation in the game intervention, while those who do not play
it will retain ethnocentric attitudes toward the conflict during the same period.
3 Methodology
3.1 Participants
100 Israeli-Jewish participants were from the Departments of Communication and

Political Science at Tel Aviv University and 80 Palestinian students were from the
Department of Political Science at Al-Quds University.
244 R. Kampf
60 Israeli-Jewish students and 45 Palestinian students played the game (experimental

group), while 40 Israeli-Jewish students and 35 Palestinian students did not play it
(control group). The experimental and control groups did not differ in key characteristics
that could provide alternative explanation for the results (Table 1).
Table 1. Characteristics of the experimental and control groups

Age Male Political attitudes Religiosity Playing a Political interest
digital
1 10 1 10 game in 1 4
the last
Left Right Religious Secular six Not at all Very
months much
M(SD) M(SD) M(SD) M(SD)
Global 23.1 (1.25) 36 % 6.12 (2.45) 6.7(1.17) 45 % 3.5 (0.47)
Conflicts
No Game 22.9 (1.23) 34 % 5.87 (2.29) 6.45(1.12) 48 % 3.46 (0.65)
3.2 Design and Procedure
The experimental condition took three hours including four parts. First, participants were
introduced to the Global Conflicts game and played a short demo (not related to the
Israeli-Palestinian conflict). Second, they filled in a short questionnaire. Third, the
participants were randomly divided to play a Western journalist representing either an
Israeli or a Palestinian newspaper in the game. Finally, the participants again filled in a
short questionnaire.
The control condition took three hours including three parts. First, participants filled
in a short questionnaire. They were then given a lecture about digital natives (not related
to the conflict). Finally, they again filled in a short questionnaire.
Both participants who played the game and those who did not play it were told that
they would be contacted by email a year later to answer a short questionnaire and they
will receive credit for their participation. The experimental group was told that the
questionnaire would examine what they remembered from the game and the control
group was told that the questionnaire would examine what they remembered from the
lecture about digital natives in order to learn about the effectiveness of the two classes.
3.3 Measures
The attitude measure examined the ‘rightness’ of each side on key issues in the conflict
including water, refugees, borders, settlements, Jerusalem, and security, using the
following scale: 1. Palestinians are absolutely right, 2. Palestinians are somewhat right,
3. Both sides are equally right, 4. Israelis are somewhat right, and 5. Israelis are abso‐
lutely right. After conducting a factor analysis, the average of answers given on the six
key issues was used as a measure of attitude change about key issues in the conflict
before and after playing the game. This measure has already been used in previous
studies conducted with computerized simulations of the Israeli-Palestinian conflict
[3, 4, 12–14], and is based on a questionnaire developed by conflict resolution scholars
from Israel and Palestine [e.g., 2].
3.4 Statistical Procedures

A Repeated Measures ANOVA was used to test the research hypotheses, investigating
the effects of playing the game (yes or no) and nationality (Israeli-Jewish or Palestinian)
on attitude values at three separate time points: immediately before and after the game
intervention and 12 months following this intervention. The important point with this
study design is that the same participants are measured three times on the same
dependent variable, so this test can detect any overall differences between related means.
4 Results
The interaction between time, playing the game and nationality was significant,
suggesting that Israeli-Jewish and Palestinian participants playing the game and those
who did not play it differed in attitude change regarding key issues in the conflict (F(3,
176) = 71.03, p < .0001, η2 = .36).
Table 2. Nationality and game-playing effects on attitudes toward the conflict
Pre-game interven‐ Post-game interven‐ 12 months following

tion M:SD tion M:SD game intervention
M:SD
Israeli-Jews
Game 3.92:.59*** 2.91:.31*** 2.93:.26***
No Game 3.71:.53*** 3.62:.53*** 3.52:.47***
Palestinians
Game 1.07:.09*** 1.44:.34*** 1.58:.36***
No Game 1.06:.12*** 1.07:.11*** 1.14:.16***
***p < .0001
Before the game intervention, Israeli-Jewish participants playing the game held a
pro-Israeli view, while Palestinian participants playing the game held a pro-Palestinian
view (i.e., ethnocentric attitudes). Similarly, Israeli-Jewish participants who didn’t play
the game held a pro-Israeli view during this time, while Palestinian participants who
didn’t play the game held a pro-Palestinian view (Table 2). Therefore, hypothesis 1 is
confirmed.
Immediately after the game intervention, Israeli-Jewish and Palestinian participants
playing the game got closer to thinking that both Israelis and Palestinians are equally
right regarding key issues in the conflict (i.e., an impartial perspective). In contrast,
Israeli-Jews and Palestinians who did not play the game retained ethnocentric attitudes
toward the conflict during this time (Table 2). Therefore, hypothesis 2 is confirmed.
246 R. Kampf
12 months following the game intervention, Israeli-Jewish and Palestinian partici‐

pants playing the game retained impartial perspective regarding the conflict. In contrast,
Israeli-Jewish and Palestinian participants who did not play the game retained ethno‐
centric perspective regarding the conflict during this time (Table 2). Therefore, hypoth‐
esis 3 is confirmed.
5 Discussion and Conclusions
By using the Global Conflicts game, which is a simulation of the Israeli-Palestinian

conflict, this study assessed whether participants could develop an impartial perspective
and whether this learning outcome persisted 12 months after the game intervention.
Results suggested that participants held ethnocentric attitudes toward the conflict before
the game intervention. In addition, participants playing the game developed more impar‐
tial attitudes toward the conflict immediately after the game intervention, while those
who didn’t play it retained ethnocentric attitudes toward the conflict during the same
time. Finally, participants playing the game retained impartial attitudes even 12 months
following the game intervention, while those who didn’t play it retained ethnocentric
attitudes toward the conflict during this time.
The results are promising in terms of showing that computer based peace games can
be used not only for teaching purposes in courses related to conflict and peace studies,
but also as part of peace building trainings. The results indicate that such games are not
only useful in teaching a more complex view of the conflict to direct parties to the conflict
with strong and ethnocentric attitudes on the issues, but also in engendering attitude
change especially in the form of taking a more impartial perspective and being able to
look at the conflict from both sides points of view, even one year following the game
intervention, despite the serious clashes between Israel and the Palestinians that occurred
during this time.
This study has a few limitations that should be addressed in future research. First,
the study focused on Political Science students who may be more interested in politics
and therefore most likely do not represent the general public. In addition, it is unclear
whether participants would have enjoyed the game and changed their attitudes to an
equal degree, if no incentive had been presented at the end. Finally, it remains unclear
to what extent the change in attitude is caused by the information the students were
exposed to, and to what degree the change is caused by the game-aspect of Global
Conflicts.
Further research can isolate different dimensions of computerized simulations like
Global Conflicts to understand how such games achieve their short term and long term
effects in the shadow of intractable conflicts such as the Israeli-Palestinian situation. It
could have also been interesting to examine the impact of Global Conflicts compared
with another format presenting the same information (e.g., a lecture, a written text, a
presentation or videos of others playing the game) in order to examine specifically
whether the game- aspect of Global Conflicts adds more than other formats to the change
in attitude.
References
1. Wolfsfeld, G., Frosh, P., Awabdy, M.: Covering death in conflicts: coverage of the Second
Intifada on Israeli and Palestinian television. J. Peace Res. 45, 401–417 (2008)
2. Rosen, I., Salomon, G.: Durability of peace education effects in the shadow of conflict. Soc.
Psychol. Educ. 14, 135–147 (2011)
3. Cuhadar, E., Kampf, R.: A cross-national inquiry into the Israeli-Palestinian and Guatemalan
scenarios in global conflicts. Negot. Confl. Manage. Res. 8(4), 243–260 (2015)
4. Kampf, R., Cuhadar, E.: Do computer games enhance learning about conflicts? A cross-
national inquiry into proximate and distant scenarios in global conflicts. Comput. Hum.
Behav. 52, 541–549 (2015)
5. Schroeder, J., Risen, J.L.: Befriending the enemy: outgroup friendship longitudinally predicts
intergroup attitudes in a coexistence program for Israelis and Palestinians. Group Process.
Intergroup Relat. 4(5), 1–22 (2014)
6. Fu, F.L., Su, R.S., Yu, S.C.: EGameFlow: a scale to measure learners enjoyment of e-Learning
games. Comput. Educ. 52(1), 101–112 (2009)
7. Kiili, K.: Digital game-based learning: towards an experiential game model. Internet High.
Educ. 8(1), 13–24 (2005)
8. Gross, K.: Framing persuasive appeals: episodic and thematic framing, emotional response,
and policy opinion. Polit. Psychol. 29(2), 169–192 (2008)
9. Stanovich, K.E., West, R.F., Toplak, M.E.: Myside bias, rational thinking, and intelligence.
Curr. Dir. Psychol. Sci. 22(4), 259–264 (2013)
10. Halperin, E., Pliskin, R., Saguy, T., Liebrman, V., Gross, J.J.: Emotion regulation and the
cultivation of political tolerance: searching for a new track for intervention. J. Confl. Resolut.
58(6), 1100–1138 (2014)
11. Halperin, E., Porat, R., Tamir, M., Gross, J.J.: Can emotion regulation change political attitudes
in intractable conflicts? From the laboratory to the field. Psychol. Sci. 24(2), 106–111 (2013)
12. Kampf, R., Cuhadar, E.: Learning about the Israeli-Palestinian conflict and negotiations
through simulations: the case of peacemaker. Int. Stud. Perspect. 15(4), 509–524 (2014)
13. Kampf, R.: Playing singly, playing in dyads in a computerized simulation of the Israeli-
Palestinian conflict. Comput. Hum. Behav. 32, 9–14 (2014)
14. Kampf, R.: Computerized Simulations of the Israeli-Palestinian Conflict and Attitude
Change. eLearning Pap. 43 (2015)
Interventions for Behavior Change
Understanding Changes in the Motivation of Stroke
Patients Undergoing Rehabilitation in Hospital
Michelle Pickrell1 ✉ , Bert Bongers1, and Elise van den Hoven1,2,3

( )
1
Faculty of Design, Architecture and Building, University of Technology Sydney,
Sydney, Australia
2
Department of Industrial Design, Eindhoven University of Technology,
Eindhoven, The Netherlands
3
Duncan of Jordanstone College of Art and Design, University of Dundee, Dundee, UK
[email protected]
Abstract. Stroke patient motivation can fluctuate during rehabilitation due to a

range of factors. This study reports on qualitative research, consisting of observa‐
tions of stroke patients undergoing rehabilitation and interviews with patients about
the changes in motivation they identified during their time completing rehabilitation
in the hospital. We found a range of positive and negative factors which affect moti‐
vation. Positive factors include improvements in patient movement and support from
other patients and family members. Negative factors include pain and psycholog‐
ical issues such as changes in mood. From this fieldwork, a set of design guidelines
has been developed to act as a platform for researchers and designers developing
equipment for the rehabilitation of stroke patients.
Keywords: Rehabilitation · Stroke · Healthcare · Feedback · Design research
1 Introduction
Rehabilitation following stroke can be a time intensive process with patients having to
commit to multiple repetitions of specific exercises [1]. Interactive equipment could be
used to help patients with their rehabilitation by offering feedback about their limb
movements and balance. Currently however, there are few interactive technologies
available to patients who are completing their rehabilitation in the hospital.
The research goals for this project are:
1. To understand the patients’ perspective of the differences between high and low
motivation towards completing their daily rehabilitation exercises
2. To understand the patient perception of the reasons for changes in motivation
This paper focuses on the best ways of encouraging patients to be motivated during
stroke rehabilitation. The aim of this paper is to support designers who are creating
interactive rehabilitation equipment for patients undergoing rehabilitation, both in the

DOI: 10.1007/978-3-319-31510-2_22
252 M. Pickrell et al.
ward and home environments. ‘Motivation’, for the purpose of this study, is defined as
the reason or reasons a person acts in a particular way.
2 Background
To support this research, the topics of stroke, stroke rehabilitation and persuasive tech‐
nology are important. It is also important to understand motivation as it is the main focus
of this research. These topics are discussed in this section, together with the existing
work that focuses on the use of interactive equipment to motivate patients.
2.1 Motivation
The level of motivation a stroke patient has will influence the outcome of their rehabil‐
itation [2]. Knowing the direct relationship between motivation and success, we can
work to design appropriate equipment to motivate patients. This paper focuses on the
theories of motivation, including goal- setting theory and self-efficacy theory.
Goal setting has been proven to have a positive effect on patient motivation [3]. It is
however, important that patients have input into the goal creation process so that they
feel a sense of ownership over their goals [4]. It is also important that goals are small
and achievable, with goals that are relevant to the patient’s interests and hobbies being
the most successful [3].
Self-efficacy theory is a central concept of Bandura’s Social Cognitive Theory and
focuses on the relationship between an individual’s degree of confidence to complete a
task and their ability to do so [5]. As stroke patients have to re-learn how to complete
basic tasks such as walking or standing up, confidence in their ability to do this learning
is important. Existing research into self-efficacy in stroke patients, which focuses on the
subject of falls, shows that there is a relationship between self-efficacy and a patient’s
rehabilitation success over time [5]. Some patients, however, do not feel the same moti‐
vation to get better. This can result from factors such as having family members that can
help with activities of daily living when they return home. Another factor is age. Younger
patients often feel more of a need to get better as they have their life ahead of them,
whereas older patients are not always motivated in the same way [3].
2.2 Persuasive Technology
The Persuasive Technology field of research focuses on the use of technology to change
behaviour. Persuasive technology research often focuses on motivation, in particular
around the topic of exercise [6–8]. Existing research identifies a number of strategies
that can be used when designing for motivation. Firstly, setting goals that are relevant
to the user and making them public to the community so the person feels an obligation
to their community [9]. Secondly, ensuring the technology channel used is relevant to
the target market [10]. Finally, a strategy that focuses on stroke rehabilitation is to
provide both immediate and long term feedback to patients [8].
In the context of stroke rehabilitation, interactive technology could support the
time-poor physiotherapists and play an important role in persuading patients to exer‐
cise [11]. As BJ Fogg points out, in some cases computers can have a number of
Understanding Changes in the Motivation of Stroke Patients 253
distinct advantages over humans when it comes to persuasion, as they are persistent
and can use many modalities for example, text, audio and sensory feedback [12].
Together with these factors, interactive technology can have a positive role as it can
be motivating and stimulating [13, 14].
2.3 Stroke and Stroke Rehabilitation

An understanding of stroke and stroke rehabilitation is important to the context of this
research. A stroke results from an interruption to the blood supply to the brain. This can
be caused by either a blockage or bleeding from a blood vessel [15]. The damage caused
by a stroke is due to the brain not receiving enough oxygenated blood and as a result the
affected brain cells cannot survive. The physical and cognitive effects of a stroke are
dependent on the size and location of the stroke in the brain. In most cases of stroke,
one side of a patient’s body will be affected. Some other effects of stroke include memory
loss, speech difficulties, changes in vision and behavioural changes [16].
Following a stroke, patients undergo rehabilitation that includes extensive interac‐
tion between a patient and their therapist [17]. However, patients are often dealing with
a range of different factors that result from their stroke. Such factors include depression
and lack of motivation, as well as their coming to terms with the massive physical change
that has just happened in their lives [18]. Patients have differing success with motivation
with 31% of patients who have motor deficiencies resulting from their stroke not
completing the exercises their physiotherapist has prescribed [19].
Stroke rehabilitation often consists of repetitive, simple exercises such as sitting
down and standing up or walking up and down stairs. The repetitive nature of these
exercises can result in patients becoming bored [20]. The objects used in the rehabili‐
tation ward tend to be everyday objects such as styrofoam cups, popsicle sticks and
wooden blocks. Whilst these are objects that patients are familiar with, they don’t offer
the patient much feedback about the way they are completing movements.
Some gaming technologies have been introduced into the rehabilitation ward such
as the Nintendo Wii and the Microsoft Xbox Kinect, however as these are designed for
able-bodied people, they often do not give patients appropriate and beneficial feedback
for their condition [20, 21].
2.4 Related Work
The changes in motivation of stroke patients who are completing rehabilitation have
been explored by studying relevant publications, particularly in the healthcare domain.
As well as healthcare, this section presents a review of motivation related human
computer interaction (HCI) studies and persuasive technology studies.
From the clinical perspective, there are few recent studies that focus on under‐
standing the motivation of stroke patients when undergoing rehabilitation. The few
studies that do focus on the topic of motivation, report on clinical implications, focusing
on the clinician’s interaction with the patient as opposed to design outcomes that focus
on equipment and technology. Important aspects of motivation, however, are identified,
including a patient’s understanding of the reasons for rehabilitation and understanding
the role of nurses and physiotherapists as well as understanding what their own role is
in the rehabilitation process [3, 18].
In the field of HCI, there are some equipment designs that have been created to
support motivation within the hospital. One such design is the stepping tiles used by
patients to measure the number of stepping repetitions they complete during their reha‐
bilitation session. This includes a simple interface, which can be easily interpreted by
patients, as well as rich and varied feedback, similar to video games [20].
Persuasive technology research focuses on motivation in a range of different contexts
including motivation for exercise, [7] healthy eating [7, 22] and helping people live more
sustainably [23]. A number of tools have been identified within this field to aid with moti‐
vation, including triggers, goal setting, comparison between subjects, asking users to pledge
a commitment and finally giving subjects useful feedback at the right time [10, 22, 23].
3 Study Design
As a result of previous work on a set of sensor floor tiles and a sensor sleeve, we identified
the importance of using design to help patients with motivation during stroke rehabili‐
tation [24]. This previous work demonstrated the positive effect on motivation of
showing patients the incremental improvements they have made. The results of this work
identified the need to conduct a study that specifically focuses on motivation by
observing and interviewing patients who have been undergoing rehabilitation for an
extended period of time. This section outlines the study design for this project.
3.1 Participants and Recruitment

The six participants in the study were inpatients at the Bankstown-Lidcombe hospital
in Sydney. All had suffered from a stroke and had been completing rehabilitation for
over four weeks. The patients ranged in age between 64 and 89 years of age. Four of the
patients were male and two were female. The physiotherapists at the hospital identified
suitable patients and the researcher started observation. After the period of observation,
if the patient was willing to participate in the interview, the researcher would obtain
informed consent. All interview data was later anonymised to comply with ethics
guidelines.
After patients had given consent, they were screened using the Mini Mental State
examination. This is a 30-point questionnaire used to assess cognitive ability and screen
for cognitive impairment [25], including situational questions, memory-based questions
and completing simple tasks such as folding a piece of paper and placing it on the floor.
Healthy individuals score 30. Patients who scored at least 23 out of 30 were accepted
for the study.
3.2 Observations and Interviews
Observations of the six patients completing various rehabilitation exercises in the reha‐
bilitation gymnasium were conducted. In particular the researcher was focusing on the
willingness of the patients to exercise, their relationship with the physiotherapists
and their interactions with other patients. This allowed the researcher to understand each
patient’s level of motivation and the differences between how they acted and how they
talked about motivation. When observing the patient, the researcher observed from the
other side of the gymnasium and did not interact with the patients.
This approach was taken to reduce any bias on the results as patients may act differ‐
ently or try harder when they know they are being observed. During the rehabilitation
sessions, family members and friends of the patients were often present in the gym, so
the researcher was able to fit in with other plain clothed people. At the time of obser‐
vation, the patients did not know the researcher. The researcher did not introduce herself
and the patients did not know what the researcher’s interests were.
After the observations were finished, the patients were introduced to the researcher
and a one-on-one interview was conducted. All interviews took place in the patient’s
room or in the physiotherapy gym at the Bankstown-Lidcombe Hospital. All interviews
explored the rehabilitation exercises that are prescribed by the physiotherapists at the
Stroke and Aged care rehabilitation gymnasium.
The interviews were structured in three sections. Firstly the patient was asked about
their stroke and the rehabilitation they were completing at the time. Secondly, the
patients was asked about their understanding and definition of motivation to ensure there
was a shared understanding. The patient was then asked to map their motivation over
the weeks they were completing rehabilitation on a scale between high motivation and
low motivation. This graph was then explored in detail with patients to explore high and
low motivation from their perspective. Finally the patient was asked about which exer‐
cises they identified as being least and most motivating.
4 Results
This section presents the combined results from the observations and interviews. Affinity
diagramming was used to extract themes from the interviews and observations.
4.1 Changes in Motivation
Most patients commented that they had ups and downs in their motivation towards
completing their rehabilitation. During the interviews patients were asked to map their
level of motivation on a scale between low motivation and high motivation during their
time in the rehabilitation ward. The results presented in Fig. 1 were self-reported during
the interview.
The graphs created by each patient showed that motivation went through ups and
downs as they completed their rehabilitation over a period of 4 weeks. In several
cases patients became slightly less motivated in the first week of exercise. Patients
commented that this was due to the fact that they had not completed exercise of this
intensity for years. The graph shows that most patients felt as though they were
highly motivated, whilst one patient, who was dealing with a lot of pain, explained
that their motivation towards rehabilitation was low due to the pain associated with
completing their exercises.
Fig. 1. Changes in patient motivation while completing rehabilitation
The patients who showed signs of being engaged and interested by their rehabilita‐
tion exercises were often observed to align that level of interest with motivation when
they graphed it on their chart. For four of the six patients this also directly aligned with
the number of repetitions they were able to do in any one session.
4.2 Progress and Improvement Aids with Motivation
Patients commented on the links they identified between their motivation and their
progress and improvement. Patients also commented that changes in their ability made
them feel better, “When there are changes with my legs and my arms then that makes
me feel good” (Participant 4). Patients also commented that it is important to them to
clearly understand if and when they have improved. As one patient commented when
talking about the benefits they saw when getting feedback, “In the arm class, at the start,
I had a bottle that only had a little bit of water in it. Now I have one that is nearly full.
I know I have achieved something” (Participant 2).
Goal setting (as discussed in Sect. 2.1) is another factor that aids with motivation.
The goal of the majority of patients is to get home and live unassisted. This was described
by all patients as being their main motivator. As one patient commented, “Sometimes I
wish I didn’t have to do the exercises, but I have been pushing myself because if I don’t
push myself I’m not going to get out of here and that is my motivation” (Participant 3).
4.3 Attitude and Understanding of the Reasons for Completing Exercises
Lifestyle and the patient’s attitude towards exercise before their stroke are other factors
affecting motivation. During the interviews all patients commented on their level and
frequency of exercise pre-stroke. There was a close link between a patient’s attitude to
exercise before and after their stroke. One patient who exercised frequently and felt they
had a good understanding of the benefits of exercise commented, “I always want to go
to the rehab gym, I know that it will make me better if I do more and more exercise”
(Participant 1). On the other hand, patients who had not done much exercise before their
stroke commented on how it was the most exercise they had done in years and that they
did not see the point in it. “It just makes my muscles hurt, I don’t see why I need to do
it if it’s just making me feel worse” (Participant 5).
Patients commented that they were not always sure why they were asked to complete
certain exercises. It was observed that the reasons for completing exercises that used
more mundane equipment such as pegs, buckets and balls were not always as clear as
the more obvious exercises such as sit to stand and walking up and down a staircase of
four steps. One patient commented, “I rather do the ‘exercises’ than the silly ones with
pegs. I prefer to do the ones that help my ability” (Participant 3).
Accepting the fact that the stroke had happened was another topic frequently
mentioned in the interviews. Patients talked about the importance of accepting the
change that has happened to them before being able to move forward and complete their
rehabilitation. As one patient commented, “You have to accept what has happened.
When I first got here I was still sad and shocked that this had happened to me. But now
I have accepted what the physiotherapists need me to do to get better. There are no days
that I don’t feel like getting up and exercising” (Participant 4). In some cases, patients
felt they did not understand enough about their stroke and what had happened.
4.4 Physical Pain and Psychological Factors Hinder Motivation
Pain was a common theme throughout the interviews. Patients talked about the pain they
faced both as a result of the effects of their stroke as well as the pain of doing intensive
exercise which they had not done for long periods of time. One patient commented, “I
am angry every day because I am in so much pain because of the arthritis in my right
knee. It is ten times worse since I had my stroke” (Participant 6). Other patients
commented on the immediate pain that resulted from them completing intensive exer‐
cise, “You have to be careful not to exercise too much, otherwise you get too sore and
can’t exercise the next day” (Participant 3).
Temporary psychological factors such as mood can also affect patient motivation.
Mood is usually temporary and can be caused by frustration, being overwhelmed or sad
[26]. One patient commented, “I have low motivation sometimes because of what has
happened (the stroke). It is very overwhelming” (Participant 4).
4.5 Social Factors
Social factors such as having other patients and family around are another factor
affecting motivation. Patients commented that the people around them; other patients,
physiotherapists and family are very motivating. One patient commented, “I am
becoming more motivated as I am getting more used to the people in the ward. When I
look around I see that you have to appreciate all the small improvements that people
make. They tell me I am improving which gets me more motivated” (Participant 5).
Another patient commented “For me, seeing my family is a big thing, so I don’t ever
lack motivation to go down to the gym because I want to show my family how much I
have improved” (Participant 1).
It was also observed that patients take a motivational role in each other’s rehabili‐
tation. In the rehabilitation ward, patients will often cheer each other on and encourage
each other to complete more repetitions of an exercise. This camaraderie was observed
to be particularly strong amongst patients who shared a ward as they had already formed
a friendship outside of the gymnasium.
The presence of family members was also observed to be particularly motivating for
patients when completing their rehabilitation. Patient’s spouses and children would often
visit the gymnasium when the patients were completing their exercise. Where possible,
family members were encouraged by the physiotherapists to support the patients by
completing the rehabilitation exercises with them. Patients would often complete more
repetitions of an exercise when their family members were completing them as well.
5 Discussion
In this section we will discuss the results of the study in relation to the research goals.
Research goal 1 - To understand the patients’ perspective of the differences between
high and low motivation towards completing their daily rehabilitation exercises
Overall, it was found that patients were able to indicate their perceived levels of
motivation when completing rehabilitation. Patients identified the causes of low moti‐
vation as being physical factors such as pain, as well as psychological factors such as
their mood. The causes of higher levels of motivation include improvement in the
patient’s range of movement and control over their movement, as well as external factors
such as the community of patients, family and friends in the rehabilitation gymnasium.
Whilst there are factors that affect motivation, most patients accepted that they needed
to complete their rehabilitation regardless of whether they felt motivated to or not. This
was because these patients had an understanding that the amount of rehabilitation they
completed would directly affect their rehabilitation outcomes.
Maclean et al. [3] also discusses the differences between high and low motivation
patients. In line with our research, Maclean identifies the patient’s desire to leave hospital
as a major goal and motivator. In line with our findings that patients did not understand
the reason that they need to complete particular exercises, Maclean also identifies the
importance of patients understanding the context and possible outcomes which will
result from completing different exercises. The Knowledge Attitude Behaviour (KAB)
model is used in the persuasive technology domain to explain the importance of knowl‐
edge for motivation, both “awareness” and “how-to” information [22]. In the context of
stroke rehabilitation, this could translate to both “awareness” of why a patient needs to
do an exercise as well as the “how to” instructional information.
Existing literature [3] also identifies the concept that motivation is a personality
factor and that some people are motivated and some are not. This was not something
that we found in our study. It was clear that patients are motivated by activities other
than exercise, so instead of being a personality factor motivation was related to what a
patient was passionate about and interested in.
In general, our study found that patients who had a good relationship with exercise
before their stroke were more likely to be motivated to complete their rehabilitation
exercises. In addition, situational factors such as mood, affected patients day to day and
resulted in fluctuating changes in motivation.
Research goal 2 - To understand the patient perception of the reasons for changes in
motivation
During the interviews, patients commented that their changes in motivation came
from both internal, body related and external factors. External factors include the
community of patients, rehabilitation staff, family and friends who encourage them when
they are in the gymnasium. All patients commented that when they improved, they felt
more motivated to continue completing their exercises. However, patients also
commented that internal factors such as being in pain from exercising the day before or
from pre-existing conditions, such as arthritis, affected their motivation as it made it
physically more difficult to complete exercises. Some older patients also commented
that it had been many years since they had done exercise of that intensity.
Existing literature that looks at patient motivation from the physiotherapists’
perspective identifies that providing information about improvements in patient’s reha‐
bilitation, as well as setting relevant and realistic goals with patients, can help with
motivation [3, 8].
This understanding of patient motivation is of particular importance for patients
when they have returned to a home environment and no longer have the help and support
of the rehabilitation staff. It is during this time that many patients discontinue their
rehabilitation exercises. As a result it is important to be able to design for these changes
in motivation so that we can persuade patients to continue with their rehabilitation when
they have varying levels of motivation.
6 Design Guidelines
The following design four guidelines (DG) are for the use of researchers and designers
wanting to create interactive equipment for stroke rehabilitation.
DG1 – Allow the user to set goals relevant to themselves
The interviews showed that that main goal of all patients is to live unassisted at home.
However, patients also discussed the positive affect that improvements and progress
have on their motivation. It is therefore important to allow the patients to set and revise
their own goals. They should be able to increase or decrease the difficulty of the exercise
as well as the number of repetitions. Therefore patients have a goal that is relevant and
achievable to them.
DG2 – Give patients feedback on their short-term and long-term improvements
During the interviews, patients discussed how important being able to see feedback
of improvements is for their motivation. Showing patients the short-term improvements
in their limb movement alongside the larger long-term improvements they have made
can help to motivate patients.
DG3 – Create a community
Both the interviews and observations showed the importance of the camaraderie and
support that the patients offer each other. Being able to support this sense of community
both in the ward and when patients have returned home could allow patients to continue
to encourage each other to complete their exercises.
Technology can also be used to facilitate the collaboration between patients and their
family members and friends. It was observed that when patient’s family members
completed their exercises with the patient, they were more motivated and in many cases
completed more repetitions of the exercises.
DG4 – Inform patients about the purpose of different exercises
During the interviews, some patients commented that they could not see the point in
certain exercises. As an example, patients questioned the value of the peg exercise where
the patient is asked to move up to fifteen pegs from the lip of one bucket to the lip of
another. However, it was observed that when patients understand the point of an exercise
and how it is helping them with their rehabilitation, they are more likely to complete the
exercise.
7 Conclusion and Next Steps
This research aims to understand the motivation of stroke patients who are undergoing
rehabilitation. By observing and interviewing patients, we gained an understanding of
the patient perception of their motivation and how it had changed during their time in
the hospital. Our findings show that there are a range of different factors which affect
motivation. Some of these factors can be designed for, for example aiding with a sense
of community, and some, such as reduction of pain, cannot.
Whilst this paper has touched on the motivation of patients who have left the hospital
and are completing their rehabilitation in a home environment, it is important to conduct
a similar study interviewing patients participating in rehabilitation at home. There may
be differences in what motivates patients when they have returned to a home environ‐
ment, as they no longer have the support of the rehabilitation team.
This research is part of a larger piece which focuses on motivation, feedback and
technology used in the rehabilitation ward from the perspective of the patient and phys‐
iotherapist. A model that focuses on feedback and motivation is being developed, for
use by designers.
Following this research phase, interactive equipment to help patients with motivation
will be designed and prototyped. To validate this equipment, we will use methods such
as a motivation questionnaire or an experience sampling method that allows us to collect
data at each time point. This technique was not used for this piece of research as we
wanted to explore a number of topics around motivation and the use of technology that
could not be explored with a questionnaire.
By observing and interviewing patients, we have also been able to understand the
context of stroke patient rehabilitation. Using this information we can better design
persuasive systems which suit the needs of patients. We can leverage the tools of
persuasive technology [23] to aid and encourage patients as they face the highs and lows
of motivation during their rehabilitation. The outcome of this study is a set of design
guidelines for researchers and designers who are designing interactive equipment for
stroke patients undergoing rehabilitation.
Acknowledgements. We would like to thank all the patients and physiotherapists who
contributed to this study. The study is completed under ethics approval (HREC/12/CRGH/185)
and (SSA/13/LPOOL/80). This research was supported by STW VIDI grant number 016.128.303
of The Netherlands Organization for Scientific Research (NWO), awarded to Elise van den Hoven.
References
1. Balaam, M., Rennick Egglestone, S., Fitzpatrick, G., Rodden, T., Hughes, A., Wilkinson, A.,
Nind, T., Axelrod, L., Harris, E., Ricketts, I., Mawson, S., Burridge, J.: Motivating mobility:
designing for lived motivation in stroke rehabilitation. In: Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems (CHI 2011), pp. 3073–3082. ACM,
New York (2011)
2. Holmqvist, L.W., Koch, L.: Environmental factors in stroke rehabilitation: being in hospital
itself demotivates patients. Br. Med. J. 322, 1501 (2001)
3. Maclean, N., Pound, P., Wolfe, C., Rudd, A.: The concept of patient motivation: a qualitative
analysis of stroke professionals’ attitudes. Stroke 33, 444–448 (2002)
4. Rosewilliam, S., Roskell, C., Pandyan, A.D.: A systematic review and synthesis of the
quantitative and qualitative evidence behind patient-centered goal setting in stroke
rehabilitation. Clin. Rehabil. 25, 501–514 (2011)
5. Hellstom, K., Lindmark, B., Wahlberg, B., Fugl-Meyer, A.R.: Self-efficacy in relation to
impairments and activities of daily living disability in elderly patients with stroke: a
prospective investigation. J. Rehabil. Med. 35, 202–207 (2003)
6. Wunsch, M., Stibe, A., Millonig, A., Seer, S., Dai, C., Schechtner, K., Chin, R.C.C.: What
makes you bike? Exploring persuasive strategies to encourage low-energy mobility. In:
MacTavish, T., Basapur, S. (eds.) PERSUASIVE 2015. LNCS, vol. 9072, pp. 53–64.
7. Toscos, T., Faber, A., An, S., Gandhi, M.P.: Chick clique: persuasive technology to motivate
teenage girls to exercise. In: CHI 2006 Extended Abstracts on Human Factors in Computing
Systems (CHI EA 2006), pp. 1873–1878. ACM, New York (2006)
8. Pickrell, M., Bongers, B., van den Hoven, E.: Understanding persuasion and motivation in
interactive stroke rehabilitation. In: MacTavish, T., Basapur, S. (eds.) PERSUASIVE 2015.
9. Consolvo, S., Klasnja, P., McDonald, D.W., Landay, J.A.: Goal-setting considerations for
persuasive technologies that encourage physical activity. In: Proceedings of the 4th
International Conference on Persuasive Technology, pp. 1–8, ACM, New York (2009)
10. Fogg, B.J.: Creating persuasive technologies: an eight-step design process. In: Proceedings
of the 4th International Conference on Persuasive Technology, pp. 1–6. ACM, California
(2009)
11. van den Berg, M., Sherrington, C., Hassett, L., Killington, M., Smith, S.T., Bongers, A.J.,
Hassett, L., Crotty, M.: Video and computer-based interactive exercises in geriatric and
neurological rehabilitation wards. J. Physiotherapy (2016, accepted for publication)
13. IJsselsteijn, W.A., de Kort, Y.A.W., Midden, C., Eggen, B., van den Hoven, E.: Persuasive
technology for human well-being: setting the scene. In: IJsselsteijn, W.A., de Kort, Y.A.W.,
Midden, C., Eggen, B., van den Hoven, E. (eds.) PERSUASIVE 2006. LNCS, vol. 3962,
pp. 1–5. Springer, Heidelberg (2006)
14. Smith, S.T., Schoene, D.: The use of exercise-based videogames for training and
rehabilitation of physical function in older adults: current practice and guidelines for future
research. Aging Health. 8, 243–252 (2013)
15. Caplan, L.: Stroke. Demost Medical Publishing, New York (2006)
16. Lincoln, N.B., Kneebone, I.I., Macniven, C.M.: Psychological Management of Stroke,
pp. 3–28. Wiley, Malden (2012)
17. Fasoli, S., Krebs, H., Hogan, N.: Robotic technology and stroke rehabilitation: translating
research into practice. Top. Stroke Rehabil. 11, 11–19 (2004)
18. Maclean, N., Pound, P., Wolfe, C., Rudd, A.: Qualitative analysis of stroke patients’ in
rehabilitation. Br. Med. J. 321, 1051–1054 (2000)
19. Shaughnessy, M., Resnick, B.M., Macko, R.F.: Testing a model of post-stroke exercise
behavior. J. Rehabil. Nurs. 31, 2048–7940 (2006)
20. Bongers, A.J., Smith, S.T.: Interactivating Rehabilitation through Active Multimodal
Feedback and Guidance. In: Rocker, C., Ziefle, M. (eds.) Smart Healthcare Applications and
Services: Developments and Practices, pp. 236–260. IGI-Global, Pennsylvania (2010)
21. Hilland, T., Murphy, R., Stratton, G.: The Feasibility and Appropriateness of Utilising the
Nintendo Wii during Stroke Rehabilitation to Promote Physical Activity, A report by the
Liverpool John Moores University (2011)
22. Orji, R., Vassileva, J., Mandryk, R.: LunchTime: a slow-casual game for long-term dietary
behaviour change. Pers. Ubiquit. Comput. 17, 1211–1221 (2013)
23. Froehlich, J., Findlater, L., Landay, J.: The design of eco-feedback technology. In:
Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI
2010), pp. 1999–2008. ACM, New York (2010)
24. Bongers, A.J., Smith, S.T., Donker, V., Pickrell, M., Hall, R.: Interactive infrastructures –
physical rehabilitation modules for pervasive healthcare technology. In: Holzinger, A., Ziefle,
M., Röcker, C. (eds.) Pervasive Health – State of the Art and Beyond, pp. 229–254. Springer,
London (2014)
25. Cockrell, J., Folstein, M.F.: Mini-mental state examination. In: Copeland, J.R.M., Abou-
Saleh, M.T., Blazer, B.G. (eds.) Principles and Practice of Geriatric Psychiatry, 2nd edn, pp.
140–141. Wiley, Chichester (2002)
26. House, A.: Mood disorders after stroke a review of the evidence. Int. J. Geriatr. Psychiatry
2, 211–221 (1987)
Developing a Virtual Coach for Chronic
Patients: A User Study on the Impact
of Similarity, Familiarity and Realism
Arlette van Wissen(&), Charlotte Vinkers, and Aart van Halteren
Philips Research, High Tech Campus 34, 5656 AE Eindhoven, The Netherlands
{Arlette.van.Wissen,Charlotte.Vinkers,
Aart.van.Halteren}@philips.com
Abstract. Healthcare costs are increasing dramatically due to disproportional

consumption of healthcare resources by chronic patients. Automated forms of
health coaching can contribute to improved patient self-management while
reducing costs due to increased scalability and availability of the use of human
health coaches. Embodied Conversational Agents (ECAs) seem to be good
candidates to function as automated coaches, as they introduce a social com-
ponent to human-computer interactions which makes them particularly suitable
to influence a user’s attitude or behavior. To date, there is limited knowledge on
the impact of appearance-related characteristics of an ECA as a virtual coach
among a chronically ill elderly population. The primary aim of this study is to
investigate the impact of three appearance cues on user acceptance: (i) similar-
ity; (ii) familiarity; and (iii) realism. Findings demonstrate that patients (a) pre-
ferred the realistic-looking ECA over the more stylized one; (b) showed no
preference for the familiar over the unfamiliar ECAs (but did evaluate the
unfamiliar ECAs as more positive than the familiar one); and (c) evaluated an
ECA as virtual coach for self-management support as useful.
Keywords: Persuasive technology Socially influencing systems

Human-computer interaction Virtual coaching Embodied conversational agent
1 Introduction
The prevalence of chronic illnesses, such as diabetes and heart failure, has strikingly
increased in the past decades [35]. One viable way to keep healthcare costs within
bounds is to help patients better manage their health and illness through health
coaching (in the context of chronic disease called self-management support [17]),
which refers to coaching towards a change in lifestyle-related behaviors for improved
health and quality of life, or towards establishing and maintaining health promoting
goals [7]. In current clinical practice, health coaching is often human effort that is
labor-intensive, costly, and therefore less scalable and cost-effective compared to
automated forms of health coaching. Moreover, human health coaches (HCs) have
limited availability and reach and therefore limited impact (as impact can be considered
reach * efficacy). The reach of a virtual agent in the role of a health coach is much
broader as it is accessible to patients whenever they require support. As such there is a

DOI: 10.1007/978-3-319-31510-2_23
264 A. van Wissen et al.
need to develop innovative and interactive virtual HC services that are able to address
those limitations associated with human HC support [3, 18].
Embodied Conversational Agents (ECAs) are a promising technique to help balance
the desire for human HC involvement with the need to increase automated interventions
in healthcare context. In the present work, we describe an ECA that will function as a
virtual (health) coach (VC). The discussed prototype is designed to be the core of a more
extensive automated coach that can be used to provide advice and guidance in health
promotion and aims to persuade users to change their self-management behaviors to
engender beneficial health-related outcomes.
For such a VC to be effective, it has to maximize the influence it has on the user. In
persuasive communication there are three components (the source, the receiver, and the
messages) that can affect the effectiveness of persuasion [20]. In this paper we examine
the influence of the source, in particular, of different appearance characteristics of the
source, on its persuasiveness and acceptance among elderly chronic patients. The main
aim of our work is (1) to develop an ECA that functions as a virtual coach for chronic
patients, i.e. that is able to support and stimulate behavior change, and (2) to identify
visual characteristics of the agent that yield the highest user acceptance among our
target population.
2 Background
ECAs can provide a particularly acceptable form of human-computer interaction for

chronic patients, which are primarily older adults (e.g. [19]). ECAs are suitable for
older adults and can address challenges such as their declining sensory and physical
abilities [34] and their limited comfort with using technology based on text rather than
voice [32]. Moreover, interactive automated HC systems have been developed that
have shown to be as effective as human healthcare providers and yield advantageous
health-related outcomes such as treatment adherence [6, 24].
Despite its potential benefits in terms of ease of use and health outcomes, there is
limited evidence about the extent to which chronic patients accept the use of an ECA as
health coach, and if they do, what the ECA should look like to engender user accep-
tance. User acceptance and perceived persuasiveness of an ECA is a crucial determi-
nant of an ECA’s effectiveness, as source characteristics can unconsciously affect the
human decision making process [22, 23, 29]. Examples of these characteristics are
trustworthiness, perceived expertise, likeability, and similarity to the user.
We have designed a user study that will yield the required insights as to which ECA
characteristics should be implemented to obtain maximum user acceptance and con-
sequently potential persuasive influence and user adherence.
3 User Acceptance and Engagement
In the study, we will systematically vary (a) familiarity and similarity; and (b) the
visual rendering (realistic vs stylized) of the ECA to examine its impact on user
acceptance. With user acceptance, we refer to the acceptance of the ECA in the role of a
Developing a Virtual Coach for Chronic Patients 265
health coach, which means that the ECA should be trustworthy, likeable, and be
perceived as having the expertise to be an effective substitute for a human HC. These
factors are also known to influence persuasiveness [11, 23, 29]. The effect of these
three characteristics have yielded inconclusive and contradictory results in earlier
research, which has been primarily conducted with younger highly educated popula-
tions and with observational study designs. We therefore use an experimental design to
study whether these characteristics influence user acceptance among older patients (i.e.
the target group for the virtual health coach).
Similarity. With regard to the ECA’s similarity to the user, it is often assumed that the
agent should look similar to the user in order to obtain user acceptance and trust
(cf. [11, 30, 31]). This is also reflected in consistent findings in research on
human-human relationships that similarity breeds attraction, as advocated by the
Similarity Attraction Theory (see [31, 36]). In this study we will focus on age as proxy
for perceived similarity, as age also has consequences for another dimension that
determines how people evaluate others: power or perceived competence [13, 28]. In our
study the ECA will have the role of an expert, and as such it is important that it
is perceived to be competent and trustworthy [4]. As such, users may prefer an
older-looking ECA due to perceptions of competence and experience. On the other
hand, although our users may identify themselves more with an older-looking ECA,
they may prefer to receive health advice from younger-looking ECAs, as younger
people are often seen as more attractive, which may yield more liking, user acceptance
and persuasiveness [16, 25].
Familiarity. Another question that is not unambiguously answered in the literature is
to which extent an ECA should be a representation of a familiar person that exists in
real-life. A vast amount of research shows that the more familiar someone is, the more
a person will like and trust him or her (e.g. [9, 10]). Literature on the importance of
familiarity in ECAs is scarce, but psychologists and counselors have begun to use
virtual representations of themselves (‘doppelgangers’) to deliver therapy to clients,
which suggest that familiarity may be an important feature to implement in an ECA as
HC for chronic patients. The effect of familiarity on user acceptance is therefore another
characteristic we will examine in the present study.
Realism. There are diverging claims in the literature about the impact of realistic
versus more stylized appearances in visual rendering [15], but few empirical studies to
back up claims on either side. On the one hand, an ECA’s strength is that it emulates
a human communicator, with all its social and emotional affordances that such
a relationship brings (e.g., support) [2, 8]. On the other hand, the ‘uncanny valley’
demonstrates a thin line between very realistic ECAs and those that look uncomfortably
creepy, leading to negative user responses and discontinued system use [21]. Moreover,
people expect virtual characters to display behavior that fits their appearance: the more
human-like an ECA looks, the less forgiving they are when the ECA does not meet
expectations (e.g., [27]).
4 Virtual Coaching: Design and Use

4.1 Technical and Functional Requirements
For a virtual coach to be effective as a health coach of chronic patients, it needs to have
functionalities that reach beyond many current approaches of assistance for chronic
patients that focus mostly on diagnosis, reminders, and direct feedback of behavior.
It should be able to deploy different behavior change techniques and using those,
persuade the user to change attitudes and behaviors. Keeping our target functionality
and domain in mind, we identified several high-level Virtual Coach Requirements that
have to be met in order to achieve this:
1. the agent should support both user-initiated and system-initiated interactions.
System-initiated interactions enable pro-active pushes of personalized information
to the user. User-initiated interactions allow for a degree of control for the user.
2. the agent should be able to collect personal data from the patient to assess the
current state of the user and to provide the health care professional with up to date
information.
3. the ECA should be able to learn real-time from user input. That is, it should be
able to dynamically update the knowledge base about the user and generate
responses on-the-fly based on new information
4. assessing and reacting to the user state should be core of the ECA’s functionality.
5. the virtual coach should be capable of providing long-term support. It should
support and guide users through longer periods of change and should aim to form a
social bond with the user.
Fig. 1. Tablet with eCareCompanion interface Fig. 2. Measurement entry within eCP
4.2 Integration and Use in a Telehealth Program

The virtual coach is intended to be a fully developed ECA that can be used to provide
advice and guidance in health promotion and disease management. As a first step
towards realization of such a system, a prototype was developed that is able to talk to
the user and respond to touch-based user interactions. The core functionality of this
prototype was to provide the user with information about the importance of physical
activity, to give encouragement and to support them in setting reminders for medication
intake and appointments with health care professionals. This VC is intended to be used
in a digital environment called eCareCompanion (eCP), which is realized as a
tablet-based interface (see Fig. 1). This interface enables patients to manually enter
measurements and respond to survey questions (see Fig. 2). Patients also receive
feedback on how their measurements (e.g., blood pressure, weight) change over time.
Additionally, patients can use the app to see what consults have been scheduled and to
connect to members of their telehealth team through video-chat.
The VC is designed to complement health coaching in a telehealth program.
Although for some types of support human coaching and interventions are necessary or
preferred, in other cases a virtual coach can take on (parts of) the work of care team
members, such as health nurses, health coaches, psychologists, social workers or
clinicians. By performing tasks such as scheduling appointments, reminder patients to
upload their physiological measurements, providing surveys, or providing some form
of social support, the VC will be able to reduce the work load of the care team
members, allowing them to focus their efforts where a human touch is most needed and
providing them with relevant and regular feedback on the patient status.
5 Method
5.1 Study Aim and Design

The primary aim of the study was to investigate the independent differential impact of
the following static ECA characteristics on user acceptance: (a) a younger vs. older
appearing ECA (both unfamiliar) vs. an ECA that resembles a familiar person in
real-life, and (b) a more realistic vs a more stylized visual rendering of the ECA. These
factors were tested in a 3X2 mixed between- and within-subjects design with 3
between-subject conditions: older (unfamiliar) vs. younger (unfamiliar) vs. familiar,
and 2 within-subject conditions: stylized vs. photo-realistic. The stylized version of the
ECA was created using a software that creates a cartoonized filter on top of the original
photographs. We choose this approach because we wanted the stylized version to
resemble the realistic one as closely as possible (in order to rule out other effects of
appearance). Additionally, we decided on a female representation for the ECA, because
prior studies show that people prefer ECAs who conform to stereotypical expectations
(e.g., [12]). Given that most (health) coaches in the US are female, these findings
suggest that the ECA as HC should be female to induce the most positive responses.
Each participant was semi-randomly allocated (in order to guarantee an equal number
of participants in the different conditions) to one of the six conditions. The conditions
were operationalized as follows (see Fig. 3):
• Younger: a representation of a woman of the age 30–35
• Older: a representation of a woman of the age 55–60
• Familiar: a representation of an actual female health coach/nurse
• “stylized” versions of each picture were created using rotoscoping
5.2 Procedure
Prior to the experiment, a health coach was recruited whose representation we used to
create the ‘familiar’ ECA. Upon arrival at the research facility, participants were asked
to sit at a table and were provided a tablet. At this point the experiment leader (i.e. the
researcher in the older/younger condition or the health coach in the familiar condition)
explained the purpose of the study, after which informed consent was signed.
The Virtual Coach Application was then launched on the tablet, after which partic-
ipants were presented with the first ECA (photorealistic or stylized), with which they were
able to interact via multiple choice responses on the touch screen. The order of presen-
tation of the stylized vs. photo-realistic version of the ECA was counterbalanced between
participants. During the interaction with the VC, the VC briefly introduces herself, creates
a reminder for medication, and asks the participant a question about physical activity.
After the interaction, participants filled out an evaluation survey on the tablet, and
subsequently were presented with the second ECA (photorealistic or stylized) and some
final surveys. All versions of the ECA (older, younger, familiar) had the same voice,
pushed the same content, and were otherwise matched on appearance features to the
extent possible (e.g., hair color, skin color). Finally, the experiment leader returned to
the room for a discussion and debriefing.
Fig. 3. Appearances of the ECA in the study. Left: older-looking (unfamiliar), middle:
younger-looking (unfamiliar), right: familiar. Upper: realistic, under: stylized
5.3 Materials
Tablet. An Android tablet was used to collect data and enable interaction with the VC.
The survey data were collected via a secure link; data that participants filled out during
the interactions with the VC was not recorded.
Virtual Coach application. The Virtual Coach app presented the torso of a fully
animated talking head of an Embodied Conversational Agent. The avatar was gener-
ated with a software package called CrazyTalk1 and deployed on Android using the
Unity platform to create the interactive user experience.
Background Survey. Socio-demographic variables such as gender, age, household
composition and education level were assessed, as well as relevant psychological
constructs, such as social support and ability for self-care.
Manipulation Check. As a manipulation check (younger vs. older vs. familiar;
realistic vs. stylized), participants were asked three questions after each presentation of
the ECA. Sample item: “How humanlike do you find the virtual health coach?”; all
response options ranging from 1 (not at all) to 5 (very).
Coach Evaluation Survey. To measure user evaluations of the ECA, we used the
questions from extensive earlier research on user acceptance of ECAs by Bickmore [5]
which are based widely acknowledged models on technology acceptance (Unified
Theory of Acceptance and Use of Technology, [33]; Technology Acceptance Model,
[1]). Sample item: “How much do you like the virtual health coach?”; all response
options ranging from 1 (not at all) to 5 (very).
Attitudes Towards Use. As the use of a VC for the purpose of support with
health-related tasks in a chronically ill population has received limited attention to date,
we developed an 8-item survey based on prior research (cf. [16]) about the perceived
use and usefulness of the VC. Sample item: “I would find the virtual health coach
useful to support me with the management of my health”; all response options ranging
from 1 (not at all) to 5 (very).
6 Study Results
6.1 Participants
In total, 64 participants, of whom 34 female, participated. The majority of participants
was between 65–70 years old (62.5 %; > 70 years: 31.3 %; < 65 years: 6.3 %), highly
educated (=> associate degree: 59.4 %), living with their spouse (54.7 %), and Cau-
casian (95.3 %). All participants had one or multiple chronic illnesses for which they
had to take medications (e.g. diabetes, cardiovascular disease). Overall, the sample
reported to be comfortable with technology (M = 4.02, SD = .88), to have a high level
of social support (M = 4.13, SD = .79), and indicated a good ability to engage in
self-management, such as self-monitoring symptoms and eating a healthy diet (M range
2.83–3.56; SD range .38–.70).
1
https://www.reallusion.com/crazytalk/.
6.2 Manipulation Check

Repeated measures ANOVAs with visual rendering (realistic vs. stylized) as within
factor, and Appearance (older vs. younger vs. familiar) and Order (realistic first vs.
stylized first) as between factors showed that, as expected, the realistic version was
perceived as more human-like, similar, and familiar than the stylized version of the VC,
all p < .01. No significant differences were found on these three variables between the
older, younger, and health coach conditions.
6.3 Coach Evaluation Survey

Separate Repeated Measures ANOVAs were performed for each of the items relating to
the users’ evaluations of the VCs, with visual rendering (Realistic vs. Stylized) as
within factor, and Appearance (Older vs. Younger vs. Familiar) and Order (Realistic
first vs. Stylized first) as between factors.
Table 1. Means of evaluations of the coach evaluations of the older, younger and familiar
virtual coach
Coach evaluations Realistic Stylized
M (SD) M (SD)
Friendly Familiar 3.22 (2.02) 2.92 (2.01)
Older 4.15 (2.07) 3.44 (2.07)
Younger 3.68 (1.97) 3.23 (1.96)
Competent Familiar 3.61 (1.91) 2.92 (1.81)
Older 3.60 (1.97) 3.28 (1.87)
Younger 3.50 (1.87) 3.32 (1.77)
Attractive Familiar 3.32 (1.94) 1.93 (1.85)
Older 3.41 (1.99) 1.80 (1.90)
Younger 3.50 (1.89) 2.32 (1.80)
Like Familiar 3.29 (1.97) 2.51 (1.75)
Older 3.51 (2.02) 2.84 (1.80)
Younger 3.59 (1.92) 3.05 (1.71)
Trust Familiar 3.34 (1.67) 2.84 (1.98)
Older 3.43 (1.72) 3.00 (2.03)
Younger 3.68 (1.64) 3.32 (1.93)
Follow advice Familiar 3.58 (1.57) 3.04 (2.01)
Older 3.54 (1.62) 3.17 (2.07)
Younger 3.77 (1.53) 3.64 (1.96)
Continue interaction Familiar 3.35 (2.16) 2.47 (2.17)
Older 3.40 (2.22) 2.77 (2.24)
Younger 3.82 (2.11) 3.27 (2.13)
Satisfied Familiar 3.31 (2.06) 2.36 (2.18)
Older 3.39 (2.11) 2.87 (2.25)
Younger 3.68 (2.01) 3.23 (2.13)
For all outcome variables, a significant difference between evaluations of the

stylized compared to the realistic appearances were found. Specifically, the participants
evaluated the realistic version more positively with respect to characteristics such as
likeability, competency, trustworthiness (all p’s < .01; see Table 1).
For one outcome variable, ‘friendly’, a marginally significant main effect of
Appearance (older vs. younger vs. familiar) appeared, F(2, 57) = 2.64, p = .08,
pη2 = .08. Post hoc analysis (Least Square Difference Test) revealed that participants in
the older condition (M = 3.80, SD = 1.80) found the VC significantly friendlier than
participants in the familiar HC condition (M = 3.07, SD = 1.75, p = .03). There were no
differences between the younger condition (M = 3.46, SD = 1.71) and the other two
conditions (p’s > .21). For all other outcome variables, no differences between the
older, younger, and familiar VC were demonstrated (all p’s > .14).
6.4 Attitudes Towards Use of a Virtual Coach

Again, separate Repeated Measures ANOVAs were performed for each of the items
relating to the users’ evaluations of the VCs, with visual rendering (realistic vs. styl-
ized) as within factor, and Appearance (older vs. younger vs. familiar) and Order
(realistic first vs. stylized first) as between factors. No main effects of visual rendering
were demonstrated. However, for four of seven outcome variables, a main effect of
Appearance (older vs. younger vs. familiar) appeared, all p’s <= .05.
We found differences between participants in the extent to which they evaluated the
(intended) functionality of the VC as useful, easy to use, enabling better health goal
achievement, supporting their health management, and whether they liked working
with it (see Table 2). Post-hoc analyses revealed that for ‘easy to use’, ‘like working
with’, and ‘enable health goal achievement’, the younger VC was evaluated as sig-
nificantly more favorable than the familiar VC (p’s < = .03), whereas no differences
were found between the younger and older, and older and familiar VC. For ‘easier to
manage health’ and ‘useful’, a significant difference was also found between the older
and familiar VC, with the older VC evaluated as more favorable than the familiar VC.
Table 2. Means of evaluations of the usefulness of the older, younger and familiar virtual health
coach
Survey item Familiar Older Younger pη2
M (SD) M (SD) M (SD)
Easy to use 4.05 (.97) 4.43 (.68) 4.55 (.51)
Like working with 3.29 (.96)a 3.71 (.64)a,b 4.05 (.84)b .13
Useful to support health management 3.48 (.93) 3.76 (.89) 4.05 (.84)
Enable health goal achievement 3.38 (.92)a 3.71 (.78)a,b 4.00 (.69)b .10
Easier to manage health 3.24 (.77)a 3.52 (.98)a 4.09 (.61)b .17
Useful 3.24 (.83)a 3.52 (.93)a 4.05 (.72)b .15
Make life more interesting 3.00 (.84) 3.00 (.89) 3.50 (1.01)
Help when necessary 3.76 (.70) 3.95 (.87) 4.27 (.63)
Note. Values with different superscripts differ at p < .05
7 Discussion and Future Work
Our results show that using a familiar face for a virtual coach may pose challenges.
Compared to the other appearances, the familiar virtual coach received lower evalua-
tions. One hypothesis for these results is that people are disappointed by a virtual
representation of someone they know, as it emphasizes technological limitations (such
as imperfect lip synchronization, lack of shared knowledge between virtual and actual
coach) and the mismatch with reality (different voice, different movements). An
alternative explanation for our findings could be that they can be contributed to the
particular images that were used. Future work could explore this further. Furthermore,
our results indicate that people have slight preference for a younger appearance over an
older one. It could be that they found her more attractive and as such they might have a
cognitive bias in favor of her (see [16, 25]).
Overall, people were positive about their interaction with the Virtual Health Coach
and they have a positive attitude towards the deployment of an ECA to support with
health related tasks. This indicates that such a virtual coach can indeed be deployed as
an aid to an elderly chronic patient population with respect to behavior change and
self-management support. This is in line with results found in [14, 26], in which the use
of ECAs as a medium for persuasive dialogue was examined. Future research will have
to show whether the ECA will be preferred over merely text- or voice-based interac-
tions, given that the acceptance of voice-based assistants has improved dramatically
over the last years (e.g. Siri by Apple, Google Now, and Cortana by Microsoft).
With respect to our study design, some limitations have to be noted. With respect to
the analyses, we are aware of the discussion as to whether 5-point Likert scales should
be considered ordinal or interval data, and analyzed as such. In this case we argue that
the items can be considered symmetric and equidistant and as such will behave like an
interval-level measurement which measures continuous concepts. Furthermore, in the
study we chose to create an environment in which patient and health coach could
become familiar with each other, rather than working with someone who was actually
known to the patients. This was decided because of (a) scalability of the study (so as
not to have to create 64 different ECAs from photos of different health coaches), and
(b) participants’ opinions about their health coach, which may bias their evaluations of
the ECA resembling this person. We do however acknowledge that the degree of
familiarity is somewhat limited and as such we would in the future like to explore
setups with truly familiar subjects and ECAs.
8 Conclusion
This paper discusses part of the development process of a virtual coach for elderly
chronic patients. The virtual coach will be a tablet-based application integrated in a
telehealth program where it will function to monitor the patient and support with taking
measurements, as well as to use more complex intervention strategies to provide (pro)
active and persuasive coaching and social support. In order to achieve maximum user
acceptance and persuasive potential, a study was conducted in which the effects of
changes in familiarity, similarity and visual rendering of the ECA were examined.
Findings demonstrated that participants preferred the realistic-looking ECA over the
more stylized one; as well as had a slight preference for the unfamiliar ECAs over the
familiar ones. These findings suggest that when designing ECAs as virtual coaches,
their appearance should be realistic in order to mimic human coaching relationships,
and that using representations of loved/familiar ones is not (yet) recommended due to
the heightened expectations that current state-of-the-art technology has difficulty to
address. Additionally, the participants considered a virtual coach for self-management
beneficial to their health. As user acceptance of technology among the target population
is a requirement for (long-term) system use, we consider this user study a crucial step in
the development of virtual health coaches for chronic patients.
References
1. Bagozzi, R.P., Davis, F.D., Warshaw, P.R.: Development and test of a theory of
technological learning and usage. Hum. Relat. 45(7), 659–686 (1992)
2. Bailenson, J., Swinth, K., Hoyt, C., Persky, S., Dimov, A., Blascovich, J.: The independent
and interactive effects of embodied conversational agent appearance and behavior on
self-report, cognitive, and behavioral markers of copresence in virtual immersive
environments. Presence 14, 379–393 (2005)
3. Barlow, J., Wright, C., Sheasby, J., Turner, A., Hainsworth, J.: Self-management approaches
for people with chronic conditions: a review. Patient Educ. Couns. 48, 177–187 (2002)
4. Beishuizen, J., Hof, E., Putten, C., Bouwmeester, S., Asscher, J.: Students’ and teachers’
cognitions about good teachers. Br. J. Educ. Psychol. 71, 185–201 (2001)
5. Bickmore, T., Caruso, L., Clough-Gorr, K., Heeren, T.: It’s just like you talk to a friend:
relational agents for older adults. Interact. Comput. HCI Older Popul. 17, 711–735 (2005)
6. Bickmore, T., Giorgino, T.: Health dialog systems for patients and consumers. J. Biomed.
Inform. 39, 556–571 (2006)
7. Butterworth, S., Linden, A., McClay, W., Leo, M.: Effect of motivational interviewing-
based health coaching on employees’ physical and mental status. J. Occup. Health Psychol.
11, 358–365 (2006)
8. Campbell, R.H., Grimshaw, M.N., Green, G.M.: Relational agents: a critical review. Open
Virtual Reality J. 1, 1–7 (2009)
9. Chu, S., Fedorovskaya, E., Quek, F., Snyder, J.: The effect of familiarity on perceived
interestingness of images. In: Human Vision and Electronic Engineering, vol. 8651 (2013)
10. Cialdini, R.B., Goldstein, N.J.: Social influence: compliance and conformity. Ann. Rev.
Psychol. 55, 591–621 (2004)
12. Forlizzi, J., Zimmerman, J., Mancuso, V., Kwak, S.: How interface agents affect interaction
between humans and computers. In: Proceedings of the 2007 Conference on Designing
Pleasurable Products and Interfaces, pp. 209–221. ACM (2007)
13. Georgesen, J., Harris, M.: The balance of power: interpersonal consequences of differential
power and expectancies. Pers. Soc. Psychol. Bull. 26, 1239–1257 (2000)
14. Grolleman, J., van Dijk, B., Nijholt, A., van Emst, A.: Break the habit! designing an e-therapy
intervention using a virtual coach in aid of smoking cessation. In: IJsselsteijn, W.A.,
de Kort, Y.A., Midden, C., Eggen, B., van den Hoven, E. (eds.) PERSUASIVE 2006. LNCS,
15. Haake, M., Gulz, A.: Visual realism and virtual pedagogical agents. In: The 3rd International
Design and Engagability Conference@ NordiCHI 2006 (2006)
16. Khan, R., De Angeli, A.: The attractiveness stereotype in the evaluation of embodied
conversational agents. In: Gross, T., Gulliksen, J., Kotzé, P., Oestreicher, L., Palanque, P.,
Prates, R.O., Winckler, M. (eds.) INTERACT 2009. LNCS, vol. 5726, pp. 85–97. Springer,
Heidelberg (2009)
17. Lorig, K., Holman, H.: Self-management education: history, definition, outcomes, and
mechanisms. Ann. Behav. Med. 26, 1–7 (2003)
18. Lorig, K., Ritter, P., Stewart, A.L., Holman, R.: Chronic disease management program:
2-year health status and healthcare utilization costs. Med. Care 39, 1217–1223 (2001)
19. Mendis, S., Puska, P., Norrving, B.: Global Atlas on Cardiovascular Disease Prevention and
Control, pp. 2–14. World Health Organization, Geneva (2011)
20. Miller, G.R.: On being persuaded: some basic distinctions. In: Roloff, M.E., Miller, G.R.
(eds.) Persuasion: New Directions in Theory and Research, pp. 11–28. Sage, Beverly Hills
(1980)
21. Mori, M.: The uncanny valley (K.F. MacDorman and N. Kageki, Trans.). IEEE Robot.
Autom. Mag. 19(2), 98–100 (1970/2012)
22. Nguyen, H., Masthoff, J.: Is it me or is it what i say? source image and persuasion. In: de
Kort, Y.A., IJsselsteijn, W.A., Midden, C., Eggen, B., Fogg, B.J. (eds.) PERSUASIVE
23. O’Keefe, J.D.: Persuasion: Theory and Research. Sage, Newbury Park (1990)
24. Revere, D., Dunbar, P.: Review of computer-generated outpatient health behavior
interventions: clinical encounters in absentia. J. Am. Med. Inform. Assoc. 8, 62–79 (2001)
25. Ring, L., Utami, D., Bickmore, T.: The right agent for the job? In: Bickmore, T., Marsella,
S., Sidner, C. (eds.) IVA 2014. LNCS, vol. 8637, pp. 374–384. Springer, Heidelberg (2014)
26. Schulman, D., Bickmore, T.W., Sidner, C.L.: An intelligent conversational agent for
promoting long-term health behavior change using motivational interviewing. In: AAAI
Spring Symposium: AI and Health Communication, March 2011
27. Slater, M., Steed, A., Chrysanthou, Y.: Computer Graphics and Virtual Environments: From
Realism to Real-Time. Pearson Education, Boston (2002)
28. Svennevig, J.: Getting Acquainted in Conversation: A Study of Initial Interactions, vol. 64.
John Benjamins Publishing, Amsterdam (2000)
29. Stiff, J.B., Mongeau, P.A.: Persuasive Communication. Guilford Press, New York (2003)
30. Van Vugt, H., Bailenson, J., Hoorn, J.F., Konijn, F.: Effects of facial similarity on user
responses to embodied agents. ACM Trans. Comput. Hum. Interact. 17, 7 (2010)
31. Van Vugt, H.C., Konijn, E.A., Hoorn, J.F., Keur, I., Eliéns, A.: Realism is not all! User
engagement with task-related interface characters. Interact. Comput. 19(2), 267–280 (2007)
32. Vardoulakis, L.P., Ring, L., Barry, B., Sidner, C., Bickmore, T.: Designing relational agents
as long term social companions for older adults. Intell. Virtual Agents 7502, 289–302 (2011)
33. Venkatesh, V., Zhang, X.: Unified theory of acceptance and use of technology: US vs.
China. J. Glob. Inf. Technol. Manage. 13(1), 5–27 (2010)
34. Wallace, J., Mulvenna, M., Martin, S., Stephens, S., Burns, W.: ICT interface design for ageing
people and people with dementia. In: Mulvenna, M.D., Nugent, C.D. (eds.) Supporting People
with Dementia Using Pervasive Health Technologies. Advanced Information and Knowledge
Processing, pp. 165–188. Springer, Heidelberg (2010)
35. Ward, B., Schiller, J., Goodman, R.: Multiple chronic conditions among U.S. adults: A 2012
update. Prev. Chronic Dis. 11, E62 (2014)
36. Zhou, S., Bickmore, T., Paasche-Orlow, M., Jack, B.: Agent-user concordance and
satisfaction with a virtual hospital discharge nurse. In: Bickmore, T., Marsella, S., Sidner, C.
(eds.) IVA 2014. LNCS, vol. 8637, pp. 528–541. Springer, Heidelberg (2014)
Improving Adherence in Automated e-Coaching
A Case from Insomnia Therapy
Robbert Jan Beun1 ✉ , Willem-Paul Brinkman2, Siska Fitrianie1,

( )
Fiemke Griffioen-Both1, Corine Horsch2, Jaap Lancee3, and Sandor Spruit1

1
Utrecht University, Utrecht, The Netherlands
{r.j.beun,s.fitrianie,f.griffioen-both,a.g.l.spruit}@uu.nl
2
Delft University of Technology, Delft, The Netherlands
3
University of Amsterdam, Amsterdam, The Netherlands
[email protected]
Abstract. Non-adherence is considered a problem that seriously undermines the

outcome of behavior change therapies, in particular of self-help therapies deliv‐
ered without human interference. This paper presents the design rationale behind
a computer system in the domain of adherence enhancing strategies in automated
e-coaching. A variety of persuasive strategies is introduced and implemented in
a mobile e-coaching system in the domain of insomnia therapy. The system inte‐
grates two types of interface elements, i.e. dedicated tools and natural language
conversation, to simplify therapy related activities and to include social strategies
to improve motivation. We focus on the crucial role of communication and
adaptation.
Keywords: e-Coaching · Adherence · Persuasive strategies · Insomnia therapy ·

Conversation · Tools · Behavior change systems
1 Introduction
One of the first software programs that automatically mimicked the behavior of a thera‐
pist is the digital Rogerian psychoanalyst ‘Eliza’ [26]. Eliza simulates a simple question-
answer conversation in natural language by rephrasing many of the patient’s statements
into questions. The program showed that a digital therapist could be implemented in a
computer system that has strong positive emotional effects with relatively simple means.
Since the appearance of Eliza the world changed radically. In the digital world,
powerful mobile systems support the majority of the population with a wide range of
activities; in health care, awareness increased that our health can be positively influenced
by our behavior and activities. In the field of consumer health care, these developments
resulted in a variety of products, ranging from various types of monitoring systems to
self-help apps that aim to change an individual’s behavior and cognition. Today over
100,000 health apps exist in the Apple and Google stores that support all types of health
related activities, but a major problem is that a proliferation of consumer health care

DOI: 10.1007/978-3-319-31510-2_24
Improving Adherence in Automated E-Coaching 277
applications arises regardless of real user needs [21]. Potentially, we now have the
knowledge and the technology to build interventions and coaching principles in smart‐
phones that are far beyond the capabilities of Weizenbaum’s Eliza, but the design of
these systems should be preceded by careful analysis of the coaching process.
In this paper, we present the design rationale behind a fully automated coaching
system that supports an intervention for behavior change in the domain of insomnia
therapy – we refer to this system as the ‘SleepCare-system’. The system behaves as a
digital agent, the so-called e-coach, and integrates various types of interaction, in partic‐
ular natural language conversation and graphical interaction with various tools. We
focus in particular on the requirements and the generic elements in the design that were
included to improve exercise adherence.
We first elaborate on the coaching process and some existing e-coaching implemen‐
tations. We then turn to the domain of the application, insomnia therapy, and the problem
of exercise adherence within the domain. Next, we discuss various adherence enhancing
strategies to improve behavior change. We then turn to the design elements of the system
that support these strategies and consider some aspects of testing. We finish this paper
with a discussion about the relevant design elements.
2 Coaching and e-Coaching
Central to most definitions of coaching is that coaching is a goal-oriented activity to help

people enhance aspects of both their personal and professional lives by fostering self-
directed learning through collaborative goal setting, action planning and feedback [14].
Assumptions in the coaching process are the absence of serious mental health problems
[6] and the notion that the client – or coachee – is resourceful and motivated to engage
in finding solutions [1, 17]. In the coaching process, two learning dimensions are essen‐
tial: (1) learning through individual subjective experiences and (2) learning as a social
and collaborative practice. The first dimension refers to the experiential and action
oriented process of the coachee; the second refers to the idea of a collaborative dialogue
that unfolds between the coaching parties and where the coachee learns in interaction
with the coach.
Since current society and daily lives of people are highly contextual and character‐
ized by a growing degree of uncertainty, a coach will be cautious in offering solutions.
What counts as a solution for one person, not necessarily counts as a solution for another.
Therefore, coaching involves a collaborative approach and respect of the coachee’s
autonomy. Personalization, contextualization and frequent adaptation are necessary
prerequisites of the coaching process. In that respect, the role of a coach seems an
excellent candidate for a digital agent that supports automated self-help therapies; we
will refer to these systems as e-coaches.
Examples of e-coaches that apply methods from behavior medicine cover a wide
range of therapy domains, ranging from obesity to depression and insomnia treat‐
ment. In [5], for example, an e-coach was developed to support overweight people
improving their lifestyle. The e-coach is able to help motivated participants adhere
to the program and lose weight. In [7], an e-coach is described that offers a fully
automated treatment for depression, based on behavioral activation, a form of
278 R.J. Beun et al.
psychotherapy. In [9] an e-coach is described that offers an insomnia treatment in six

weekly sessions. The authors successfully tested the intervention against a placebo
and treatment as usual in a randomized controlled trial.
Previous studies mainly focus on one specific domain. An important exception to
domain specificity is the work described in [4] who describe a theory-driven computa‐
tional model to deliver health behavior change counseling. The authors present an
ontology containing concepts from behavioral medicine and communication. Various
models are introduced that enable the representation of knowledge about, for instance,
the user and the underlying theory of the intervention. We also aim at developing a
reusable mobile framework based on generic coaching principles, but in contrast to [4]
who focuses on modeling counseling knowledge from which dialogue actions can be
inferred, our perspective starts from communication and cybernetics theory, and is based
on concepts such as alignment, negotiation, evaluation, feedback and adaptation. This
enables us to include persuasive strategies such as tunneling, adaptation, reduction and
increase of motivation. The insomnia domain will be used as a proof of concept.
3 Insomnia and Insomnia Therapy
Insomnia is a sleep disorder with a high prevalence (about 10 % of the population) and
can have severe individual and societal consequences (e.g., concentration problems,
increased risks of accidents, depression); people with insomnia have difficulty initiating
and/or maintaining sleep. Today, it is widely accepted that behavior change treatments
such as Cognitive Behavior Therapy (CBT) produce sustainable positive changes in the
condition of insomnia [22].
CBT for insomnia (CBT-I) offers a variety of exercise types that differ in aim and
properties [23]: sleep restriction, stimulus control, relaxation, cognitive therapy and
sleep hygiene. Sleep restriction involves curtailing the time spent in bed to stabilize the
sleep pattern and lengthening sleep time as sleep efficiency improves. Stimulus control
aims at restoring the coachee’s positive association of the bed and the bedroom with
sleep. Relaxation training involves methods aimed at reducing somatic tension. Cogni‐
tive therapy aims at changing dysfunctional beliefs and attitudes. Sleep hygiene educa‐
tion aims to make the person aware of practices and environmental factors that may
either be detrimental or beneficial for sleep. Treatment protocols usually take between
6 and 10 weekly consultation sessions. The actual intervention is preceded by a one- or
two-week baseline sleep diary monitoring period.
4 Exercise Adherence
Central in self-management approaches to behavior change for health – and CBT-I is

no exception to this – is the idea that the individual invests a reasonable amount of effort
and time in the activities. However, people have many reasons not to perform the
assignments: lack of energy, motivation and willpower, mistrust, procrastination, simply
forgetting, etcetera. As a result, low adherence rates are an ever present and complex
problem, and seriously undermine the outcome of CBT [25]. In [15], it has been shown
that the average treatment adherence for technology mediated insomnia treatments is
approximately 52 %, meaning that almost half of the people did not finish the treatment.
It may even be expected that in reality these rates are considerably lower. First, since
these results were based on self-reports and user logs, the reports by the participants
may be overrated and based on a positive answer bias. Second, adherence is a complex
notion and has no clear-cut boundaries (e.g., total time in bed, number of relaxation
exercises per day, completion of the assignments). Low adherence rates can be expected
in particular with exercise types that require strong willpower.
Preventing the process of a downward motivation spiral and adequately responding
to it is a mandatory element in any behavior change intervention. For that, we should
realize that the exercises in insomnia treatment include a large variation of activities in
terms of actual content, duration, timing, frequency, presentation and intensity (cf. [12]).
Some of these activities take time and should be performed on a daily or weekly basis
such as relaxation exercises or consults. Sleep restriction and stimulus control require a
great deal of willpower and can even be dangerous, but may have high treatment effects
if well adhered to. The sleep hygiene exercise may require changing habits such as
refrain from drinking coffee or alcohol before bedtime; in other cases, the exercise
requires a one-time behavior such as cleaning the bedroom or changing the bedroom
temperature. Improving adherence to this variation of therapy elements requires a well-
dosed combination of various adherence enhancing strategies.
5 Adherence Enhancing Strategies
5.1 Tunneling, Feedback and Adaptation
Basically, every CBT-coach offers evidence based behavior change techniques that
should be experienced by the coachee for a particular period of time. The behavior in
the CBT-I domain consists of introducing or changing activities in the coachee’s daily
life such as filling in a sleep diary, having a conversation for consultation or changing
bedtime; in other cases, activities and habits are discouraged such as drinking coffee or
alcohol just before bedtime. In other words, a coach tunnels therapy related activities:
it determines the exercise types, the properties of these exercises and communicates this
information to the coachee. We call this process tunneling (cf. [11]).
In contrast to self-help books, a human coach adds an important quality to the
tunneling process: a feedback loop. To cope with the many uncertainties, a frequent
feedback loop is included to reshape the offered techniques and the communication about
these techniques. Hence, the information flow between coach and coachee not only
pertains to tunneling (e.g., ‘Go to bed at 12’, ‘Fill in your sleep diary before 8’), but also
to the adaptation process. A human coach observes, asks questions, negotiates, agrees,
aligns and tailors the exercises to the multifarious coachee characteristics such as age,
habits, preferences, intermediary results and experiences [3]. Adaptation is a crucial
element in the tunneling process. A person who, for whatever reason, is unable to go to
bed at twelve and repeatedly hears from her coach that she should go to bed at twelve,
most likely decides that the therapy is not applicable and may even consider the coach
as ignorant or unreliable. This is one of the reasons that the results of many self-help
books are rather disappointing [24]: they are targeted at large user groups and do not
offer information or exercises tailored to the individual.
5.2 Persuasive Strategies

Coaches also apply a range of behaviors to support the behavior change: they motivate,
encourage, challenge, and explain; they show progress and confront with discrepancies
in real and committed behaviors. In that sense, coaching is a constant process of
tunneling, tailoring and support. We refer to a collection of coherent activities that
support the process of behavior change as a persuasive strategy. In line with [13, 20],
we assume that persuasive strategies aim at increasing the ability and/or motivation of
the coachee with respect to the intended behavior change. Loosely speaking, improving
ability implies that the coachee may decrease the amount of effort that should be put
into the therapy; improving motivation implies that the coachee is willing to put more
effort into the therapy. The distinction between the two concepts is not always that clear,
however.
We assume that the coachee is located in a complex multidimensional space, the
current state, and aims to achieve another complex space, the desired state. In insomnia
therapy, both states can be expressed in concrete values for sleep variables such as total
sleeping time and quality of sleep. In the desired state, the coachee has ‘improved’ on
at least some of these values. To stress the importance of the difference between the use
of tools and language (cf. [11]), persuasive strategies are now classified in two categories
(cf. [16]): (1) polishing strategies and (2) meta-level strategies.
ad 1. The road to the desired world is full of obstacles. An important function of the
coach is to take away these obstacles and polish the way to success. Metaphorically
speaking, the coach offers a slide to improve the flow of activities that have to be
performed to achieve the desired world. It may simplify behavior and help to avoid
boredom and stress; it may include beauty or a positive user experience to attract
attention and willingness to participate. In practice, this may imply that the coach
offers tools, e.g., an electronic sleep diary, an agenda, relaxing music or a game. This
is also related to what is often called ‘reduction’ and ‘nudging’. In insomnia therapy,
for instance, a tool may be implemented that automatically lowers the intensity or
changes the color of the light to induce homeostatic processes in the body for bedtime
preparation.
ad 2. The meta-level adds a symbolic level to the previous category. It is what we often
call ‘social influence’ and may, for instance, be represented in the interface as an
embodied character that uses natural language dialogue and nonverbal signs. The
coach is now able, for instance, to communicate that ‘if the coachee performs an
activity, then the coachee will be rewarded’. The use of symbols enables the coach to
approve or disapprove the coachee’s behavior, to communicate about the road to the
desired state and to explain matters in terms of cause and consequence. The meta-level
also enables to negotiate the intensity of the intervention, or to show some type of
authority (e.g., ‘This is an evidence based intervention’). Including the meta-level
enables us to incorporate social elements that have a high impact on a person’s
behavior: collaboration, transparency, competition, trust, commitment, reciprocity

(see e.g., [8]) and accompanying speech acts such as promise, praise, welcome, bid,
finish the conversation, etcetera. The meta-level substantially increases the expres‐
sivity of the coaching system; in practice, the levels are constantly intertwined.
In CBT-I two types of activities can be distinguished: a. main activities that have to
be experienced by the coachee such as sleep restriction and b. supporting activities such
as activity scheduling. To restructure the coachee’s sleep architecture, it is important to
experience sleep pressure, but it is unimportant to experience calculating sleep effi‐
ciency. Consequently, all activities that are part of CBT-I and that do not have to be
experienced by the coachee can be simplified. In CBT-I, for instance, sleep data can be
registered by means of an electronic diary instead of paper and pencil notation; calcu‐
lations of sleep efficiency can be automated and integrated in a proposal for wake up
and bedtimes. In the ideal case, supporting activities are fully automated. Because main
activities, such as decreasing total bedtime, should be experienced, it is important to
increase motivation for these types of activities.
To conclude, an e-coach that is intended to offer an intervention for behavior change
should include a variety of adherence enhancing strategies that aim at:
• tunneling the intended activities,
• adaptation of exercises and interaction,
• reduction by simplification of supporting activities,
• increasing motivation to perform main activities (cf. [19]).
We now turn to the design of the SleepCare-system and explain how these requirements
were applied in the system.
6 The SleepCare System
6.1 The Functionality of the e-Coach

The SleepCare-system offers a therapy on an Android smartphone that includes a variety
of CBT-I exercises: relaxation, sleep restriction and sleep hygiene. Sleep restriction has
been chosen, because it is an effective element of CBT-I; relaxation and sleep hygiene
were chosen because these exercises were relatively easily to implement and vary
substantially from sleep restriction. Exercises consist of assignments that have to be
carried out during the therapy. There are five types of assignments: going to bed, getting
out of bed, filling in the sleep diary, doing a relaxation exercise and taking part in a
consult. A consult is a conversational activity about various therapy related topics such
as the introduction and evaluation of a particular exercise type.
In the therapy, three interaction stages are distinguished: opening, intervention and
closure [3]. The goal of the opening phase is twofold: improving transparency through
a process of alignment and establishing commitment to the therapy. In the intervention
phase the actual therapy is conducted and the coachee is supposed to carry out the
assignments. The closure phase starts when all assignments have been performed or
when the coachee indicates the desire to withdraw. In this phase, the e-coach and coachee
evaluate the offered therapy and may say goodbye.
6.2 Tunneling the Coachee’s Activities

A fundamental element in the process of tunneling is the activity schedule that is intro‐
duced by the e-coach in one of the early introduction sessions. The primary task of the
activity schedule is to keep track of the various assignments that have been or should
be carried out during the therapy. All scheduled assignments result from an agreed
contract between coach and coachee. The end time of the activities is usually unknown,
but in practice the time interval of assignments is relatively short (e.g., between 1 and
16 min). Scheduled assignments may trigger the generation of reminders and other
communication acts. Activities by the e-coach may also be triggered by non-scheduled
events, in particular when the system detects violations of pre-defined ‘constraints’ such
as obvious non-adherence of the coachee [3].
To observe the properties of the scheduled assignments, such as starting-time and
performance status, there exists a corresponding interaction tool. The most important
function of the tool is to display the coachee’s previous commitments and adherence to
the assignments. For that, the interaction tool not only displays the scheduled starting
time and date in the past, present or future, but also the performance status of an assign‐
ment by a colored line in front: red means missed, orange means near missed, green
means completed and grey means planned or actual. The left picture in Fig. 1 shows a
part of the interaction tool presented in the system’s home screen. As a consequence,
the tool not only enables a coachee to have a prompt overview of the future commit‐
ments, but also of the (non-)adhered assignments in the past. In that way, the schedule-
tool not only tunnels the coachee’s activities, but also refers to meta-level properties
such as agreed commitments.
Fig. 1. The left picture depicts the home screen of the SleepCare-system and a part of the schedule
tool; colored vertical lines indicate the performance status of the exercises. The middle and right
picture show examples of the sleep diary tool and an evaluation dialogue about the use of the sleep
diary. On top of the interface notification icons appear in the notification area (the small black
beam).
6.3 Feedback and Adaptation

Starting from the opening phase, there is a constant activity of feedback and adaptation.
From the e-coach’s point of view, the goal of the opening phase is to construct a user
model to adapt the therapy to the characteristics of the coachee. The adaptation process
pertains to information at both the therapy and the communication level (e.g., sleep
characteristics, age and name). Based on a set of exclusion criteria, the coach may also
advise the coachee to withdraw from the therapy; the coachee, however, makes the final
decision about continuation.
During the intervention phase, exercises always follow the same four steps [3]:
1. introduction, 2. planning and commitment, 3. task execution and 4. evaluation. In the
introduction step, the exercise type is explained. In the planning and commitment step,
e-coach and coachee aim at agreement about the exercise properties. During task execu‐
tion the coachee performs the assignments. In the evaluation step, results are discussed and
the exercise properties may, in deliberation with the e-coach, be adapted to the coachee’s
desires. In case of exercise adaptation, the process is repeated, starting from the planning
and commitment phase. If no constraints are violated, the planning-evaluation cycle takes
about one week. All exercise steps are included in the activity schedule, but in the course
of the therapy, assignments may be added, rescheduled or even be removed in deliberation
with the e-coach.
6.4 The User Interface
In the system, four categories of interface elements are distinguished: conversations,

tools, notifications and the menu. Conversations enable the e-coach to perform meta-
level natural language interaction activities such as introductions and weekly consults.
Tools are relatively independent modules that are less language oriented and contain
graphical elements such as buttons, wheels and tables; we included the schedule-tool,
an electronic sleep diary and a relaxation tool. Notifications function as attention grab‐
bers to start a particular activity and are displayed outside the system’s normal user
interface (see Fig. 1 middle and right pictures). The menu offers access to the various
tools, performed conversations, sleep exercises, advice and information, and general
issues such as notification settings, privacy and safety.
Although simplistic compared to human F2F-conversations, the use of natural
language interaction enables us to include a plethora of conversational structures that
resemble well-known collaboration and persuasion strategies at the meta-level. As such,
conversations are the central interaction concept in the system. They not only explain
the assignments (what, how and why) and what can be expected (e.g., difficulties,
results), but also function as an instrument for adaptation, shared commitment and
explicit motivational support such as praise and encouragement.
To elucidate a conversational interaction, we briefly focus on one of the meta-level
strategies used in the SleepCare-system: the establishment of a shared commitment by
negotiation. An adapted version of a conversation about the sleep restriction exercise is
shown in the dialogue fragment below (translated from Dutch; E: e-coach, C: coachee).
In this fragment, the coachee options are given between brackets (selected options in
bold); values of variables based on the e-coach’s knowledge base are represented in
italics. In this case, we have chosen a persuasive strategy of shared decision making,
because sleep restriction is an intensive exercise and can even be dangerous in certain
circumstances (e.g., [18]). The negotiation space depends on the user model, in particular
total bedtime and sleep efficiency, and a number of rules, for instance, total bedtime
should never be shorter than 5 h.
E Good morning John. This week we will start the sleep restriction exercise. You have
to keep regular bed times and should shorten the total time in bed. Because you have
less time to sleep, you will be more tired and that helps for a better sleep. Would you
like to know more before we start?
C [Let’s start] [What do I have to do?] [What can I expect?]
E We will now determine your bedtimes. On the basis of your sleep diary I suggest
that the total bedtime will be 5 h. Do you agree?
C [Yes] [I prefer longer][I don’t want to do bedtime restriction]
E How long do you want to stay in bed?
C [5.30][6.00][6.30]
E You suggested 6.30 h. The exercise works better when you stay in bed as long as
your mean sleeping time last week. I therefore suggest that the total time in bed will
be 5.30. Do you agree?
C [Yes][I prefer longer]
E …
To cover the main interaction needs in CBT-I, 18 different conversational types were
included (e.g., intro, planning and commitment, evaluation). These conversations can
be considered as tree like structures where selections of the coachee or the value of
particular sleep and behavior data causes a branch. Half of them are scheduled based,
and the others are constraint-based conversations. Conversations are generated using
XML-based predefined templates for describing the discourse moves of the e-coach [10].
All conversations (including the branches) were evaluated by an expert in Dutch
communication and a sleep therapist.
6.5 Testing
The complex nature of the system and the application domain, the multimodal nature
of the interaction, the large amounts of collected user knowledge and the many
uncertainties in the use of mobile applications make the testing procedure a chal‐
lenge. In particular, the variety in the circumstances of use is one of the main diffi‐
culties in the development of a smartphone application. Different versions of
Android and smartphone manufacturers and changes thereof during use, network
availability, unexpected user behavior and changes in the environment may cause a
great deal of uncertainty in the outcome of the offered therapy. This not only
requires that the system functions flawlessly, but also that the e-coach offers a
therapy that is safe and in line with the advice of human therapists even under
extreme circumstances.
For testing purposes, scenarios were written together with a sleep psychologist, making
sure that the scenarios and user data covered the most common sleeping patterns and
extremes on both ends as well. The scenarios were simulated and recorded in videos. A group
of four sleep experts watched and commented on the behavior of the e-coach in the videos,
providing us with information on the validity of the offered therapy. They focused in partic‐
ular on the critical scenarios in which people sleep less than five hours per night.
Next, a pilot experiment was organized to test the SleepCare-system with intended
users. 24 people asynchronously started using the app in their daily life and focused in
particular on usability issues, the interaction with the e-coach and the circumstances of
a possible system break down. Main success criteria for continuation were: a. no
decrease in sleep efficiency and sleep quality, and b. no technical failure. The results
gave us enough confidence in the system and the training program to execute a random‐
ized controlled trial (RCT) as a next step in the research process.
7 Discussion
In this paper, strategies were discussed for enhancing adherence in behavior changing
therapies delivered by digital systems. As a proof of concept, some of the strategies were
implemented in an automated e-coaching system that supports insomnia therapy – the
so-called SleepCare-system. An interface was realized that distinguishes between graph‐
ical tools and natural language conversations. Conversations enable us to include meta-
level persuasive strategies based on social influence, such as commitment and shared
decision making. We have stressed the importance of a design rationale and a careful
analysis of the coaching process.
We briefly recapitulate the strategies discussed in this paper and their implementation
in the SleepCare-system:
Tunneling: the e-coach should communicate the intended activities at the right moment
in the right form. In SleepCare a sophisticated activity schedule, notifications and
reminders were included.
Adaptation: exercises and interaction should be personalized and adapted to the
circumstances. In SleepCare an explicit alignment phase and frequent feedback loops
were included.
Reduction: therapy related activities that do not have to be experienced can be
supported by tools or automated. In SleepCare tools were included for handling sleep
data (e.g., registration, overviews), scheduling and relaxation.
Increasing motivation: therapy related activities that must be experienced can be
supported by increasing the coachee’s motivation. Examples in SleepCare are:
showing (non-)adherence, performance and progress, shared decision making, explicit
user commitment, explanation of therapy and exercise rationale, and expectation
management.
Our first activity for the future is experimental verification of the SleepCare-system
by means of the RCT. Next we will extend the adherence enhancing strategies to other
domains and user groups, for example shift workers, and include automated sleep
registration by non-obtrusive sensor information.
Finally, we may ask what we have learned since the appearance of Weizenbaum’s
digital therapist Eliza. We believe that in the future, consumers in e-health will not only
be confronted with a variety of tools for monitoring health related characteristics and
telecommunication with human caretakers, but also with automated e-coaches or e-
doctors that direct a user’s behavior. Computer systems aiming at behavior change will
be equipped with adherence enhancing techniques to ‘seduce’ their users to do the
therapy activities. This not only raises interesting philosophical questions, but also
questions with respect to ethical design. Clearly, extensive testing is a mandatory
element, but we also believe that explicit strategies should be included that protect the
user’s autonomy. We have the tools and the language, we now need the knowledge.
Acknowledgements. This research is supported by Philips and Technology Foundation STW,

Nationaal Initiatief Hersenen en Cognitie NIHC under the Partnership programme Healthy
Lifestyle Solutions.
References
1. Berg, I.K., Szabo, P.: Brief Coaching for Lasting Solutions. W.W. Norton, NY (2005)
2. Beun, R.J.: Persuasive strategies in mobile insomnia therapy: alignment, adaptation, and
motivational support. Pers. Ubiquit. Comput. 17(6), 1187–1195 (2013)
3. Beun, R.J., Griffioen-Both, F., Ahn, R., Fitrianie, S., Lancee, J.: Modeling interaction in
automated e-coaching: a case from insomnia therapy. In: 6th International Conference on
Advanced Cognitive Technologies and Applications (2014)
4. Bickmore, T.W., Schulman, D., Sidner, C.L.: A reusable framework for health counseling
dialogue systems based on behavioral medicine ontology. J. Biomed. Inform. 44, 183–197
(2011)
5. Blanson Henkemans, O.A., van der Boog, P.J.M., Lindenberg, J., van der Mast, C.A.P.G.,
Neerincx, M.A., Zwetsloot-Schonk, B.J.H.M.: An online lifestyle diary with a persuasive
computer assistant providing feedback on self-management. Technol. Health Care 17(3),
253–257 (2009)
6. Bluckert, P.: The similarities and differences between coaching and therapy. Ind. Commercial
Training 37(2), 91–96 (2005)
7. Both, F., Cuijpers, P., Hoogendoorn, M., Klein, M.: Towards fully automated psychotherapy
for adults: BAS - behavioral activation scheduling via web and mobile phone. In: 3rd
International Conference on Health Informatics 2010, pp. 375–380 (2010)
8. Cialdini, R.B.: Influence: Science and Practice. Pearson, New York (1993)
9. Espie, C.A., Kyle, S.D., Williams, C., Ong, J.J., Douglas, N.J., Hames, P., Brown, J.S.L.: A
randomized, placebo-controlled trial of online cognitive behavioral therapy for chronic
insomnia disorder delivered via an automated media-rich web application. Sleep 35(6), 769–
781 (2012)
10. Fitrianie, S., Griffioen-Both, F., Spruit, S., Lancee, J., Beun, R.J.: Automated dialogue
generation for behavior intervention on mobile devices. Procedia Comput. Sci. 63, 236–243
(2015)
12. Fogg, B.J.: The behavior grid: 35 ways behavior can change. In: Persuasive 2009. Claremont,
California (2009a)
13. Fogg, B.J.: A behavioral model for persuasive design. In: Persuasive 2009. Claremont,
California (2009b)
14. Greene, J., Grant, A.M.: Solution-Focused Coaching: Managing People in a Complex World.
Momentum Press, London (2003)
15. Horsch, C.H.G., Lancee, J., Beun, R.J., Neerincx, M.A., Brinkman, W.P.: Adherence to
technology-mediated insomnia treatment: a meta-analysis, interviews with users, and focus
groups with users and experts. J. Med. Internet Res. 17(9), e214 (2015)
16. Horsch, C.H.G., Brinkman, W.P., van Eijk, R.M., Neerincx, M.A.: Towards the usage of
persuasive strategies in a virtual sleep coach. In: UKHCI 2012 Workshop on People,
Computers and Psychotherapy (2012)
17. Hudson, F.M.: The Handbook of Coaching. Jossey-Bass, San Francisco (1999)
18. Kyle, S.D., Miller, C.B., Rogers, Z., Sirwardena, A.: Sleep restriction therapy for insomnia
is associated with reduced objective total sleep time, increased daytime somnolence, and
objectively-impaired vigilance: implications for the clinical management of insomnia
disorder. Sleep 37(2), 229–237 (2014)
19. Lancee, J., van den Bout, J., Sorbi, M.J., van Straten, A.: Motivational support provided via
email improves the effectiveness of internet-delivered self-help treatment for insomnia: a
randomized trial. Behav. Res. Ther. 51, 797–805 (2013)
20. Michie, S., van Stralen, M.M., West, R.: The behavior change wheel: a new method for
characterizing and designing behavior change interventions. Implementation Sci. 6(42)
(2011)
21. Middelweerd, A., Mollee, J.S., van der Wal, N., Brug, J., te Velde, S.J.: Apps to promote
physical activity among adults: a review and content analysis. Int. J. Behav. Nutr. Phys. Act.
11(1), 97 (2014)
22. Morin, C.M., Bootzin, R.R., Buysse, D.J., Edinger, J.D., Espie, C.A., Lichstein, K.L.:
Psychological and behavioral treatment of insomnia: update of the recent evidence (1998–
2004). Sleep 29, 1398–1414 (2006)
23. Morin, C.M., Espie, C.A.: Insomnia: A Clinical Guide to Assessment and Treatment.
Springer, New York (2003)
24. Van Straten, A., Cuijpers, P., Smit, F., Spermon, M., Verbeek, I.: Self-help treatment for
insomnia through television and book: a randomized trial. Patient Educ. Couns. 74(1), 29–
34 (2009)
25. Vermeire, E., Hearnshaw, H., Van Royen, P., Denekens, J.: Patient adherence to treatment:
three decades of research a comprehensive review. J. Clin. Pharm. Ther. 26, 331–342 (2001)
26. Weizenbaum, J.: ELIZA – a computer program for the study of natural language
communication between man and machine. Commun. ACM 9(1), 36–45 (1966)
Online Peer Groups as a Persuasive Tool
to Combat Digital Addiction
Amen Alrobai ✉ , John McAlaney, Keith Phalp, and Raian Ali

( )
Faculty of Science and Technology, Bournemouth University, Bournemouth, UK

{aalrobai,jmcalaney,kphalp,rali}@bournemouth.ac.uk
Abstract. Digital Addiction (DA) denotes a problematic usage of digital devices

characterised by properties such as being compulsive, impulsive, excessive and
hasty. DA is associated with negative behaviours such as anxiety and depression.
“Digital Detox” programs have started to appear and are mainly based on a rela‐
tively expensive and heavyweight in-patient care utilising traditional solutions
such as motivational interviews and cognitive behavioural therapies. For
moderate addiction, persuasive technology could have potential, as a brief inter‐
vention, to assist users to regulate their usage. This paper explores the design of
online peer groups as a persuasive technique that puts together people who share
a common interest in combating their DA or in helping others to do so. We
conducted empirical research to explore design aspects of this mechanism. The
results raise a range of questions and challenges to address when developing such
a technique for the behaviour change needed against DA.
Keywords: Digital addiction · Digital health persuasion · Online peer groups
1 Introduction
Despite obvious benefits, the emergence of digital technologies and social networking
services has also led to negative consequences on modern societies. A recent meta-
analysis study covered 80 empirical reports from 1996 to 2012 and concluded that
Internet Addiction, which is a form of digital addiction, affects 6 % of people worldwide
[1]. In 2013, 3.2 % of British students were considered to be addicted [2]. In South Korea,
over 140 Internet Addiction treatment recovery centres opened by 2011 [3]. According
to the recent statistics from the China Youth Association for Network Development, the
number of Chinese teenage addicts increased to 24 million (14.1 %) by 2009 [4]. This
is over double the number in 2005, which was around 10 million (13.2 %). It is worth
noting that these studies followed different criteria on assessing addictive usage.
These trends have led to growing interest in research to combat DA using self-regu‐
lation systems. These systems are seen as supportive means and emphasize that addicts
have an active role in changing their own behaviour. These systems could include for
example interactive warning labels containing persuasive techniques like timers and
avatars to combat DA, as proposed in [5]. In another study [6], a new approach to ICT-
facilitated self-regulation was proposed based on social cognitive theory to limit smart‐
phone usage. The approach facilitates creating groups of users to share their usage

DOI: 10.1007/978-3-319-31510-2_25
Online Peer Groups as a Persuasive Tool 289
information. Generally, these systems are based on the assumption that people have the
individual ability to adjust and optimise their own behaviour to maximise their gains
according to their particular circumstances.
Despite such efforts, the introduction of these software systems to health-related
behaviour, including addictive behaviours such as DA, has led to many controversial
arguments. Most notable is the lack of strong scientific proof for their potential effec‐
tiveness. In a recent study [7], researchers found that many app-based psychological
interventions including those hosted by governmental bodies, such as the National
Health Service in the UK, fail to demonstrate clinical evidence of a long term change.
A longitudinal research study found that delivering interventions within peer group
settings could be harmful due to different factors relevant to group structure, which led
to reinforcing the negative behaviour [8] such as loafing and compensation [9], and
conformity effect [10]. Some negative attributes of persuasive technologies were also
reported in [11], such as frustration, anxiety, peer pressure and feeling of guilt for the
participants. This suggests a need for further research on the design of such software-
based solutions in order to exploit their power whilst attempting to avoid or reduce
negative side effects.
In this paper, we introduce online peer groups as a persuasive mechanism based on
self-regulation systems, to support an effective and long-term behavioural change to
combat DA. This paper conducts an exploratory research on the different aspects that
need to be considered when designing online peer groups and reflects on the applicability
and potential as well as risks of such a mechanism.
2 Background and Research Motivation
A motivational peer group is where people “voluntarily come together to help each other
address common problems or shared concerns” [12]. Linking addicts to peer support
groups prior to the professional treatment may reduce the duration needed in the initial
episodes of treatment and increase recovery rates [13]. Also, extending the participation
in peer support groups reduced the need for subsequent treatment episodes [13]. The
peer groups approach can also be utilised in the post-treatment to reduce relapse rates
[14]. The strength of this approach lies in its distinct persuasive and motivational mech‐
anisms to sustain behavioural change; mainly commitment and consistency, reciprocity,
and social proof [15] as well as surveillance which reinforce all these mechanisms. Peer
groups technique could utilize the helper-therapy principle [16] which suggests that it
can be personally beneficial for addicts to assist others deal with own addictions.
Behavioural change theories such as the theory of planned behaviour [17] and
goal-setting theory [18] are used to bridge the gap between attitudes and behav‐
iours. These theories aim at reducing discrepancies between these two conceptual
constructs such as, for example, the gap between the intention to change a behav‐
iour and the act of actually doing so [19]. This is achieved by encouraging individ‐
uals to create a plan to achieve the targeted behaviour. In self-regulation systems,
monitoring is a fundamental design element. It provides a useful basis for effective
intervention design by enabling users to track their performance and support them
290 A. Alrobai et al.
in achieving their goals, whilst also maintaining their regulated behaviours [20].
Self-monitoring is a “process of having individuals record data regarding their own
behaviour for the purpose of changing its rate” [21]. When doing so, correctly in
social settings, such as peer groups, it can further support the positive change.
Goal setting is a key element to guide monitoring processes and make it more mean‐
ingful. In DA, it is still a research question as to which type of goals, i.e. proximal and
distal, would be more effective to develop self-efficacy and be easier for users to setup.
Proximal goals are essential sub-goals to achieve the distal goals. For example,
increasing offline social connection is a proximal goal to regulate digital usage and
combat DA, the associated distal goal. In DA, we still have no models on the decision
rights of goal settings for a particular group and whether it should be set up by the
individuals or an authorized moderator or perhaps collectively. Likewise, the decision
on setting up the goals is still a research issue, e.g., whether it should be self-set, provider-
set, participatory-set, or set up by recommender systems [22].
Self-regulation systems can either monitor behaviour, e.g., the user shared 40 posts
this week on their social network, or monitor change in the behaviour, e.g., the user
shared fewer posts than last week [23]. It is fundamental to investigate which type of
monitoring would motivate users. Some studies, such as [24], concluded that interven‐
tion systems for addictive behaviours may fail due to poor application of goal-setting
theory, e.g., difficulty in setting standards as well as poor consideration of conflicting
goals, such as regulating mobile usage and enjoying the moment, and also distorted
goals, such as surfing the Internet to improve mood. These risks pose challenges for
group’s governance. While there are some successful intervention cases, they are often
short term, as such interventions are expensive and hard to maintain [25]. This suggests
the need for complementary strategies to support long-term interventions and to reduce
relapse rate. This could be achieved through in-patient care, which is expensive and
heavy weight for the early stages of DA.
Hence, we suggest investigating a persuasive technique that combines technology
with human support to achieve sustainable behaviour change in a flexible and efficient
style, hence the suggestion of online peer groups. Peer groups approach can be an
appropriate program for users in the transition to addiction stage due to their need for
less action-oriented strategies in which immediate change is not expected [26]. It can
also benefit those who are also unaware of their level of addiction as it can make them
more informed of the consequences occurred to their peers. This paper will explore the
potential of online peer groups as a persuasive technique in that regard and focus on
different design aspects.
3 Method and Research Settings
This paper reports upon work to explore users’ perceptions of online peer groups, with
respect to their possible use to help digital addicts. We adopted several qualitative
methods in two studies to triangulate the findings and to generate more comprehensive
understanding.
For the first study, we had a relatively broad remit, to investigate how users would
perceive self-monitoring and peer monitoring to combat DA. We first conducted a diary
study with 14 participants, 5 male and 9 female, aged between 18 and 50. A convenience-
sampling technique was used. A pre-selection questionnaire test was used to ensure that
participants had at least one aspect of problematic usage of their smart phones. The pre-
selection test was an adapted version of the CAGE questionnaire customized to fit the
properties and remit of DA [27].
The participants were asked to install one of three commercial digital diet smart‐
phones applications and use it for 14 days, and to record their observations and feelings
about the application and their usage style. These diaries were used to guide the follow-
up interviews with the same participants. To be selected, the application needed to have
a rating of at least three stars out of five, a high number of downloads (no less then 500);
at least 5 persuasive techniques, e.g. goal-settings, monitoring reminders, and rewarding,
coercion and surveillance. This was to allow users to engage with a wide range of
persuasive features in order to assess their influence and suitability for DA. After 14
days of usage and writing reflections and feedback, we conducted semi-structured inter‐
views with the same participants to elaborate on their diaries. One of the interviews
questions was about the social features of the applications, which represent various
aspects of online peer groups.
The second study focused on understanding different perspectives on online peer
groups and their interactive design. We conducted a two sessions focus group study.
The first one included 6 participants, 3 males and 3 females. The participants had pre-
existing social relationships with each other so they aligned with the concept of a peer
group, which requires some degree of shared interest and trust. Ages ranged between
20 and 26. The CAGE-like questionnaire was again used as a pre-selection test. Partic‐
ipants were given an engaging task in which they had to comment on and construct
different online peer group designs and interaction styles including the use of persuasive
techniques to regulate addiction.
We used all obtained data in the first session to come up with a new peer group design
and made it the subject of discussion in the next session in which the same participants
would engage as potential users of such online peer group aiming to regulate their
addiction. However, one of the participants, who could not participate, was replaced
with another one who met the selection criteria and had good experience in 3D anima‐
tion, which was advantageous to give some ideas on a creative design.
Finally, we performed a survey study to get further confirmation and insights through
comments from a wider sample on our findings. A total of 73 completed responses were
returned from the sample of 42 male and 31 female, aged between 18 and 65, recruited
through an open call via several academic mailing lists.
In order to scope our analysis and as an initial template, we used the Cialdini’s six
principles of influence [15] to investigate the potential influential aspects in social
settings from users’ perspective as well as Fogg’s behavioural model [28] to focus on
the technology-facilitated features that can maximise the persuasion of online peer
groups.
4 Results
Different aspects and areas of concerns about peer groups have been explored such as
the study in [8]. This includes the personal characteristics that can heavily influence the
effectiveness of ex-addicts and non-professional participation in counselling activities
[29]. In online peer groups, such concerns need to be revisited by exploring what digital
addicts prefer in terms designing online peer groups. Figure 1 presents the main aspects
of online peer groups for DA. This conceptual map reflects the areas of concerns that
are considered important from users’ perspective. Governance, Structuring and Moder‐
ation will be discussed in separate subsections while Risks will be discussed as a cross-
cutting aspect through all subsections.
Fig. 1. Online peer groups for digital addiction
In the next subsections, we present general findings including the usefulness of peer
groups to regulate DA, users’ motivations to join them and how group structuring is
seen and preferred by the users. We will also discuss various concerns and design issues
related to the role of groups’ moderators and the application of behaviour change theories
and persuasive techniques within social settings including social norms.
4.1 Attitudes to the Overall Concept of on-Line Peer Groups
In our study, we assessed the perception of digital addicts of the usefulness of peer groups
to regulate their usage. The overall impression was positive, and 71 % felt that a peer
group would be useful (26 % certainly, 45 % somehow). Participants liked the idea of
technology-limiting technology and found “online system to cure online addiction is an
interesting concept”. However, it had not escaped the authors, nor had it escaped some
of the participants, that there is a paradox in supporting online addiction by inviting
people to partake of online support, as part of another community. Perhaps one of the
comments that highlight this best was one person who likened the approach “inviting
alcoholics down the pub to chat about their alcoholism”. Of course, there are key
differences here. Firstly, the medium, the online, is for many, something that they cannot
choose to avoid. As such, being connected is a fundamental part of their life, or indeed,
for many, their professional life. Secondly, digital addiction is a software-mediated
behaviour. Hence, with the aid of software means, it is possible to actively monitor and
intervene when necessary.
To identify the primary appealing characteristics of online peer groups, participants
were asked about their motivation to join peer groups. Interestingly, most of the
comments were clearly stating that providing moral support to others would be the main
motivation. This motivation can be triggered by reciprocity norm [15] which suggests
that the equality of power and exchanging help are the essential ingredients to build
effective groups. This is well understood in the “helper therapy” principle [16]. That is,
participants find it more motivating to be also useful for others, not only addicts who
are seeking for help. This is perhaps the key element of peer groups.
Some other techniques used in persuasive systems seem not to be an important aspect
for online peer groups used to assist the behavioural change for combating DA. For
example, the Cialdini’s Liking principle suggests that people are easily influenced by
those who they admire, such as celebrities. That is to say popular persuasive techniques
may not be seen efficient and even accepted when applied for behavioural change in
addictive behaviours, such as DA, and they need to be revisited for that context of use.
4.2 The Moderation Role in Online Peer Groups

Participants highlighted the important role of having the human element such in software
systems advocating techniques like peer groups to build a sense of trust and commitment
to support long-term change. For example, one participant commented that in young age
groups, having parental involvement would benefit more as they have a sense of
authority to regulate the usage. For all of the individuals within the group, the role of
the moderator was clearly understood as of paramount importance.
For example, there was a legitimate concern that there is a high risk factor of peers
developing deviant behaviours due to normative influence [30]. This is one of the main
reasons to introduce the moderation as an essential process in online peer groups.
Moderator can also play other typical governance roles such as those related to member‐
ships and rewards allocation as well as addressing the influence of the non-matched
members such as grouping members who belong to different levels of change.
Characteristics of Moderators. Participants were asked to consider a range of poten‐

tial characteristics for the crucial role of moderator. Of these, only one question gained
greater than 50 % agreement, 58 % believing the moderator must be “accredited/profes‐
sional/should be professional for [advanced digital] addiction”. The comment above
appears to endorse the need for a ‘professional’ moderator, though an exception was
highlighted when groups or moderators have “successful support history” regardless of
their professional knowledge. Thus, the experiential knowledge was perceived as an
appealing attribute of the moderation role. In addition, two further clear themes are clear
from the comments.
The first observation is that the participants made a distinction between ‘light’ and
‘advanced stage’ digital addiction. That is, they suggest that for early stage addiction,
friends or less qualified people might be helpful in a peer group, and this further suggests
that such a peer group approach as an early, low cost, intervention is something that they
consider to be useful even among genuine peers. Two comments exemplify this
commonly stated view: (1) “for groups with advanced addiction, moderator should be
a therapist or a digital addict or an ex digital addict but with therapist expertise so they
know what to say and how to say it”, and (2) “For light addiction anyone really, does
not matter, I would say the same for early addiction that means to prevent it first or to
recover from it. I see difference”.
The second observation is that the attitude or approach of the moderator is seen as
paramount, friendliness and liberal styles being mentioned, one argued that “friendliness
in the group is a main requirement”. However, again these comments being qualified
by those cases where ‘professional treatment’, requires a therapist.
Finally, the question of whether moderators should themselves be ex-addicts drew
far more mixed response. While some, 20 % of the survey, considered that ex-addicts
would have more empathy, a greater number suggested that the moderator should not
be an ex-addict, “addicts might dictate their opinion and be biased to their own
experience”.
The Role of Moderators. Participants wanted to have collaborative moderators, who

have the ability to guide the behavioural change by providing inspirational motivation.
For example, a participant commented that a moderator should be “someone who is
respectable and can take charge; but also sympathises”. The most positive responses
for activities of the moderator, were: create and suggest rules (of engagement) 58 %,
support motivation (55 %), provide advice to members (54 %), and create real life events
(54 %) – but of course this final suggestion contrasts with anonymity, and reward
members for complaint behaviour/usage (though note this is taken on trust) 50 %.
Hence, the moderator’s role was seen, as in other forms of addiction, as primarily
about setting out and controlling how people interacted, suggesting rules, motivating
members, and giving advice. A perhaps surprising finding is that many wanted moder‐
ators to create real life events, contrasting with other questions for anonymity. Finally,
while the moderator giving some kind of rewards was favoured by 50 % of the survey
respondents, penalties was, a much lower score, with only 30 % believing that the
moderator should give penalties. An interesting comment from the survey, which again
tallied with our other studies was: “No penalty but probably confrontation with their
status”, since one of the perceived benefits is concrete evidence and heightened aware‐
ness of actual usage. These characteristics of the role of moderation align well with the
transformational leadership paradigm [31].
The use of “rational or economic means” to strengthen the probability of members’
compliance with group’s goals, suggests a moderation role that follows transactional lead‐
ership paradigm [31]. In this type of moderation, two approaches can be taken. The first is
about active moderation and requires monitoring groups’ interaction to ensure continued
enhancement of the performance through applying corrective actions. Second is a passive
moderation in which a moderator intervenes and applies operant conditioning when
group’s goals and standards are violated [31]. Other users preferred to have the moderator
as a councillor, so members request their interventions when needed. As such, no moni‐
toring and direct intervening are required.
Using persuasive techniques for behavioural change might lead some design issues
related to the moderator role in peer groups. Participants argued that in some groups a
moderator should be enabled to guide the change through positive reinforcements and
light penalties as persuasive techniques. For example, one participant commented:
“people may leave a peer group if too much penalty is enforced”. Another one high‐
lighted that if penalties must be implemented should take more influential approach such
as “confrontation members with their status”.
The careful implementation of persuasion techniques and moderation role will have
profound a impact on groups’ self-esteem. One of main components in Fogg’s model
for persuasive design [28] is the ability to perform the targeted behaviour or to reduce
negative behaviour. When reducing the negative behaviour is very challenging goal such
as in severe addiction, we would expect the design of online peer groups to increase the
motivational influence and to apply the right triggers. This is to increase the probability
of behavioural change to occur. For example, providing means to express the confidence
in member’s ability to change or applying the right social norm would increase the
perceived self-efficiency which will act as a powerful motivational tool [32]. Peer group
design should provide moderators with means to enact such policies and enable an
effective persuasion to change behaviour.
4.3 Structuring of Online Peer Groups

Professional involvement in severe addiction cases suggests that the stage of addiction,
i.e. early, intermediate and advanced, has an influence on how to customise online peer
groups. Thus, we would expect the design on online peer groups to sense users’ addiction
status and adapt different facets of this mechanism accordingly to provide more persua‐
sive effects.
Ultimately, each stage of addiction represents different level of self-control and
distinct attitudes and behaviours. Regardless of the extent to which the object of addic‐
tion dominates decision-making processes, individuals with less severe addictive behav‐
iour can be guided through the stage of change [26]. The stage of change, correspond
to the stages of the Transtheoretical model (TTM) to behavioural change [26].
A critical assumption is that persuasive software-mediated interactions are more
suitable for those who are open to the change, honest and do not have denial of reality.
On the other hand, users who exhibit severe addiction symptoms require different course
of action and more comprehensive treatment regardless of the stage of change they are
at [26]. However, peer groups can still play a role in different phases of that compre‐
hensive treatment, e.g. pre-treatment phase to support problem recognition “non-addicts
have no idea but they may give a perspective and may learn how it feels” and post-
treatment phase to support relapse avoidance as highlighted in [13]. Structuring peer
groups should also consider the theme of addiction. For example addiction to online
pornography would require certain degree of anonymity. This aspect will be discussed
in the next section.
The social structure within online peer groups seems a very important aspect to be
considered in the design. For example, a participant commented “friends are not always
the good thing here but unknown people with no direct contact or a friend of friend might
be better and more relaxed” and he continued, “family members would be distracting
in the group as I may need to behave differently”. Only, 9 % of the responses were in
favour of having family members. However, another one commented: “family members
are fine to have in the online peer groups but not as moderators”. This suggests that the
design needs to consider the impact of including family and friends versus unknown
individuals in the group. This could be linked to the severity and domain of DA. Thus,
the design of online peer groups must also consider the domain of addictions, such as
gambling and pornography, which would require higher level of anonymity.
The social norms approach has become a major focus of research in recent years
and is widely adopted in different developmental sectors such as educational settings
in the United States [33]. The approach has been successfully used for behavioural
change in the domain of addiction as well as a number of health and socially rele‐
vant behaviours [33].
As has been demonstrated extensively throughout social psychological research,
individuals are strongly motivated to alter their own thoughts and behaviours to match
the norms of the group [34]. This can include descriptive norms, which refers to how
often or extensively we perceive our peers to engage in a behaviour or injunctive
norms, which refers to the attitudes we believe our peers to hold. In a case of reciprocal
causality individuals will also seek out social groups whose behaviours and attitudes
they perceive to reflect their own [35]. Explicit attempts to manipulate groups, particu‐
larly by out-group members, can lead what known as a reactance response in which
individuals engage even more strongly in the original behaviour [36]. However people
also tend to underestimate how easily influenced they are by the groups they belong to
[37]. As such by challenging the perceived norms within a group or encouraging the
group to aspire towards a healthier norm behaviour change may be achieved.
Research into the use of peer networks to bring about behaviour change would
suggest that they can indeed create new and more positive social norms [38]. There is
overall though a lack of research on how social norms may operate within online peer
groups.
The degree to which social norms may operate differently in online groups could be
expected to reflect the complexity of the social relationships between the members of
the peer groups. In the case of peer groups where members feel emotionally distant from
one another they are less likely to conform to the norm [39]. However conformity to the
social norm is more likely to occur in groups where there is a shared sense of a common
goal and a belief that each member plays an important role in the achievement of this
goal [10]. If peer groups are therefore to be used to address DA it is important that this
is done in a way that engages the group, creates an agreed norm to aspire too and involve
all the members of this group into the process. A participant commented, “if a group of
people I knew were all trying to cut down their phone usage then I think it would
[motivate me to cut down my usage”.
4.4 Governance and Social Aspects in Online Peer Groups

Enabling computer-mediated interactions among peers raises several social-related
concerns. Participants highlighted the level of anonymity as key motivation to join peer
groups. Several participants commented that such platforms should be a “safe space” in
which users can maintain certain level of anonymity not necessarily complete, as self-
disclosure is a key aspect in such social software platforms. For example, anonymity
might need to be maintained at the level of members’ interactions only, i.e. a member
cannot be identified by other members but still identifiable by the system to monitor his
progress over time. As such, online peer groups should accommodate various degrees
of anonymity [40]. We expect the design of online peer groups to consider the addiction
theme as an important aspect in deciding the suitable levels of privacy. More work is
still needed to look into how addicts perceive anonymity and the influential privacy
aspects that plays a role in persuading them to join and sustain their participation.
The experiential knowledge attribute mentioned in Sect. 4.2 suggests visibility
concept in groups functioning as an important persuasive feature to enhance the intra-
group’s trust. Participants used the visibility term to denote the notion of having acces‐
sible service history and overall performance of groups and moderators. In this sense,
anonymity and visibility are not conflicting requirements as the latter revolves around
participation visibility rather than participant’s visibility.
Other participants showed interest in sharing the role of moderator to maximise
group’s outcomes. The observation from the lack of having a particular trend in electing
moderators is perhaps due to people in peer groups wanting a participatory style and to
hear from peers rather than authority figures, such as parents, which take more central‐
ised style.
As with the whole concept of using online to regulate DA drew some mixed views,
and many noted the apparent paradox of having gamification and online approaches due
to the risk from over-engagement. However, on the whole participants were positive
about bringing some ‘fun’ to the peer group, and transfer of activities is often something
useful within traditional addictions. On the whole, competition inside the group was
seen as potentially problematic whereas they wanted to support rather than to compete
other members. However, from all stages of the research there was a mention of the
possibility of the group having an overall usage, from all members, and that they might
wish to see this as a collective goal, or even compete with other groups. Clearly, this is
an often used gamification tactic, call centre teams compete with each other, and often
weight loss has been tackled with such an approach. Having the element of competition
suggests that the group functioning can take task-oriented approach to either meet indi‐
viduals’ or collective goals. Other users wanted more flexible and supportive medium
that is free from competition to “support rather than beat each other”.
In terms of specific tactics, 53 % wanted points for compliance with ‘healthy’
activity, whereas only 40 % were happy with the idea of something like a leader board.
Of course one person noted the monitoring issue, stating “points are good but how will
you monitor off-line behaviour?”
Finally, and perhaps to be expected for peer support, the most desirable feature,
(64 %), was online chat, further reinforcing the impact of people in support.
5 Conclusions and Future Work
In this paper we explored different aspects of online peer groups, as a motivational

mechanism, from users perspectives. We demonstrated its prominent persuasive consid‐
erations. Although peer groups technique aims at supporting individuals in all addiction
levels, we argue the need for careful re-evaluation of the online version of it from the
perspectives of behavioural change theories. We also argue that tailoring such social
software platform to support those with advanced stage addiction would be a very chal‐
lenging task. Users who exhibit severe addiction are more vulnerable to relatively
unconscious distorted, conflicting, changing requirements and could be accompanied
with denial of reality. Our understanding of how social norms operate within peer groups
is based primarily on offline interactions. Online environments may have unique char‐
acteristics that need to be better understood if change is to be achieved.
The current state of classical methods in software development, e.g. requirements
engineering, is not efficient enough to deal with users in that stage. Thus, to design
persuasive systems for DA, future studies are required to re-visit software engineering
methods to customise existing elicitation models to the domain of behavioural addiction.
Ultimately, adapting online peer groups to different users’ needs and expectations should
eventually produce more persuasive effects. Using simple metrics such as time and
frequency to measure the level of addiction would perhaps provide misleading asser‐
tions. As such, we argue that measurement models should consider the psychological
research on the addiction severity based on clinical criteria, such as salience, conflict
and relapse [41]. Thus, users feedback should feed into future measurement models to
provide meaningful and suitable configurations for online peer groups. The design of
such user feedback acquisition and its peculiarities in this domain, e.g. to detect and
react against denial of reality, is a challenge to address.
Acknowledgment. The research was supported by Bournemouth University through the PGR
Development fund. We would like also to thank Asad Khan and Yasmeen Abdalla for their
valuable contributions in conducting the focus group and diary studies in the early stages of this
research.
References
1. Cheng, C., Li, A.Y.-L.: Internet addiction prevalence and quality of (real) life: a meta-analysis
of 31 nations across seven world regions. Cyberpsychology Behav. Soc. Netw. 17, 755–760
(2014)
2. Kuss, D.J., Griffiths, M.D., Binder, J.F.: Internet addiction in students: prevalence and risk
factors. Comput. Hum. Behav. 29, 959–966 (2013)
3. Young, K.S., de Abreu, C.N. (eds.): Internet Addiction: a Handbook and Guide to Evaluation
and Treatment. John Wiley & Sons (2010)
4. China Youth Association for Network Development. 2009 statistical Report on Internet?
Addictions of Chinese Adolescents. Beijing (In press, 2010)
5. Ali, R., Jiang, N., Phalp, K., Muir, S., McAlaney, J.: The emerging requirement for digital
addiction labels. In: Fricker, S.A., Schneider, K. (eds.) REFSQ 2015. LNCS, vol. 9013, pp.
6. Ko, M., Yang, S., Lee, J., Heizmann, C., Jeong, J., Lee, U., Shin, D., Yatani, K., Song, J.,
Chung, K.-M.: NUGU: a group-based intervention app for improving self-regulation of
limiting smartphone use. Presented at the February (2015)
7. Leigh, S., Flatt, S.: App-based psychological interventions: friend or foe? Evid. Based Mental
Health 18, 97–99 (2015)
8. Dishion, T.J., McCord, J., Poulin, F.: When interventions harm: peer groups and problem
behavior. Am. Psychol. 54, 755–764 (1999)
9. Karau, S.J., Williams, K.D.: Social loafing: a meta-analytic review and theoretical integration.
J. Pers. Soc. Psychol. 65, 681–706 (1993)
10. Allen, V.: Situational factors in conformity. Adv. Exp. Soc. Psychol. 2, 133–170 (1965)
11. Hamari, J., Koivisto, J., Pakkanen, T.: Do persuasive technologies persuade? - a review of
empirical studies. In: Spagnolli, A., Chittaro, L., Gamberini, L. (eds.) PERSUASIVE 2014.
12. Davidson, L., Chinman, M., Kloos, B., Weingarten, R., Stayner, D., Tebes, J.K.: Peer Support
among individuals with severe mental illness: a review of the evidence. Clin. Psychol.: Sci.
Pract. 6, 165–187 (2006)
13. Moos, R.H., Moos, B.S.: Help-seeking careers. J. Subst. Abuse Treat. 26, 167–173 (2004)
14. Moos, R.H., Moos, B.S.: Paths of entry into alcoholics anonymous: consequences for
participation and remission. Alcohol. Clin. Exp. Res. 29, 1858–1868 (2005)
15. Cialdini, R.B.: Influence. HarperCollins, New York (2009)
16. Riessman, F.: The “helper” therapy principle. Soc. Work 10, 27–32 (1965)
17. Ajzen, I.: The theory of planned behavior. Organ. Behav. Hum. Decis. Process. 50, 179–211
(1991)
18. Locke, E.A., Latham, G.P.: A Theory of Goal Setting and Task Performance. Prentice-Hall,
Englewood Cliffs (1990)
19. Webb, T.L., Sniehotta, F.F., Michie, S.: Using theories of behaviour change to inform
interventions for addictive behaviours. Addiction 105, 1879–1892 (2010)
20. Torning, K., Oinas-Kukkonen, H.: Persuasive system design: state of the art and future
directions. In: PERSUASIVE, p. 30 (2009)
21. Coleman, M.C., Webber, J.: Emotional and Behavioral Disorders: Theory and Practice. Allyn
& Bacon, Boston (2002)
22. Strecher, V.J., Seijts, G.H., Kok, G.J., Latham, G.P., Glasgow, R., DeVellis, B., Meertens,
R.M., Bulger, D.W.: Goal setting as a strategy for health behavior change. Health Educ.
Behav. 22, 190–200 (1995)
23. Maitland, J., Chalmers, M.: Self-monitoring, self-awareness, and self-determination in
cardiac rehabilitation. In: CHI, pp. 1213–1222 (2010)
24. Vohs, K.D., Baumeister, R.F.: Handbook of Self-Regulation, 2nd edn. Guilford Press,
New York (2013)
25. Green-Demers, I., Pelletier, L.G., Ménard, S.: The impact of behavioural difficulty on the
saliency of the association between self-determined motivation and environmental
behaviours. Can. J. Behav. Sci./Revue Canadienne des sciences du comportement. 29, 157–
166 (1997)
26. Prochaska, D.J.O.: Transtheoretical model of behavior change. In: Gellman, M.D., Turner,
J.R. (eds.) Encyclopedia of Behavioral Medicine, pp. 1997–2000. Springer, New York (2013)
27. Ewing, J.A.: Detecting alcoholism: the CAGE questionnaire. JAMA 252, 1905–1907 (1984)
28. Fogg, B.J.: A behavior model for persuasive design. In: PERSUASIVE, p. 40 (2009)
29. Snowden, L., Cotler, S.: The effectiveness of paraprofessional ex-addict counselors in a
methadone treatment program. Psychotherapy: Theor. Res. Pract. 11, 331–338 (1974)
30. Allen, J.P., Chango, J., Szwedo, D., Schad, M., Marston, E.: Predictors of susceptibility to
peer influence regarding substance use in adolescence. Child Dev. 83, 337–350 (2011)
31. Bono, J.E., Judge, T.A.: Personality and transformational and transactional leadership: a
meta-analysis. J. Appl. Psychol. 89, 901–910 (2004)
32. Shamir, B., House, R.J., Arthur, M.B.: The motivational effects of charismatic leadership: a
self-concept based theory. Organ. Sci. 4, 577–594 (1993)
33. McAlaney, J., Bewick, B., Hughes, C.: The international development of the “social norms”
approach to drug education and prevention. Drugs. Educ. 18, 81–89 (2011)
34. Kelman, H.C.: Interests, relationships, identities: three central issues for individuals and
groups in negotiating their social environment. Annu. Rev. Psychol. 57, 1–26 (2006)
35. Ennett, S.T., Bauman, K.E.: The contribution of influence and selection to adolescent peer
group homogeneity: the case of adolescent cigarette smoking. J. Pers. Soc. Psychol. 67, 653–
663 (1994)
36. Fuegen, K., Brehm, J.W.: The intensity of affect and resistance to social influence. In:
Resistance and Persuasion, pp. 39–64 (2004)
37. Darley, J.M.: Social organization for the production of evil. Psychol. Inq. 3, 199–218 (1992)
38. Wright, K.B., Bell, S.B., Wright, K.B., Bell, S.B.: Health-related support groups on the
internet: linking empirical findings to social support and computer-mediated communication
theory. J. Health Psychol. 8, 39–54 (2003)
39. Greene, J.D., Sommerville, R.B., Nystrom, L.E., Darley, J.M.: An fMRI investigation of
emotional engagement in moral judgment. Science 293, 2105–2108 (2001)
40. Kobsa, A., Schreck, J.: Privacy through pseudonymity in user-adaptive systems. TOIT 3,
149–183 (2003)
41. Griffiths, M.: A “components” model of addiction within a biopsychosocial framework. J.
Subst. Use 10, 191–197 (2005)
Design Strategies and Techniques
Red Radiators Versus Red Tulips: The Influence
of Context on the Interpretation
and Effectiveness of Color-Based Ambient
Persuasive Technology
Shengnan Lu(B) , Jaap Ham, and Cees Midden
Human Technology Interaction, Eindhoven University of Technology,

P.O.Box 513, 5600 MB Eindhoven, The Netherlands
{s.lu,j.r.c.ham,c.j.h.midden}@tue.nl
Abstract. Colors are widely used as feedback in ambient persuasive

technology. In current research, we argue that the information that color-
based feedback carries is highly context dependent. Two studies investi-
gated effects of context (in which color-based feedback was presented) on
user’s interpretation of feedback messages, and more importantly, on the
effectiveness of this feedback (for influencing energy conservation behav-
ior). Results of both studies showed that participants perceived the color
red in an energy-related context to be warmer and as related to a higher
energy consumption level than red in an energy-unrelated context. Also,
participants receiving color-based feedback in an energy-related context
consumed the lowest amount of energy. These findings extend our insight
into the psychological mechanisms of ambient persuasive technology by
making clear that ambient stimuli (e.g., color) are part of a broader con-
text (e.g., projected on a radiator) that influences user’s interpretation
and the effectiveness of these stimuli.
Keywords: Ambient persuasive technology · Context · Color · Sustain-

ability
1 Introduction
In day-to-day life, people constantly interact with and reflect on their surround-
ings [1]. Indeed, the physical environment has a strong influence on how people
experience the world, and also on how people function in many ways [2]. In other
words, people’s subjective experiences are dependent on their bodily sensations
and their perceptions of their physical environments [3].
The other way around, psychological factors including subjective experi-
ences can also influence people’s assessment of their physical environments
[4,5]. For example, earlier research showed that social exclusion experiences can
make people literally feel colder [6]. Likewise, other research showed that peo-
ple’s beliefs regarding the current temperature settings and their own abilities

DOI: 10.1007/978-3-319-31510-2 26
304 S. Lu et al.
to adjust to warm or cold conditions can influence their thermal experiences [7,8].
Also, research showed that a reported (relatively) high temperature can increase
perceived comfort even when the actual room temperature remained the same
[9]. Moreover, recent research showed that a sustainable mindset (e.g., people
being aware of acting environmentally friendly) can induce people to perceive
the room temperature to be physically higher [10].
In this interaction between physical and psychological factors, colors pre-
sented in the physical environment play an important role [11,12]. Various earlier
theories and research studying the influence of color on people presented evidence
for a link between colors and temperature-related experiences. For example, Ben-
nett and Rey [13] argued that colors with ‘cold’ hues (e.g., blue, green and vio-
let), lead to the perception of temperatures being lower, while colors with ‘warm’
hues (e.g., red, yellow and orange), lead to the perception of temperatures being
higher (known as “Hue-Heat” hypothesis). Confirmed by Franger and colleagues
[14], a red room led to the perception of higher temperature, whereas a blue room
created the perception of a colder temperature (even though room temperature
was objectively the same). Also, earlier research showed that yellow can induce
a higher perceived room temperature than blue [15].
However, in some other studies that also used colors to test the “Hue-Heat”
hypothesis, different results have been reported refuting or not confirming this
hypothesis [16]. For instance, Ho and colleagues [17] found effects apparently
in contrast with the red-hot/blue-cold association, which indicated that a blue
object is more likely to be judged as warm than a red object of the same physical
temperature. Also, several studies were unable to demonstrate that colors had
any effect on the perception of temperature [16,18]. Separate from the “Hue-
Heat” hypothesis literature, other research also showed results that conflict. That
is, some studies presented evidence in line with this hypothesis by indicating that
warm colors can evoke more arousal than cool colors [19]. Still other research
presented evidence opposing this hypothesis, that is, brain research showed that
blue light (in comparison to red light) leads to more arousal [20]. In addition,
several researchers pointed out that red enhances cognitive performance [21,22]
while, in contrast, others found that red reduces cognitive performance [21,23],
or no relation between color and performances [12,24,25]. Finally, some earlier
research showed that colors can influence human emotional state [26], while
others studies did not find such evidence [27].
These conflicting results might be due to various causes. In the current
research, we explore as a reason Elliot and Moller’s Color-in-Context Theory
[28]. This theory proposed that the effects of colors are highly context depen-
dent. That is, it argues that color is an integrated part of the environment and
cannot be separated from its form, texture and other surroundings. Therefore,
influences of colors can only be observed in its context. For instance, the color
red can carry a negative meaning (failure, danger, threat) for individuals in
achievement contexts [28]. That is, research [21] showed that in a context in
which people have to avoid making mistakes, priming them with the color red
made participants more afraid to make mistakes. In other contexts, the color
Red Radiators Versus Red Tulips: The Influence of Context 305
red can carry a positive meaning and have affiliation-related implications, for
example in contexts involving heterosexual interaction [28].
2 Current Research
The current research was designed with the aim to investigate the influence
of a color’s context (i.e., the context in which colors were presented as energy
feedback) on the interpretation of the information that color-based persuasive
technology carries, and furthermore on the effectiveness of this color-based per-
suasive technology to influence users’ energy consumption behavior.
In the research area of persuasive technology [29], colors have been widely
used as media to support behavioral changes (e.g., [30,31]). Midden and
colleagues [31] indicated that persuasive technologies, that utilize forms of sen-
sory information like changing colors, can inform users at an intuitive level,
demanding little cognitive effort. The colors, such as red and green, have been
employed to present energy feedback to promote energy conservation behavior
[32–34]. For example, earlier research provided individual comparison feedback
by changing the shirt color of onscreen avatars that represented participants.
The avatar with a red shirt represented the person who used the largest amount
of energy, while the avatar with the green shirt represented the person who used
the smallest amount of energy. Results suggested that using these colors as feed-
back in this context was very effective for motivating participants to save energy
[35]. Likewise, Jentsch and colleagues [33] used colored ambient display to convey
information about energy consumption in office environments. That is, green was
used to convey that an office environment did not have energy-saving potential,
and red meant that an office environment did have energy-saving potential. More
recent research showed comparable results, now using colored lighting to pro-
vide feedback [36,37]. More specifically, Lu and colleagues [37] used red and green
lighting to provide ambient lighting feedback about energy consumption. Their
results indicated that colors that have pre-existing associations with energy con-
sumption (e.g., red/green that is associated with high/low energy consumption)
can enhance the persuasive power of ambient feedback.
Importantly, in the current research we argue that the meanings that colors
carry may vary dependent on the context in which a color is perceived. So,
although in certain context colors such as red are interpreted to indicate high
energy consumption and thereby help users to understand the feedback messages,
in different context the color red can be interpreted as indicating good and alive,
as for example, in a nature context (e.g., when detecting ripe fruit) [28].
Based on a pretest, we specified two contexts: one is an energy-related context
in which colors were projected on the form of a heating radiator, and the other
is an energy-unrelated context in which colors were projected on the form of
a tulip. We based these context choice on a pretest, in which participants had
indicated that an heating radiator is an object is strongly associated with energy
consumption, whereas a tulip is an object that is weakly associated with energy
consumption.
306 S. Lu et al.
Specifically, in the first study, we argue that presenting colors (e.g., red) in an
energy-related context (e.g., projecting red on an object – the form of a radiator
– to indicate high energy consumption) will make feedback messages (e.g., high
energy consumption level) more clear and can enhance the correct interpretation
of the color (e.g., warm), compared to presenting colors in an energy-unrelated
context (e.g., projecting red on an object – the form of a tulip – to indicate
high energy consumption). In the second study, we aim to replicate the effects
of a color’s context on the interpretation of the information that colors carries
(as in the first study), and more importantly, we argue that presenting colors in
energy-related contexts can increase the effectiveness of the color-based ambi-
ent persuasive technology, leading to stronger energy conservation behavioral
changes
3 Study 1
In Study 1, we focused on the influence of context (in which colors were per-
ceived) on the messages and users’ interpretations of the color when using it as a
feedback mechanism in colored-based ambient persuasive technology. Two spe-
cific colors (i.e., red and blue) were presented in either an energy-related context
or an energy-unrelated context respectively. We expected (H1.1) that partici-
pants would interpret the color red in an energy-related context as indicating
the highest energy consumption level, compared to the color red in an energy-
unrelated context and the color blue in an energy-related or energy-unrelated
context. Also, we expected (H1.2) that participants would perceive the color red
in an energy-related context to be the warmest, compared to the participants in
other conditions.
3.1 Method
Participants and Design. Fifty-one students (average age 21.6 years old,
SD = 2.15) of Eindhoven University of Technology (23 male and 28 female)
were recruited by using a local participant database. We invited only partici-
pants who were not color blind. All participants were native Dutch speaker and
were randomly assigned to one of the four experimental conditions: 2 (context:
energy-related vs. energy-unrelated)×2 (color: red vs. blue). The experiment
lasted approximately 15 min, and participants received 3 euros for their
participation.
Experimental Procedure. Before participants entering the lab, each par-

ticipant was asked to read and sign an informed consent form stating the
general purpose of the research and their willingness to participant in this
research. Meanwhile, the room was changed to fit one of the four experimen-
tal conditions, that is, we changed only the picture presented on the TV screen
present in the room to show the related lighting stimulus (i.e., a red radiator,
a red tulip, a blue radiator or a blue tulip; see Fig. 1).
(a) red radiator (b) red tulip (c) blue radiator (d) blue tulip
Fig. 1. Four experimental conditions: (a) the color red in an energy-related context,
(b) the color red in an energy-unrelated context, (c) the color blue in an energy-related
context and (d) the color blue in an energy-related context (Color figure online).
First of all, participants were asked to look around the room and then remem-
ber (for later recall) as many objects in this room as possible (with the purpose of
making them spend attention to the ambient environment). Then, participants
were asked to fill out a questionnaire. This questionnaire contained a variety of
questions (in a random order) about the ambient environment and the object on
the TV screen, amongst them the two questions of focus. That is, this question-
naire asked participants for their interpretations of the color (i.e., either red or
blue) on either the radiator or tulip (using the question “How do you judge the
meaning that this object conveys? I think this object means...”, to which par-
ticipants could answer by choosing an option on a 7-point bipolar scale ranging
from 1 = very low energy consumption to 7 = very high energy consumption).
Also, the questionnaire asked participants for the warmth perception of the color
on either radiator or tulip (using the question “How warm do you feel the object
projected on the screen is? I feel the object is something...”, to which ranging
from 1 = very cold to 7 = very warm). Finally participants were debriefed and
thanked for their participation.
3.2 Results and Discussion
To analyze the effects of the context in which colors were presented on par-
ticipant’s interpretation of the color meaning and warmth of the object, we
used an analysis of variance (ANOVA), in which 2 (context: energy-related vs.
energy-unrelated)×2 (color: red vs. blue) experimental conditions were manipu-
lated between participants.
Confirming our first hypothesis (H1.1), planned contrasts revealed that par-
ticipants interpreted the color red in an energy-related context as indicating
the highest energy consumption level, compared to participants in other condi-
tions, t(47) = 4.2, p < .001. Furthermore, an additional analysis using a two-
way ANOVA confirmed that participants reported the color red as related to
a higher energy consumption level (M = 4.98, SE = .30) than the color blue
(M = 3.59, SE = .31), F = 10.5, p = .002, η 2 = .18. Also, participants inter-
preted the color in an energy-related context (i.e., radiator) as indicating a higher
energy consumption (M = 4.78, SE = .29) than in an energy-unrelated context
(i.e., tulip) (M = 3.79, SE = .31), F = 5.31, p = .026, η 2 = .10.
308 S. Lu et al.
Also, confirming our second hypothesis (H1.2), planned contrasts supported

that participants perceived the color red in an energy-related context as the
warmest, compared to participants in other conditions, t(47) = 7.10, p < .001.
Likewise, two-way ANOVA confimed that participants perceived the color red
to be warmer (M = 5.64, SE = .24) than the color blue (M = 3.00, SE = .24),
F (1, 47) = 59.5, p < .001, η 2 = .56. In addition, participants rated the color in
an energy-related context (i.e., radiator) as warmer (M = 4.72, SE = .24) than
in an energy-unrelated context (i.e., tulip) (M = 3.92, SE = .25), F (1, 47) =
5.46, p = .024, η 2 = .10.
In sum, results of Study 1 showed that presenting colors (e.g., red) in an
energy-related context indeed made the feedback message (e.g., high energy con-
sumption) more clear, and moreover, enhanced the correct interpretation (e.g.,
warmer) represented by the color.
Still, although Study 1 showed that participants can easily understand the
information that color-based persuasive technology conveys in a fitting context,
the important question remains of whether this fitting context (in which the
color-based persuasive technology was presented) can influence users’ energy
conservation behavior. Therefore, the second study was designed with the fol-
lowing two purposes: (1) to replicate the effect of context on users’ interpretation
of the feedback messages that color-based persuasive technology carries, and (2)
to further investigate the influence of the context on the effectiveness of color-
based ambient feedback for promoting users’ energy conservation behavior.
4 Study 2
In the second study, first of all, to replicate effects of the context on users’ inter-
pretation of feedback messages found in Study 1, we argue that a fitting context
would strengthen the associations of color with high energy consumption and
the concept of warmth, which makes the feedback messages directly understand-
able. Therefore, we expected (H2.1) that participants would interpret the color
red in an energy-related context to indicate a higher energy consumption level
than the color red in an energy-unrelated context. Also, we expected (H2.2) that
participants would perceive the color red in an energy-related context as being
warmer than the color red in an energy-unrelated context.
Secondly, and more importantly, to investigate the context effect on behav-
ioral change, we argued that a fitting context can increase the effectiveness of this
color-based energy feedback to facilitate energy conservation behavior. There-
fore, we developed an interactive energy feedback system, in which the object
(i.e., either a radiator or tulip) could show colors varying from very weak red
to very intense red depending on people’s energy use. And we expected (H2.3)
that participants would save the most energy when they receive the color-based
ambient feedback in an energy-related context.
4.1 Method
Participant and Design. Fifty students (average age 23.1 years old, SD =
3.14) of Eindhoven University of Technology (32 male and 18 female) were
recruited by using a local participant database. We invited only participants
who were not color blind. They participated in the current experiment in either
an energy-related context (i.e., projecting red as energy feedback on the radia-
tor) or an energy-unrelated related context (i.e., projecting red on the tulip). All
participants were native Dutch speakers. The experiment lasted approximately
30 min, and participants received 5 euros for their participation.
Experimental Procedure. Participants were welcomed at the entrance of

the lab. Each participant was asked to read and sign an informed consent form
stating the general purpose of the research and their willingness to participate
in this research.
After entering the lab, participant was seated in front of a laptop on a desk.
All the instructions and tasks we used in this experiment were fully computerized
and were all in Dutch. First of all, participants were asked to read a magazine
for a period of five minutes, while either a red radiator (in the energy-related
context condition) or a red tulip (in the energy-unrelated context condition) was
presented on the TV (see Fig. 2). After five minutes, the computer indicated that
the reading period was over and participants were asked to fill out a questionnaire
(similar one as used in Study 1) about the object on the TV screen and the
ambient environment.
(a) feedback system via radiator (b) feedback system via tulip
Fig. 2. The interactive ambient energy feedback system by presenting colors in (a) an
energy-related context and in (b) an energy-unrelated context.
After answering these questions, we assessed energy consumption behavior

by asking participants to perform a virtual thermostat task on the laptop while
immediately receiving the interactive color-based energy feedback (see Fig. 3).
That is, participants received instructions on how to program a virtual room
310 S. Lu et al.
thermostat interface, and were asked to program this thermostat for two prac-
tice scenarios. After two practice scenarios, participants completed programming
tasks on the thermostat for ten experimental scenarios. For each scenario, the
scenario itself was first described to the participants (e.g., It is night and you are
going to bed. It is −10◦ C outside), and then participants were asked to program
the thermostat by setting the temperature for the each room, while receiving
feedback (see Fig. 3) after each change of the temperature settings, until they
pressed the ‘Ready’ button. Finally, participants were asked to answer several
demographic questions, thanked for participation, and debriefed.
(a) (b) (c) (d)
Fig. 3. The interactive color-based energy feedback system showed colors varying
between very weak and intense red depending on people’s energy use from (a) indicat-
ing very low energy consumption to (d) very high energy consumption (Color figure
online).
4.2 Result and Discussion

Replicating the effects of context on user’s interpretation of the feedback mes-
sages (found in Study 1), results confirmed the our first two hypotheses (H2.1 &
H2.2). That is, participants interpret the color red in an energy-related context
as indicating a higher energy consumption level (M = 6.12, SE = .24) than the
color red in an energy-unrelated context (M = 4.68, SE = .24), F (1, 48) = 18.3,
p < .001, η 2 = .28. Also participants perceived the color red in an energy-related
context to be warmer (M = 6.44, SE = .17) than the color red in an energy-
unrelated context (M = 5.64, SE = .17), F (1, 48) = 10.7, p = .002, η 2 = .18.
Most importantly, to analyze the influence of the context (in which color-
based ambient feedback was presented) on the effectiveness of this color-based
ambient persuasive technology (for influencing user’s energy conservation behav-
ior), we collected the energy consumption score of each participant (based on
his or her performance during the thermostat task) as dependent variable. These
dynamic energy consumption scores were based on their actions during the ther-
mostat task and were submitted to a linear mixed effects analysis to assess the
relationship between the manipulation factor context (radiator vs. tulip) and a
participant’s scores of energy consumption. So, as fixed effects, we entered con-
text into the model. As random effects, we entered scenario items. Confirming
our third hypothesis (H2.3), results showed that participants who received color-
based energy feedback in an energy-related context (i.e., radiator) decreased
their energy consumption more (M = −13.6, SE = 2.25) than participants who
received color-based energy feedback in an energy-unrelated context (i.e., tulip)
(M = −10.1, SE = 2.3), F (1, 1715.9) = 4.84, p = .028.
So, results of Study 2 showed that participants could clearly interpret the
information of color-based energy feedback carries (e.g., red indicated high
energy consumption and was more associated with warm) in an energy-related
context. Furthermore, color-based energy feedback in an energy-related context
led to the most energy conservation behavior. Thereby, these findings revealed
that color context can help users to correctly interpret the information that
color-based feedback carries, and moreover, can increase the effectiveness of this
ambient persuasive technology to facilitate energy consumption behavior.
5 General Discussion
Colors have been widely used as media to promote behavioral changes. In the
current research, we argue that the information that color-based ambient persua-
sive technology carries is highly context dependent. Two studies were designed to
investigate the effects of a color’s context (i.e., the context in which colors were
presented as energy feedback) on users’ interpretation of the information that
colors carried. Also, we investigated the effect of color context on the effectiveness
of this color-based ambient persuasive technology to influence users’ energy con-
sumption behavior. Results of both studies showed that participants to whom we
presented colors in an energy-related context (e.g., projecting red on an object
in the form of a radiator in indicate high energy consumption) could most opti-
mally interpret the information of a color carried (e.g., red indicated high energy
consumption and was more associated with warm). More importantly, results
showed that participants to whom we presented colors in an energy-related con-
text saved the most energy when they were receiving color-based energy feedback
as compared to participants in an energy-unrelated context.
In line with earlier research and the Color-In-Context theory [28,38], our
findings confirmed that the context can shape the associations of color. In addi-
tion, using colors as ambient feedback to promote energy conservation behavior,
we found that a fitting context (in which this color-based ambient persuasive
technology is presented) can strengthen the associations of the color red with
high energy consumption (or concept of warm), and subsequently can enhance
the energy-saving performance. Importantly, the current research implies that
by taking into consideration the context of ambient persuasive technology, the
effectiveness of ambient persuasive technology (e.g., colored lighting ambient
feedback used in [37]) could be increased.
The current research opens up a new field of research. That is, based on
the first findings of the current studies, future research might investigate the
influence colors, in context, on user comfort perceptions and thereby on user’s
(energy consumption) behavior. Indeed, comfort experience was stated as one
of the dominant influences on people’s (energy conservation) behavior [39,40].
Besides using colors (in different context) as feedback to indicate energy infor-
mation, future research can also investigate whether user’s comfort experiences
will be influenced by using color-based persuasive technology, via a comfort-
experience-based persuasive strategy [41].
312 S. Lu et al.
Another important consideration related to studying conform experience is

that in our second study, participants performed the temperature-setting task on
a simulated thermostat without the physical room temperature being adapted.
However, in a day-to-day environment, changes on the settings of a heating
thermostat directly lead to changes in room temperature. Future research could
investigate the relationship between ambient persuasive technology, comfort
experiences and energy consumption behavior, in an environment in which these
variables are connected to one another. For example, room temperature changes
after people change settings of a thermostat. In such research, a simulated ther-
mostat could be connected to the heating system in the lab. In addition, by
applying this thermostat system and color-based energy feedback in real setting,
future research can also compare the effect size between lab studies and field
studies.
Additionally, future research may extend the current findings by not only
presenting color-based ambient energy feedback on a TV screen. That is, future
research might replicate and also extend these findings by investigating the influ-
encing of colors and the role of their context when colors are presented in a more
ecological context, for example when presenting an actual red glow on an actual
physical radiator.
In sum, our studies showed that a fitting context in which ambient persuasive
technology is presented can shape the associations of color with high energy con-
sumption and the concept of warm, which makes the feedback messages directly
understandable. And more importantly, this fitting context can increase the per-
suasive power of ambient persuasive technology to promote energy conservation
behavior. These findings extend our insight into the psychological mechanisms
of ambient persuasive technology by making clear that contextual stimuli (e.g.,
colored lighting used as ambient feedback) are part of a broader context (e.g.,
projected on a radiator vs. a tulip) that influences user’s interpretation and the
effectiveness of these stimuli. This could also have implications for the design
of future ambient persuasive technology as the current findings show that such
design should take into consideration the context in which ambient persuasive
technology is used.
References
1. Ford, F.L.: Political Murder: From Tyrannicide to Terrorism. Harvard University
Press, Cambridge (1987)
2. Schneider, W.: Sinn und Un-Sinn: Umwelt Sinnlich Erlebbar Gestalten in Architek-
tur und Design. Bau-Verlag, Berlin (1957)
3. Barsalou, L.W.: Perceptual symbol systems. Behav. Brain Sci. 22(04), 577–660
(1999)
4. Helson, H., Blake, R.R., Mouton, J.S., Olmstead, J.A.: Attitudes as adjustments
to stimulus, background, and residual factors. J. Abnorm. Soc. Psychol. 52(3), 314
(1956)
5. Sherif, M.: An Outline of Psychology. Sage Publications, New York (1948)
6. Zhong, C.B., Leonardelli, G.J.: Cold and lonely does social exclusion literally feel
cold? Psychol. Sci. 19(9), 838–842 (2008)
7. Howell, W.C., Kennedy, P.A.: Field validation of the fanger thermal comfort model.
Hum. Factors: J. Hum. Factors Ergon. Soc. 21(2), 229–239 (1979)
8. Howell, W.C., Stramler, C.S.: Contribution of psychological variables to the predic-
tion of thermal comfort judgments in real world settings. ASHRAE Trans. 87(5),
609–619 (1981). (United States)
9. Stramler, C.S., Kleiss, J.A., Howell, W.C.: Thermal sensation shifts induced by
physical and psychological means. J. Appl. Psychol. 68(1), 187 (1983)
10. Strack, F., Deutsch, R.: Acting green elicits a literal warm glow. Nat. Clim. Change
5, 37–40 (2014)
11. Tofle, R.B., Schwarz, B., Yoon, S.Y., Max-Royale, A., Des, M., Thanks, S.: Color
in healthcare environments-a research report (2004)
12. Elliot, A.J., Maier, M.A., Moller, A.C., Friedman, R., Meinhardt, J.: Color and
psychological functioning: The effect of red on performance attainment. J. Exp.
Psychol.: Gen. 136(1), 154–168 (2007)
13. Bennett, C.A., Rey, P.: What’s so hot about red? Hum. Factors: J. Hum. Factors
Ergon. Soc. 14(2), 149–154 (1972)
14. Fanger, P., Breum, N., Jerking, E.: Can colour and noise influence man’s thermal
comfort? Ergon. 20(1), 11–18 (1977)
15. Winzen, J., Albers, F., Marggraf-Micheel, C.: The influence of coloured light in the
aircraft cabin on passenger thermal comfort. Lighting Research and Technology
1477153513484028 (2013)
16. Candas, V., Dufour, A.: Thermal comfort: multisensory interactions? J. Physiol.
Anthropol. Appl. Hum. Sci. 24(1), 33–36 (2005)
17. Ho, H.N., Iwai, D., Yoshikawa, Y., Watanabe, J., Nishida, S.: Combining colour
and temperature: A blue object is more likely to be judged as warm than a red
object. Sci. Rep., vol. 4 (2014)
18. Berry, P.C.: Effect of colored illumination upon perceived temperature. J. Appl.
Psychol. 45(4), 248 (1961)
19. Jacobs, K.W., Suess, J.F.: Effects of four psychological primary colors on anxiety
state. Percept. Motor Skills 41(1), 207–210 (1975)
20. Yoto, A., Katsuura, T., Iwanaga, K., Shimomura, Y.: Effects of object color stimuli
on human brain activities in perception and attention referred to eeg alpha band
response. J. Physiol. Anthropol. 26(3), 373–379 (2007)
21. Mehta, R., Zhu, R.: Blue or Red? Exploring the effect of color on cognitive task
performances. Sci. 323, 1226–1229 (2008)
22. Kwallek, N., Lewis, C.M., Robbins, A.S.: Effects of office interior color on workers’
mood and productivity. Percept. Motor Skills 66(1), 123–128 (1988)
23. Soldat, A.S., Sinclair, R.C., Mark, M.M.: Color as an environmental processing
cue: External affective cues can directly affect processing strategy without affecting
mood. Soc. Cogn. 15(1), 55–71 (1997)
24. Etnier, J.L., Hardy, C.J.: The effects of environmental color. J. Sport Behav. 20(3),
299 (1997)
25. Ainsworth, R.A., Simpson, L., Cassell, D.: Effects of three colors in an office interior
on mood and performance. Percept. Motor Skills 76(1), 235–241 (1993)
26. Kaya, N., Epps, H.: Relationship between color and emotion: A study of college
students. Coll. Student J. 38(3), 396 (2004)
27. O’Connor, Z.: Colour psychology and colour therapy: caveat emptor. Color Res.
Appl. 36(3), 229–234 (2011)
28. Elliot, A.J., Maier, M.A.: Color-in-context theory. Adv. Exp. Soc. Psychol. 45,
61–125 (2012)
314 S. Lu et al.
29. Fogg, B.: Persuasive Technology: Using Computers to Change What We Think
and Do. Morgan Kaufmann, San Francisco (2003)
30. Arroyo, E., Bonanni, L., Selker, T.: Waterbot : Exploring Feedback and Persuasive
Techniques at the Sink. In: Proceedings of the SIGCHI Conference on Human
Factors in Computing Systems, pp. 631–639 (2005)
31. Midden, C., Kaiser, F.G., Mccalley, L.T.: Technology’s four roles in understanding
individuals’ conservation of natural resources. J. Soc. Issues 63(1), 155–174 (2007)
32. Maan, S., Merkus, B., Ham, J., Midden, C.: Making it not too obvious: The effect
of ambient light feedback on space heating energy consumption. Energy Effi. 4(2),
175–183 (2011)
33. Jentsch, M., Jahn, M., Pramudianto, F., Simon, J., Al-Akkad, A.: An energy-saving
support system for office environments. In: Salah, A.A., Lepri, B. (eds.) HBU 2011.
34. Merkus, B., Ham, J., Midden, C.: A Shower Meter to Save Water and Energy :
the Influence of Two Feedback Sources, Goals, and the Role of Comfort on Con-
servation Behavior in a Lab and Field Experiment in partial fulfilment of the
requirements for the degree of Master of Science in Human. Ph.D. thesis (2012)
35. Midden, C., Kimura, H., Ham, J., Nakajima, T., Kleppe, M.: Persuasive power
in groups: the influence of group feedback and individual comparison feedback on
energy consumption behavior. In: Proceedings of the 6th International Conference
on Persuasive Technology: Persuasive Technology and Design: Enhancing Sustain-
ability and Health, p. 1. ACM (2011)
36. Ham, J., Midden, C.: A persuasive robotic agent to save energy: the influence
of social feedback, feedback valence and task similarity on energy conservation
behavior. In: Ge, S.S., Li, H., Cabibihan, J.-J., Tan, Y.K. (eds.) ICSR 2010. LNCS,
37. Lu, S., Ham, J., Midden, C.J.H.: Using ambient lighting in persuasive communi-
cation: the role of pre-existing color associations. In: Spagnolli, A., Chittaro, L.,
Gamberini, L. (eds.) PERSUASIVE 2014. LNCS, vol. 8462, pp. 167–178. Springer,
Heidelberg (2014)
38. Busch, M., Schrammel, J., Flu, M.A., Kruijff, E., Tscheligi, M.: Persuasive strate-
gies report. CURE-Center for Usability Research & Engineering (2012)
39. Becker, L.J., Seligman, C., Fazio, R.H., Darley, J.M.: Relating attitudes to resi-
dential energy use. Environ. Behav. 13(5), 590–609 (1981)
40. Heijs, W., Stringer, P.: Research on residential thermal comfort: Some contribu-
tions from environmental psychology. J. Environ. Psychol. 8, 235–247 (1988)
41. Lu, S., Ham, J., Midden, C.: Persuasive technology based on bodily comfort expe-
riences: the effect of color temperature of room lighting on user motivation to
change room temperature. In: MacTavish, T., Basapur, S. (eds.) PERSUASIVE
Investigating Politeness Strategies and Their
Persuasiveness for a Robotic Elderly Assistant
Stephan Hammer1(B) , Birgit Lugrin2 , Sergey Bogomolov1 , Kathrin Janowski1 ,

and Elisabeth André1
1
Human Centered Multimedia, Augsburg University, Augsburg, Germany
{hammer,janowski,andre}@hcm-lab.de
2
Human Computer Interaction, University of Wuerzburg, Wuerzburg, Germany
[email protected]
Abstract. This work is targeted towards the development of a Robotic

Elderly Assistant (REA) system that provides assistance in the form of
recommendations to support single-living elderly people in their domes-
tic environment. To avoid potential face threats the REA should be as
polite as possible whilst keeping a certain persuasiveness to promote its
recommendations. This paper investigates different verbalizations of the
REA’s recommendations regarding their perceived politeness as well as
their persuasiveness. We present the results of a laboratory study with
younger adults and a user study with the inhabitants of a retirement
home. Results suggest that the different politeness strategies reflected
different levels of politeness in both studies, while their perceived per-
suasiveness needs further investigation in the domain of elderly care.
Keywords: Elderly people · Social robots · Recommendations · Polite-

ness strategies · Persuasiveness
1 Motivation
The demographic change, especially in industrialized countries, results in impor-
tant societal challenges to be dealt with. Since elderly people often suffer from
physical and mental restrictions, they lose living independence. Especially single-
living elderly often lack societal interactions and thus get isolated. Both problems
get amplified if the elderly are inactive and lack self-initiative.
Studies have shown that regular physical and cognitive activities can help mit-
igate many age-related diseases [1,2]. For example, pain due to arthrosis could be
alleviated by moving the affected joints on a regular basis, and activities, such
as gardening or painting, can positively influence elderly people’s well-being [3].
However, some seniors tend to not engage in these activities. Therefore, Seiderer
et al. [4], for example, tried to promote a healthier lifestyle and to increase seniors’
overall well-being by providing concrete recommendations. However, developing
such systems for seniors should be done with care, as people of their generation
may be less familiar with newer electronic devices and thus have a rather high
barrier of use or, even worse, are afraid of using them [5,6].

DOI: 10.1007/978-3-319-31510-2 27
316 S. Hammer et al.
Through their embodiment, among other features, social robots are very
well suited to support socio-emotional factors. There is evidence that humans
perceive them as social actors and are likely to respond to them in a similar
way that people respond to each other [7]. Furthermore, through their ability to
adapt to a large variety of users and remain attractive to users on a long-term
basis, social agents are typically useful for intuitive and persuasive interaction
[8]. Therefore, it seems to be promising to develop a Robotic Elderly Assistant
(REA) that could contribute to the establishment of a positive emotional and
social relationship between user and system.
A number of studies have been conducted that explore social robots, such
as the iCat or Nao, in terms of their acceptance by older adults, e.g. [9,10]. In
general, seniors seem to evaluate social robots mainly in terms of appearance,
intellectuality, and friendliness [11]. Robots should appear less threatening, but
kind-hearted [12,13] and more realistic faces are perceived as more trustworthy
and sociable [11]. Elderly people prefer discrete and small robots with human or
pet-like traits over large humanoid robots [12,14]. Furthermore, robots are more
likely to be accepted if they move slowly, act less autonomously, have a female
voice and a monochrome and serious appearance [12]. There is also evidence
that elderly people respond positively to robotic companions if they emulate
social behavior that matches the seriousness of a current task or situation [15].
However, if robots look too human-like, but do not match the high expectations
in terms of behavior, people tend to get disappointed and distrustful of them [16].
Special attention should be paid to the fact that many elderly find themselves
in a situation where they can no longer take full care of themselves and need
the help of others to accomplish daily tasks. Consequently, the feeling of embar-
rassment or loss of control over their lives become important issues. Therefore,
providing recommendations often comes along with a certain threat of the users’
face. For example, the statement “Drink some water.” could invoke the feeling of
being patronized. In contrast, the question “How about drinking some water?”
would keep the users in control whilst reminding them to drink enough water.
Therefore, we think that verbal politeness strategies, as an aspect of social and
respectful behavior, can further enhance the acceptance of a REA and can foster
a good working relationship between the elderly user and the REA.
In the following we will introduce related work that investigated the role of
politeness in verbal social interactions. Afterwards, our approach to a REA will
be presented. The main part of the paper will explain details on two studies that
were performed to investigate the perception of different politeness strategies and
their persuasiveness. The second study was conducted in a local retirement home
with our REA system and also investigated the elderly participants’ reaction to
the robot. The paper concludes with a discussion and some lessons learned from
the observations made during our studies.
2 Politeness as a Social Factor

Sidner and Lee [17] designated politeness as an important factor when robots
initiate engagement with people, and Nomura et al. [18] showed that even small
Investigating Politeness Strategies and their Persuasiveness for a REA 317
differences in a robot’s non-verbal behavior, such as motions, could influence peo-

ple’s perception of it as well as their behavior towards it. In the domain of elderly
care, politeness seems to be of special importance, as the perceived politeness of
a system varies with the user’s experience with technical systems [19].
We think that the way recommendations are verbalized by the robotic com-
panion can help mitigating facial threats, e.g. by using suggestions, hints or
proposals instead of commands. Brown and Levinson [20] describe politeness
as a means to preserve the reputation of conversation partners even in critical
situations. Their politeness theory builds on the fact that every interlocutor has
two basic wants concerning their face or public self-image: (1) to be approved
of by the conversation partner (positive face) and (2) to be unrestricted by the
conversation partner (negative face). In order to avoid threads to their (posi-
tive or negative) face during a conversation, interlocutors tend to apply different
types of politeness strategies: (1) to emphasize approval (positive politeness),
(2) to emphasize the interlocutor’s freedom of choice (negative politeness) and
(3) indirect statements in case an action is necessary (off-record statements).
Johnson et al. [21] incorporated these politeness strategies in different ver-
balizations of an artificial tutor that used politeness for motivating a learner
to accomplish different tasks. Eight categories of verbalizations were presented
that relate to Brown and Levinson’s notation of positive and negative face. These
categories are: (A) direct commands, (B) indirect suggestions, (C) requests, (D)
actions expressed as the tutor’s goals, (E) actions as shared goals, (F) questions,
(G) suggestions of student goals, and (H) Socratic hints.
Results of Johnson’s experiment showed that participants rated positive and
negative politeness with a high degree of consistency and that their ratings were
consistent with Brown and Levinson’s assessment of the strategies’ politeness.
In the context of an elderly care assistant, it should be noted that depending
on the importance of the given advice, more or less polite wordings might be
required. Thus, the present contribution investigates all eight categories and
rates them with regard to their perceived politeness as well as their perceived
persuasiveness.
3 The REA System
Our overall goal is to develop a sensitive and personalized system to support

elderly people who would generally be able to live independently, but lack a cer-
tain autonomy due to forgetting appointments, daily tasks or not showing the
initiative for activities. Assistance is given in the form of situationally appro-
priate recommendations to encourage physical, mental, and social activities, all
aimed at increasing the users’ well-being. To ensure an interaction that is as
natural and intuitive as possible, and to reduce barriers to entry, the system
is impersonated by an expressive robot that acts like a social companion. The
robot chosen for the REA system was a Robopec Reeti 1 , see Fig. 1. It meets the
1
http://www.reeti.fr.
most important criteria that were mentioned in Sect. 1. It is comparatively small

(44 cm), plain white and, apart from the head, it does not have any movement
capabilities that could be perceived as threatening. Its basic appearance is that
of a cartoon-like extraterrestrial or fantasy creature. The facial expressions are
human-like, which allows the user to relate to the robot more easily, yet styl-
ized enough to reduce expectations about realistic behavior. For German speech
synthesis, Reeti uses the Loquendo text-to-speech software by Nuance2 .
Fig. 1. Reeti, a social robot with an expressive head created by Robopec. It supports
gaze behavior with three movement axes in the neck and two in each eye, as well as
eight degrees of freedom for animating the mouth, eyelids and ears.
Based on Brown and Levinson’s politeness strategies and Johnson et al.’s

taxonomy of verbalizations (see Sect. 2), we prepared three recommendations
Table 1. Different wordings of the recommendation ‘Drink some water’
Politeness strategy Verbalization Translation

Direct command Trinken Sie etwas Wasser Drink some water
Indirect suggestion Ihr Ernährungsplan sieht vor, Your dietary plan suggests that
dass Sie etwas Wasser trinken you drink some water
Request Ich hätte gern, dass Sie etwas I would like you to drink some
Wasser trinken water
Actions expressed as Ich würde etwas Wasser trinken I would drink some water
the system’s goals
Actions expressed as Wir sollten etwas Wasser trinken We should drink some water
shared goals
Questions Wie wäre es, wenn Sie etwas How about drinking some
Wasser trinken würden? water?
Suggestions of user Sie möchten bestimmt etwas You would probably like to
goals Wasser trinken drink some water
Socratic hints Haben Sie daran gedacht etwas Did you think about drinking
Wasser zu trinken? some water?
2
http://www.nuance.com.
that would naturally fit in a scenario of a REA: 1) drink some water, 2) open

the window, and 3) go for a walk. Examples of the different wordings can be
seen in Table 1. The German version as used in our system is included, since an
equally perceived degree of politeness of the translations cannot be guaranteed.
For presenting the recommendations, the VisualSceneMaker tool [22,23] was
used to implement a semi-automatic dialog application. Via keyboard inputs the
experimenter was able to move to the next phrase, to repeat the current phrase,
or to stop the whole process.
4 Laboratory Study
To ensure that our experiment is not too subtle or too abstract to grasp for the
inhabitants of an elderly home, we conducted a text-based study at our lab first
to verify that the different verbalizations are perceived as expressing different
levels of politeness and persuasiveness.
4.1 Participants and Procedure
The study was conducted with 5 female and 15 male native German speakers
aged from 25 to 45, in a lab environment. After a short introduction, participants
answered a few demographic questions. Then they were shown eight textual ver-
balizations for each of the implemented recommendations (1: drink some water,
2: open the window, and 3: go for a walk) in an incomplete counterbalanced order.
For each verbalization, participants had to rate their perceived politeness as well
as persuasiveness on a 7-point Likert-scale from 1 = “not polite/persuasive at
all” to 7 = “very polite/persuasive”.
4.2 Results
An analysis of the mean ratings, see Fig. 2, showed that questions were rated as
most polite. Actions that were expressed as shared goals or the system’s goals,
and requests were perceived as polite, too. In contrast, direct commands were
assessed as impolite. Interestingly, direct commands were perceived to have a
similar degree of persuasion as questions. Socratic hints and suggestions of user
goals were rated as least convincing. Strategies that were perceived as polite as
well as persuasive were actions expressed as shared goals and requests.
A repeated-measures ANOVA analysis of the provided ratings for all
recommendations showed significant differences for the perceived politeness
(F (7, 308) = 37.82, p < .001) as well as the perceived persuasiveness
(F (5.25, 231.05) = 8.14, p < .001). A subsequent Bonferroni post-hoc test, see
Table 2, confirmed the impressions of the descriptive analysis. Amongst others,
questions were perceived as significantly more polite than any other strategy,
and direct commands scored significantly worse on politeness than the other
strategies.
Fig. 2. Results of the laboratory study (Mean ratings for all recommendations). 1:
direct command, 2: indirect suggestions, 3: requests, 4: actions as system goals, 5:
actions as shared goals, 6: questions, 7: suggestions of user goals, 8: Socratic hints
Thus, the study showed that the investigated politeness strategies were per-
ceived as significantly differently polite as well as persuasive. However, strate-
gies that were perceived as polite were not necessarily perceived as persuasive or
unpersuasive. Therefore, a good strategy for a REA might be to (1) choose ver-
balizations that are both, polite and persuasive, (2) neglect verbalizations that
are neither polite nor persuasive, and (3) adapt verbalizations to the current
situation to keep a certain variability in behavior, e.g., by choosing the verbal-
ization depending on the importance of the given recommendation. For example,
in cases where it is necessary to follow the advice of the robotic companion (such
as remembering to take the prescribed medicine), a more convincing strategy
Table 2. Results of the laboratory study: M = Mean; SD = Standard Deviation;

Sig = significantly better than
Politeness Persuasiveness
M SD Sig M SD Sig
1. Direct command 2.27 1.03 4.11 1.76
2. Indirect suggestion 3.47 1.41 1∗∗ 4.29 1.63 7∗∗ ; 8∗
3. Request 4.42 1.60 1∗∗∗ 4.40 1.48 7∗∗∗ ;8∗∗
4. Actions expressed as system goals 4.84 1.11 1,2,8∗∗∗ 3.89 1.45 7∗
5. Actions expressed as shared goals 4.98 1.22 1,2,8∗∗∗ 4.64 1.45 7∗∗∗ ; 8∗∗
6. Questions 5.78 1.11 5∗ ; 1,2,3,4,7,8∗∗∗ 3.76 1.32 7∗
7. Suggestions of user goals 4.09 1.41 1∗∗∗ 2.96 1.38
8. Socratic hint 3.58 1.20 1∗∗∗ 3.16 1.51
∗ significant with p <.05; ∗∗ significant with p <.01; ∗∗∗ significant with p <.001
could be chosen over a more polite but less persuasive version. In other cases,
such as a suggestion to drink water regularly, the recommendation should focus
more on politeness than on persuasiveness.
5 Study with Target User Group

After the politeness strategies were evaluated by younger adults in text form in
a laboratory environment, a second study was conducted with the target user
group, elderly inhabitants of a local retirement home who are needing assis-
tance in certain daily situations, but do not suffer from serious mental diseases
such as dementia. This time, the recommendations were presented by the REA
described in Sect. 3. The aim was to investigate whether the elderly would dis-
tinguish between the perceived politeness and persuasiveness of the different
verbalizations. Furthermore, we wanted to observe the seniors’ acceptance of
the presented REA system.
5.1 Participants and Procedure
In cooperation with the retirement home’s head, we recruited 11 female and

three male native German speakers aged between 50 and 100 that required care,
but were still capable of accomplishing many of their daily tasks themselves. To
facilitate the questionnaires’ completion for the elderly, they were adapted to
their requirements. Questions were formulated shortly and clearly, and when-
ever possible the Likert-Scales were reduced to five options. Each item of the
questionnaire was printed on a separate page. Thus, at any time, participants
only had to deal with the information that was necessary for rating the current
verbalization.
Figure 3 shows the experimental setup. After a short introduction to the REA
system, the robot introduced itself and allowed the participant to choose a com-
fortable volume and speed of its voice, to ensure that the recommendations were
easy to understand. Then the robot explained the course of the study. In the
next phase, the robot presented the eight different wordings for each of the three
recommendations. To reduce the risk of order effects, the verbalizations were
presented in an incomplete counterbalanced order. After each item the dialog
paused while the participant answered the related questions. During the expla-
nation and the waiting phases, the robot’s head and face were subtly animated
with idle movements such as blinking or moving its head to make it appear
more alive and therefore more accessible. To avoid an influence on the ratings,
the robot stopped moving and returned to a neutral position when presenting a
recommendation. At the end, the robot thanked the participants for their help
and said goodbye. This last phase could be triggered earlier in case a participant
did not want to complete the entire study.
Fig. 3. Experimental setup: The study took place in a dining room at the retirement
home. The REA was placed on a table in front of the participant. The application
controlling the REA was running on a laptop which was positioned behind the robot
and turned away to hide the screen and keyboard from the participant’s view.
5.2 Results
Ten of the elderly people finished the study which took 45 to 60 min to complete.
Unfortunately, four participants wanted to terminate their participation early
because the procedure was too tiring for them.
In general, the participants provided relatively high ratings for the perceived
politeness as well as the perceived persuasiveness for all verbalizations. The
mean ratings of both criteria are graphically represented in Fig. 4. Similarly to
the first study, questions were perceived as most polite. Actions expressed as the
system’s goal also received high ratings regarding the perceived politeness. How-
ever, Socratic hints were perceived as much more polite than in the first study.
Direct commands also received much higher ratings, but were still perceived as
least polite. Regarding the perceived persuasiveness, most of the verbalizations
scored around 5.00. Only suggestions of user goals achieved a slightly lower
mean rating of 4.73. Strategies that achieved high ratings concerning both cri-
teria include indirect suggestions, requests and Socratic hints.
A repeated-measures ANOVA analysis showed significant differences between
the different wordings for the perceived politeness (F (7, 203) = 4.69, p < .0001),
but not for the perceived persuasiveness (F (4.39, 127.38) = .53, p > .05). A
subsequent Bonferroni post-hoc test showed that questions and actions expressed
as the system’s goals were rated significantly more polite than direct commands,
see Table 3. Nevertheless, the three strategies were rated as equally persuasive.
Observations of the seniors interacting with the REA indicated that none of
them was afraid of the robot or rejected it, although most of them had never seen
a system like REA before. After a few minutes of skepticism and doubts (“That’s
a machine. I can’t talk to it.”) almost all seniors did like the REA and talked to
it as if it would understand them. They stated that the robot is “very friendly”,
“very nice”, or even “really polite”. Several participants complimented it directly
and told stories about their families, their own childhood, adolescence, or current
life. One woman even wanted to hug it. Furthermore, some of the elderly people
tried to interpret the REA’s random non-verbal behavior when it incidentally
moved its head after they had talked to it.
Fig. 4. Results of the field study (Mean ratings for all recommendations). 1: direct
command, 2: indirect suggestions, 3: requests, 4: actions expressed as system goals, 5:
actions as shared goals, 6: questions, 7: suggestions of user goals, 8: Socratic hints
Table 3. Results of the user study: M = Mean; SD = Standard Deviation; Sig =

significantly better than
Politeness Persuasiveness
M SD Sig M SD Sig
1. Direct command 4.58 1.37 4.98 1.39
2. Indirect suggestion 5.19 1.04 5.16 1.13
3. Request 5.40 1.28 5.10 1.21
4. Actions expressed as system goals 5.58 .96 1∗∗ 5.06 1.18
5. Actions expressed as shared goals 5.10 1.38 4.97 1.58
∗∗
6. Questions 5.80 .96 1 4.97 1.38
7. Suggestions of user goals 5.30 1.47 4.73 1.66
8. Socratic hint 5.58 1.28 5.16 1.25
∗ ∗∗ ∗∗∗
significant with p <.05; significant with p <.01; significant with
p <.001
5.3 Discussion
The studies’ results showed that people do distinguish between the perceived
politeness and persuasiveness of different politeness strategies. However, com-
pared to the study with younger adults, with inhabitants of an retirement
home we found smaller and fewer significant differences concerning the perceived
politeness and no significant differences concerning the perceived persuasiveness.
One reason could be that the interrelation between politeness and persua-
siveness was perceived differently by the seniors. There is evidence that, while
several people found wording that they rated more polite also more persuasive,
other people rated wording that they found more polite less persuasive.
Furthermore, the comparably small differences between ratings of verbaliza-
tions may be caused by some limitations of the elderly. For example, since two
of the participants were no longer physically able to open a window, or go for
a walk, it was difficult for them to compare the different wordings. In addition,
some participants felt tired after a while, and people suffering from impaired
hearing had problems to understand the robot at certain points.
Nevertheless, all ratings provided by the elderly were relatively high and
all politeness strategies received higher ratings from the elderly than from the
younger people in the first study. This is in line with the findings of Nomura
et al. [24]. In a series of experiments they found that elderly people have more
positive impressions of a robot than younger people. Some interactions such as
telling private stories to the REA showed that, despite the short interaction
time, the elderly people already established some trust in the robot, which is an
important prerequisite to apply recommendations provided by a robotic system.
6 Conclusion
In the present contribution, we investigated linguistic variations to convey dif-
ferent levels of politeness in human-robot interaction. In the long run, a Robotic
Elderly Assistant (REA) should provide assistance for elderly people in the
form of recommendations to support them in their domestic environment. When
robots provide recommendations to the user, they permanently risk threatening
the user’s face. For example, a recommendation such as “You should be more
active.” could be perceived as offensive and demotivate users. Politeness strate-
gies may help mitigate face threats that might arise in such situations.
The results of a laboratory study with younger adults and a study with
elderly people in a retirement home showed that different politeness strategies
were perceived differently. In the laboratory study some wordings were perceived
neither polite nor persuasive, e.g. suggestions of user goals and Socratic hints.
Therefore, we think that these verbalizations should be ignored by a REA. In
contrast, wordings such as requests and especially actions formulated as shared
goals were perceived as polite and persuasive. They could be utilized as standard
strategies whenever a recommendation should be applied. Other wordings that
were not rated as very polite or very persuasive could nevertheless be applied by
a REA, e.g., to maintain the seniors’ interest in the REA by providing a wider
variety of behavior that could be adapted on the current situation. For example,
wordings that were perceived as being polite but not very persuasive, such as
questions or actions expressed as the system’s goal, could be utilized in uncritical
situations, where the recommendation does not necessarily have to be applied.
In this way, the relationship between the elderly person and the REA could be
further established. Vice versa, in critical situations, the system could emphasize
the need to apply a specific action by utilizing, for example, direct commands or
indirect suggestions that were assessed rather impolite, but convincing.
While some of the findings from the laboratory study could be confirmed in
the study with elderly users regarding the perceived politeness, the perceived
persuasiveness of the recommendations needs further investigation. Hence, we
will refine our studies to better match the requirements of our target user group.
For example, some ratings seemed to be influenced by the fact that certain
recommendations were not suitable for some of the participants due to physical
impairments, e.g. not being able to go for a walk. In a follow-up study, more
recommendations will be included that will be chosen based on the participant’s
preferences and abilities. Furthermore, it would be of interest to investigate
which of the recommendations of the REA will be carried out. It seems likely
that there is a gap between the abstract thinking of rating a recommendation
as being convincing and the real drive to carry out an action.
Serendipitously, the elderly people’s behavior during the user study showed
that they were very open minded towards our REA system and acted very
positively during the interaction. Therefore, we think that elderly people are
generally willing to accept recommendations provided by our REA, and aim
on contributing to the field of robotic elderly assistant systems by sharing our
insights on the usage of different politeness strategies in this domain.
Acknowledgments. The authors would like to thank the staff and inhabitants of the
retirement home “Paritätisches St. Jakobs-Stift Seniorenheim” in Augsburg, Germany,
for enabling the main study and providing valuable feedback on the design of the REA.
This research was partly funded by the Bavarian State Ministry for Education, Science
and the Arts (STMWFK) as part of the ForGenderCare research association.
References
1. Teri, L., Lewinsohn, P.: Modification of the pleasant and unpleasant events sched-
ules for use with the elderly. J. Consult. Clin. Psychol. 50(3), 444–445 (1982)
2. Mahncke, H.W., Bronstone, A., Merzenich, M.M.: Brain plasticity and functional
losses in the aged: scientific bases for a novel intervention. In: Møller, A.R., (ed.):
Reprogramming of the Brain, Progress in Brain Research, vol. 157, pp. 81–109.
Elsevier (2006)
3. Schmid, T. (ed.): Promoting Health Through Creativity: For Professionals in
Health, Arts and Education. Whurr Publishers, London (2005)
4. Seiderer, A., Hammer, S., André, E., Mayr, M., Rist, T.: Exploring digital image
frames for lifestyle intervention to improve well-being of older adults. In: Proceed-
ings of the 5th International Conference on Digital Health, DH 2015, pp. 71–78.
ACM, New York (2015)
5. Phang, C.W., Sutanto, J., Kankanhalli, A., Li, Y., Tan, B., Teo, H.H.: Senior
citizens’ acceptance of information systems: A study in the context of e-government
services. IEEE Trans. Eng. Manage. 53(4), 555–569 (2006)
6. Monk, A., Hone, K., Lines, L., Dowdall, A., Baxter, G., Blythe, M., Wright, P.:
Towards a practical framework for managing the risks of selecting technology to
support independent living. Appl. Ergon. 37(5), 599–606 (2006)
7. Reeves, B., Nass, C.: The Media Equation: How People Treat Computers, Televi-
sion, and New Media Like Real People and Places. Cambridge University Press,
New York (1998)
8. Fong, T., Nourbakhsh, I., Dautenhahn, K.: A survey of socially interactive robots.
Robot. Auton. Syst. 42(3–4), 143–166 (2003)
9. Heerink, M.: Assessing acceptance of assistive social robots by aging adults. Ph.D.
thesis, University of Amsterdam (2010)
10. Looije, R., Neerincx, M.A., Cnossen, F.: Persuasive robotic assistant for health
self-management of older adults: Design and evaluation of social behaviors. Int. J.
Hum. Comput. Stud. 68(6), 386–397 (2010)
11. Spiekman, M.E., Haazebroek, P., Neerincx, M.A.: Requirements and platforms for
social agents that alarm and support elderly living alone. In: Mutlu, B., Bartneck,
C., Ham, J., Evers, V., Kanda, T. (eds.) ICSR 2011. LNCS, vol. 7072, pp. 226–235.
12. Broadbent, E., Stafford, R., MacDonald, B.: Acceptance of healthcare robots for
the older population: Review and future directions. Int. J. Soc. Robot. 1(4), 319–
330 (2009)
13. Frennert, S., Östlund, B., Eftring, H.: Would granny let an assistive robot into her
home? In: Ge, S.S., Khatib, O., Cabibihan, J.-J., Simmons, R., Williams, M.-A.
(eds.) ICSR 2012. LNCS, vol. 7621, pp. 128–137. Springer, Heidelberg (2012)
14. Wu, Y.H., Fassert, C., Rigaud, A.S.: Designing robots for the elderly: appearance
issue and beyond. Arch. Gerontol. Geriatr. 54(1), 121–126 (2012)
15. Goetz, J., Kiesler, S., Powers, A.: Matching robot appearance and behavior to tasks
to improve human-robot cooperation. In: The 12th IEEE International Workshop
on Proceedings of Robot and Human Interactive Communication, ROMAN 2003,
pp. 55–60, October 2003
16. Walters, M., Syrdal, D., Dautenhahn, K., te Boekhorst, R., Koay, K.: Avoiding
the uncanny valley: robot appearance, personality and consistency of behavior in
an attention-seeking home scenario for a robot companion. Auton. Robot. 24(2),
159–178 (2008)
17. Sidner, C., Lee, C.: The initiation of engagement by a humanoid robot. In: AAAI
Spring Symposium on Multidisciplinary Collaboration for Socially Assistive Robot-
ics (2007)
18. Nomura, T., Saeki, K.: Effects of polite behaviors expressed by robots: A psycho-
logical experiment in Japan. Int. J. Synth. Emot. 1(2), 38–52 (2010)
19. Mayer, R., Johnson, W., Shaw, E., Sandhu, S.: Constructing computer-based tutors
that are socially sensitive: Politeness in educational software. In: Conference of the
American Educational Research Association (2005)
20. Brown, P., Levinson, S.: Politeness: Some Universals in Language Usage. Studies
in Interactional Sociolinguistics. Cambridge University Press, Cambridge (1987)
21. Johnson, W.L., Mayer, R.E., André, E., Rehm, M.: Cross-cultural evaluation of
politeness in tactics for pedagogical agents. In: Proceedings of the 2005 Conference
on Artificial Intelligence in Education: Supporting Learning Through Intelligent
and Socially Informed Technology, Amsterdam, Netherlands, IOS Press, pp. 298–
305 (2005)
22. Gebhard, P., Mehlmann, G., Kipp, M.: Visual Scenemaker - a tool for authoring
interactive virtual characters. Multimodal User Interfaces 6(1–2), 3–11 (2012)
23. Mehlmann, G., Janowski, K., André, E.: Modeling grounding for interactive social
companions. KI - Künstliche Intelligenz 30, 1–8 (2015)
24. Nomura, T., Takeuchi, S.: The elderly and robots: From experiments based on
comparison with younger people (2011)
RightOnTime: The Role of Timing and Unobtrusiveness
in Behavior Change Support Systems
Piiastiina Tikka ✉ and Harri Oinas-Kukkonen

( )
Oulu Advanced Research on Service and Information Systems Research Group (OASIS),
University of Oulu, Oulu, Finland
{piiastiina.tikka,harri.oinas-kukkonen}@oulu.fi
Abstract. Influencing people’s behavior by means of technology is achievable

under many technological guises from websites to mobile devices to activity
trackers, making them practically ubiquitous. The timing of persuasive messages
has been found to be influential in itself, but the omnipresence of modern
computing also raises questions about the effects of randomly timed interruptions
and perceived obtrusiveness. The presented research is an explorative experiment
regarding the unobtrusiveness of a behavior change support system. The study
compares two timing strategies: random timing and user-defined timing for
system interaction. While the results of the mixed ANOVA analyses in this pilot
study did not yield statistically significant results between the timing strategies,
the correlations found do seem to indicate that random interruptions are perceived
as more obtrusive than user-timed interaction. Furthermore, a correlation was
observed between task success, task satisfaction and perceived unobtrusiveness.
Keywords: Persuasive systems · Behavior change · Timing strategies ·

Interruptions · Persuasive systems design · User experience · Unobtrusiveness
1 Introduction
Use of information systems to influence behavior has been increasing in popularity

in recent years. These systems are available in various forms both on desktops and
in mobile devices and range from applications in the realm of the quantified self
movement to applications and systems aimed at helping with lifestyle changes such
as weight loss or even clinical or therapeutic systems that deal with conditions such
as depression [1, 2] or diabetes management [3]. The behavioral problem domains
addressed by the many systems are various, but in persuasive technology research the
concept of Behavior Change Support System (BCSS) [4, 5] draws together the key
elements of the systems themselves as information systems that are designed to form,
alter and/or reinforce compliance, attitudes, and/or behaviors, and do so without
using coercion or deception [5].
In the present experiment the research focuses on one important element of
persuasive systems: the timing of persuasive communications. Such timing of persua‐
sive messages has been suggested to have an effect on behavior [6, 7]. How the timing

DOI: 10.1007/978-3-319-31510-2_28
328 P. Tikka and H. Oinas-Kukkonen
of persuasive messages is managed can also affect an end-user’s experience of obtru‐

siveness, since system-originated messages can produce interruptions and demand
attention at moments that an end-user deems irritating [8].
The aim of the present study, therefore, is to shift the focus away slightly from trying
to accurately spearfish for those opportune moments, and instead try and trawl for a
coarser understanding of how persuasive message timing strategies affect behavior,
behavior change and the user experience as regards the system. The key research ques‐
tion in the present pilot study, then, is: How do timing strategies affect behavior change
and perceived unobtrusiveness when using BCSSs?
2 Background
There are essentially two perspectives on persuasive message timing that are present in
this paper: the perceived interruptions from a system demanding attention, and the search
for the opportune moment for demanding that attention. The objective here is not to
pinpoint those opportune moments as such, but rather that the need to find them should
be taken into account when studying timing issues. The present study aims to support
that quest by exploring the broader scope of timing strategies. First, there is the effect
of increased cognitive load that results from random interruptions demanding active
attention [e.g. 8–10]. The second element is the concept of “an opportune moment” (or
‘kairos’ from the Greek) in the design of persuasive systems: the problem of when best
to communicate a persuasive message to the system user for maximum effect [6, 7].
Cognitive load theory [11, 12] proposes that high working memory effort will have
a negative effect on problem solving and other cognitive processes. The implications
are notable for, say, learning [11] and decision-making [13]. Research into attention and
interruptions has shown that attention is a limited cognitive capacity and that switching
attention from one task to another comes at a cost to task performance [14]. Randomly
timed interruptions have been found to have a negative effect on the performance of a
main task [8, 9, 15] and such interruptions also increase negative affect as regards the
system [8, 9]. Even in a broader context than specific tasks, some research suggests that
extended and continuous exposure to constant interruptions by messaging and social
media can have a detrimental effect on learning [16].
Miyata and Norman [10] argue that notifications from a system would have a lower
interruption cost if delivered at a moment of lower mental workload. In other words,
they advocate identifying kairos for sending a notification to a system user. Drawing
upon this, a host of further research aims at identifying such moments for various systems
to make use of [8, 9, 15]. For persuasive systems, the other type of ‘opportune moment’
has been identified as a factor in increasing the persuasive power of a message [6, 7].
Using the definition by Kinneavy [17], ‘kairos’ indicates an opportune moment and also
the right measure of action. How much, how often and how frequently would all seem
reasonable measures when discussing elements affecting perceived unobtrusiveness.
Earlier studies into factors affecting use and intention to use persuasive systems
indicate that perceived unobtrusiveness affects at least the intention to use, if not the
actual use of a system [18], and that it has an effect also on perceived persuasiveness of
a persuasive system [19]. The PSD model [20] proposes that unobtrusiveness should be
RightOnTime: The Role of Timing and Unobtrusiveness 329
a key consideration in the design of persuasive systems: these systems should not disturb
or interrupt users when they are carrying out their main tasks, and poorly executed timing
of persuasive communications can lead to less than ideal results. Irritation resulting from
interruptions may lead to levels of negative affect that influence one’s intention to use
a system or its actual use. The balance between too low involvement, where the neces‐
sary effort associated with learning and reflection (both in-action and on-action) [21]
does not take place, and too active demand on attention through interruptions is of great
interest in efforts to design better persuasive systems. A balance between reflection as
a behavior change tool and the level of attention demand is necessary: after all, for self-
reflection to take place at all, some attentional resources must be allocated to the process.
Context-aware systems are an absolute necessity in ensuring at least a minimum
level of control over interruptions [22]. Irritation or frustration from interruptions is not
conducive to successful persuasion, as we have been found to be more open to persuasion
when we are in a good mood [6]. Furthermore, having control over when to take a break
has been found to produce better problem-solving results than if break times are dictated
by someone else [23].
3 Research Setting
3.1 Hypothesis
The research aimed at discovering whether the type of persuasive communication timing
strategy had any affect on persuasion outcome and Perceived Unobtrusiveness (PU) of
the system. Based on earlier research on the topic, as presented in the Background
section, the following hypotheses were formed: (1) Random Timing (RT) is perceived
to be more obtrusive than User-defined Timing UT; (2) RT has a stronger positive
influence on behavior change both in self-assessed task success than UT; and (3) UT
results in a more positive response in terms of Perceived Unobtrusiveness than RT.
Should the presented hypotheses hold true, an interesting conflict emerges in persua‐
sion outcome and perceived unobtrusiveness: a better result in terms of behavioral
influence may lead to higher perceived irritation associated with the system in question.
The discussion should then focus on how to address this imbalance in persuasive systems
design.
3.2 Study Design
The experiment was a mixed design (N = 13) comparing two timing strategies, imple‐
mented in a purpose-built iPhone app. The experimentation period was five days, with
additional days for pre- and post-test assessments. For the basic analysis of how the
different types of interruption affected punctuality performance and perceived satisfac‐
tion over a period of time, a mixed ANOVA (analysis of variance) design was selected.
The within-subjects factor was self-assessment ratings collected over the experimenta‐
tion period, with the between-subject factor being the timing strategy. The dependent
variables there were self-reported success ratings and perceived satisfaction ratings.
Participants were divided into one of two groups as per timing strategy.
Five measures of both dependent variables were used. The participants reported on
more than one punctuality task each day, and the average of the ratings for each day was
used as the daily score. As the daily tasks were carried out independently by each partic‐
ipant without close supervision by researchers there were individuals who did not always
complete three tasks per day and individuals who may have reported more than three
tasks on some days. Listwise deletion approach was deemed most suitable for handling
missing values.
3.3 Procedure (Data Collection)

Participants (N = 13) in the experiment represent a sample of opportunity. As an incen‐
tive to take part, completion of the experiment entered participants in a prize draw. At
the time of recruitment participants were not told what the exact topic of the experiment
was: the only information offered was a broad outline of the effort expected of them and
the explanation that the experiment was to do with behavior change. As participants
signed up, they were informed that they could withdraw from the study at any time if
they so wished.
The participants were screened for punctuality schematicism. In short, self-schema
refers to a person’s perception of his or her self in terms of his or her specific environment
[24] and it is a construct linked with transition of intention into behavior [24–26].
Research into behavioral intention and self-schemas have established that a positive self-
schema towards a specific behavior acts as a moderator in turning intention into behavior
[25] and that schema-matched persuasion can result in more effective change [e.g. 27].
The purpose of using such an attitude scale was to ensure that there were participants
who would fall within the target group for an app that addressed punctuality problems.
Short of finding volunteers who freely admitted to having problems with time keeping,
using a punctuality self-schema assessment meant that we were able to identify partic‐
ipants who did not, perhaps, have the need to be on time as a prominent part of their
self-identity; we could then assign such participants to both test conditions (RT and UT).
While the sample size in the present experiment is not overly large (N = 13), it was
estimated using G*Power analysis tool [28]. Initially, 40 potential participants signed
up, but for varying (personal) reasons many of them dropped out and did not complete
the experiment. Aiming at effect size of .4 and power of .95 for two groups and five
measures, the estimated total sample size required was 14. We were unable to meet this
target minimum sample size in the time available for the study. Latin squares were used
for balancing punctuality schematics and non-schematics between the conditions (Group
A and Group B) at the start of the experiment, but owing to participants’ freedom to
drop out of the experiment at any time it was not possible to control the final outcome
of schematics/non-schematics per condition or to keep the test groups exactly the same
size. Overall, 14 participants completed the tasks in a satisfactory manner, but one
participant failed to complete the required information regarding schematicism and
therefore had to be dropped from the analysis (Table 1).
Table 1. Schematic and non-schematic test participants in the experiment’s timing conditions.
Timing/punctuality type Schematic Non-schematic Total

Random 5 3 8
User-defined 3 2 5
Total 8 5 13
After the schematicism screening the participants also filled in a background infor‐
mation questionnaire and were instructed how to download the app and use it with the
relevant settings. Finally, after completing the tasks on the app, the participants were
sent links to the final questionnaires by e-mail.
The two groups of participants were each instructed as to the specific settings they
had to use on the test app in the experiment. Group A were the ‘Random Timing’ group,
where the system would send them tips and hints throughout the day and also query their
responses regarding task success and satisfaction some time after the completion of each
task. Group B were asked to set one time for daily interactions with the system; in this
way all the communications and requests to fill in the self-assessments were sent to the
user once a day at a time of the user’s own choosing. For both groups the number of tips
and hints was limited to three so that frequency would remain the same between the
groups.
During the five experiment days all participants set themselves three timed tasks per
day and one time-estimation practice task per day. The timed tasks could be anything
that the participants determined they needed to be on time for or anything that they
decided they wanted to do at some specific time. In the practice task the participants
were asked to select some routine such as breakfast, journey to work, etc. and learned
to estimate how long the activity actually took. The participants submitted the self-
evaluations to the research database using a ‘send’ button in the test app.
3.4 Materials
Mobile Device Test App. A purpose-made native iOS app called ‘RightOnTime’ was
designed and developed by the research team and was entered into Apple Store as a free
download. The particular purpose of the app was to help in rehearsing time management
skills that are associated with many cases of chronic tardiness and it also provided
features for actively practicing time-awareness and to encourage personal time-manage‐
ment related self-reflection. The app, then, was designed to support users in developing
and/or maintaining punctual behavior patterns by making them see when their behavior
was not based on realistic perceptions of time (or, indeed, when they were punctual and
timely). The problem-domain content used in the app was a collection of existing
knowledge and suggestions from the literature [29] and self-help websites1. In the
experiment the app’s two settings configuration possibilities were used in studying the
difference in response as regards perceived unobtrusiveness (Fig. 1).
Fig. 1. RightOnTime app: Home screen (left) for setting tasks, and an example of a self-
evaluation view (competed after a task is done).
The app had two main parts: user-set punctuality targets/tasks and time-awareness prac‐
tice. In the first category the user would set a number of timed events for each day. These
could be any tasks or activities that the user wanted to be on time for. Each activity
would be given an exact starting time and an estimation of duration. The main objective
was to increase awareness of important meetings or tasks and to learn in that way to
make the effort of getting there on time. Once the task was over, the system would prompt
users to evaluate their performance in terms of timeliness. The prompt was set to appear
within a 15-min frame after completion of the task so that it was not entirely predictable.
Time-awareness assessments were not in the focus of the present pilot study.
A selection of Persuasive Systems Design [20] features was used in the design of
the app: reflection (in action as well as on action [21]) is very much in the foreground
not only in the form of Self-monitoring as such, but by generally offering the user ways
of keeping the behavior change target in mind through continuous interactions, practice,
etc. Further PSD features [20] used in the app included reduction, reminders, suggestion
and real-world feel (see Table 2).
The app itself was constructed using iOS components and style as much as possible
in order to minimize the effect of design on UX while maintaining an acceptable level
of interface familiarity, functionality and usability.
1
How To Be Punctual. http://www.wikihow.com/Be-Punctual. (Date of reference 4.8.2015).
Prolific Living. The Importance of Being Punctual. http://www.prolificliving.com/the-impor‐
tance-of-being-punctual/ (Date of reference 4.8.2015).
Table 2. Persuasive system design features used in the RightOnTime app.
Primary task support Reduction An overall goal of “be

more punctual” may
seem too big to manage,
so it is good to split the
target into smaller and
more manageable indi‐
vidual tasks
Self-monitoring Self-assessment data and
summaries offer an
opportunity to monitor
one’s development as
regards the behavior
change targets
Computer-human dialogue support Reminders The system reminds the
user to complete the
self-assessments and to
view daily assessment
summaries
Suggestion The system sends the user
tips and hints (informa‐
tion, encouragement
and guidance) as push
notifications
System credibility support Read-world feel An “About” section in the
system provides infor‐
mation about the
designers, the origins
and the purpose of the
app
Attitude Questionnaire Development. A self-schema assessment questionnaire was

devised based on an established model by, for example, [24]. The questionnaire gauged
each participant’s self image as regards punctuality by querying representative punctu‐
ality statements (measures). The use of an 11-point scale on the two measures (“describes
me/does not describe me” and “is important to me/is not important to me”) has been
used in earlier research on the relationship between self-schemas and behavior by, for
example, [24, 30]. As such, the questionnaire style both in the listed studies and in the
present experiment is a direct measure, but filler items were used in order to disguise
the focus of the questionnaire at the beginning of the experiment. Also, the participants
were not told what behavior specifically was going to fall within the scope of the test
and the filler items referred to other lifestyle topics such as healthy eating. As mentioned,
attitude was queried.
Perceived Unobtrusiveness UX Questionnaire Development. In order to gauge the

final experience of unobtrusiveness after using the test app for five days, a questionnaire
to that effect was developed. By using established UX questionnaires2,3 a perceived persua‐
siveness questionnaire [19, 31] the PSD model [20] and other related material [32, 33], five
expert evaluators scored the statements, directing thus which statements were included in
the final questionnaire. Again, when administering the questionnaire, filler items were
added so as to diffuse the focus of the respondents.
4 Results
4.1 Participants and Sample Characteristics
Initially, a total of 40 people signed up to take part in the experiment initially. Of those,
31 completed the schematicism screener and were assigned to the test groups. Subse‐
quently, more than half dropped out at various stages during the experiment period. 13
participants completed the entire experiment. Of the total initial number of interested
participants (40) nine were non-schematics as regards punctuality. In the end, five of
these nine completed the experiment and were included in the analysis.
Sample Description. Average age for the sample was 36 years, with the youngest
participant being 23 and the oldest 72 years of age. The oldest participant was notably
older than the rest, but the participant was an experienced iPhone user, had no charac‐
teristics in choice of daily tasks or otherwise that stood out from the rest of the sample
and so the participant was not considered an outlier. Nine participants out of the 13 were
female, but the analysis does not consider gender a critical factor within the scope of
the experiment. A clear majority (11) had used in iPhone for more than six months,
making them familiar with the test device itself. As regards punctuality schematicism,
the sample includes more schematics than non-schematics, meaning that the majority
of the participants identify themselves as people who value punctuality and timeliness.
4.2 Data Analysis
The collected data was analyzed for variance both for task success (Tsucc) and task
satisfaction (Tsat). These scores were also checked for correlation with Perceived Unob‐
trusiveness (PU). The PU scores were also analyzed for variance between the test groups.
Analysis of variance showed no significant effect of time of measurement in the Task
success scores (time of measurement here referring to each of the five days of the
experiment). The daily score used in the analysis is an averaged score from the three
2
NASA TLX: Task Load Index. http://humansystems.arc.nasa.gov/groups/tlx/.
3
System Usability Scale (SUS). http://www.usability.gov/how-to-and-tools/methods/system-
usability-scale.html.
tasks participants reported each day in order to show a more consolidated day-by-day
view of the development over time. The RT group’s task scores were measured at
different times of the day, depending upon what time their set tasks took place, while
the UT group reported all three scores at one self-set time. The extended delay between
completing a task and reporting it in retrospect was part of the timing strategy difference
between the conditions.
The collected means in Fig. 2 show the development of means for each variable and
factor over the five days. The interesting movements in the chart (Fig. 2) concern the
beginning and the end of the period. At the beginning User-defined timing group scores
drop after the first day before they gradually climb back up to the starting levels and
even beyond. The Random timing group is seen to climb up to the same levels at the
end as the user-defined timing group, but task satisfaction in particular appears to climb
up in a steady and even manner.
Fig. 2. Self-evaluation scores: means for each group per day.
UT group participants’ Tsucc and Tsat scores compared with PU scores pointed to
a very small linear relationship (r = −31 for Tsucc, r = −.20 for Tsat). In the RT group
the correlation coefficients indicated a stronger positive linear relationship, with r = .62
for Tsucc and r = .56 for Tsat. These indications of strength in the linear relationship
are extremely interesting considering that the PU score is given on negative statements:
the higher the score, the stronger the agreement with statements that indicate obtrusive‐
ness. In other words, participants in Group A (Random Timing) tended to rate the system
more obtrusive as their Task Success and Task Satisfaction increased. In the UT group,
on the other hand, the system was rated less obtrusive as the Task Success and Task
Satisfaction increased.
One of the initial hypotheses was that Random Timing of persuasive messaging may
be experienced as being more obtrusive than timing set by the system user. We do not have
exact information as to the reasons why participants dropped out from the experiment,
as it was their right to do so without explanation, but it is worth noting that of the 31 initial
participants assigned to test groups 17 were assigned to Random Timing and 15 to User-
defined Timing groups. For these groups, the completion rate was only 5 participants
(approx. 30 %) in the RT group and 8 (approx. 50 %) in the UT group. One possible avenue
of thought for explaining the difference in the drop-out rate could be the difference in the
obtrusiveness between the two experiment conditions.
A one-way ANOVA of the PU scores does not show statistical significance variance
between the groups either, but as predicted, it is the RT group whose scores point towards
a more pronounced experience of obtrusiveness. Therefore, while performance related
scores improved over the experiment period, the participants in the RT group perceived the
system to be more obtrusive than the participants in the UT group. In this questionnaire a
higher score (the ‘agree’ end of the scale) indicated agreement on negative statements. In
other words, here, too, there is some slight indication that Random timing might be expe‐
rienced as more obtrusive than User-defined timing of persuasive messaging.
The primary purpose of the experiment was to explore the problem field around the
concept of unobtrusive user experience with persuasive systems so that a more robust
exploration could then be carried out by means of a well-informed model of influencing
factors.
The difference in the correlation coefficients between the test groups is a particularly
interesting avenue of analysis. When using a system that sends messages, reminders and
evaluation requests at random times, better perceived performance (Tsucc) and higher
satisfaction in one’s own achievement (Tsat) do not seem to translate into an unobtrusive
experience of the system. By contrast, there was an indication of the opposite effect
when system users dealt with the system only at a time they had chosen themselves.
Admittedly, the linearity in this case was weak, but it was observably in a different
direction than with the Random Timing group. The difference prompts various questions
as regards the role of cognitive load in perceived unobtrusiveness. How strongly does
the awareness and sense of being disturbed, interrupted or otherwise reminded of an
on-going task at random moments result in negative affect towards the system? In this
small sample and over this short period of time performance and satisfaction did not
differ significantly, and neither did the final experience scores; but does the correlation
result suggest potential problems in terms of perceived unobtrusiveness in a longer time
frame? Or does this suggest that it might be possible to achieve acceptable behavior
change results even when minimizing interruption-based cognitive load?
As it was, statistically we ended up with an inconclusive set of results: the differences
between the groups in terms of Task Success and Task Satisfaction ratings are not
significant, and while there is an observable difference between the Perceived Unob‐
trusiveness rating between the conditions, the difference is not statistically significant.
The significance of the drop-out rate difference between the two groups calls for further
exploration in order to determine if the perceived unobtrusiveness was even more
pronounced in the Random Timing group than presently reported. As it was, the freedom
to drop out of the experiment does not allow for more than speculative conclusions based
on the fact that more participants dropped out from the Random Timing group than from
the User-defined Timing group.
Further research is certainly required, but the findings from this study can be used in
identifying meaningful factors that are at play in defining appropriate timing strategies. One
way of looking at the relationship between ‘timing strategies’ and an ‘opportune moment’
is to regard ‘opportune moment’ as one end of a continuum and an abysmally inappro‐
priate moment as the other. Where a persuasive system falls on that continuum can depend
on the timing strategies, whether random vs. set, sensor-based or context-based, etc. The
ideal is to hit upon the kairos moment and achieve behavior change with minimal cogni‐
tive load, but such rarities apart, we must search for ways of providing an unobtrusive user
experience without compromising the behavior change target.
References
1. Kuonanoja, L., Lanqrial, S., Lappalainen, R., Lappalainen, P., Oinas-Kukkonen, H.: Treating
depression with a behavior change support system without face-to-face therapy. AIS Trans.
Hum. Comput. Interact. 7(3), 192–210 (2015)
2. Lappalainen, P., Langrial, S., Oinas-Kukkonen, H., Tolvanen, A., Lappalainen, R.: Web-
based acceptance and commitment therapy for depressive symptoms with minimal support:
treatment efficacy and participants’ experiences: a randomized controlled trial. Behav. Modif.
39(6), 805–834 (2015)
3. NovoNordisk. Cornerstones4Care Diabetes Healthcoach, 30 July 2015. https://
www.cornerstones4care.com
4. Oinas-Kukkonen, H.: Prerequisites for successful measuring of ambient persuasive
technology, pp. 491–494 (2010)
5. Oinas-Kukkonen, H.: A foundation for the study of behavior change support systems. Pers.
Ubiquit. Comput. 17(6), 1223–1235 (2013)
7. Räisänen, T., Oinas-Kukkonen, H., Pahnila, S.: Finding kairos in quitting smoking: smokers’
perceptions of warning pictures. In: Oinas-Kukkonen, H., Hasle, P., Harjumaa, M.,
Segerståhl, K., Øhrstrøm, P. (eds.) PERSUASIVE 2008. LNCS, vol. 5033, pp. 254–257.
8. Bailey, B.P., Iqbal, S.T.: Understanding changes in mental workload during execution of
goal-directed tasks and its application for interruption management. ACM Trans. Comput.
Hum. Interact. 14(4), 1–28 (2008)
9. Bailey, B.P., Konstan, J.A.: On the need for attention-aware systems: measuring effects of
interruption on task performance, error rate, and affective state. Comput. Hum. Behav. 22(4),
685–708 (2006)
10. Miyata, Y., Norman, D.A.: The control of multiple activities. In: Norman, D.A., Draper, S.W.
(eds.) User-Centered System Design: New Perspectives on Human-Computer Interaction.
Lawrence Erlbaum Associates, Hillsdale (1986)
11. Paas, F., Renkl, A., Sweller, J.: Cognitive load theory: instructional implications of the
interaction between information structures and cognitive architecture. Instru. Sci. 32(1/2),
1–8 (2004)
12. Sweller, J.: Cognitive load during problem solving. Cogn. Sci. 12, 257–285 (1988)
13. Vohs, K.D., Baumeister, R.F., Schmeichel, B.J., Twenge, J.M., Nelson, N.M., Tice, D.M.:
Making choices impairs subsequent self-control: a limited-resource account of decision
making, self-regulation, and active initiative. J. Pers. Soc. Psychol. 94(5), 883–898 (2008)
14. Kahneman, D.: Attention and Effort. Prentice-Hall Inc., Englewood Cliffs (1973)
15. Czerwinski, M., Cutrell, E., Horvitz, E.: Instant messaging: effects of relevance and timing. In:
People and Computers XIV: Proceedings of HCI, British Computer Society, pp. 71–76 (2000)
16. Frein, S.F., Jones, S.L., Gerow, J.E.: When it comes to Facebook there may be more to bad
memory than just multitasking. Comput. Hum. Behav. 29(6), 2179–2182 (2013)
17. Kinneavy, J.L.: Kairos: a neglected concept in classical rhetoric. In: Moss, J.D. (ed.) Rhetoric
and Praxis: the Contribution of Classical Rhetoric to Practical Reasoning, pp. 79–105. The
Catholic University of America Press, Washington (1986)
18. Drozd, F., Lehto, T., Oinas-Kukkonen, H.: Exploring perceived persuasiveness of a behavior
change support system: a structural model. In: Bang, M., Ragnemalm, E.L. (eds.)
19. Lehto, T., Oinas-Kukkonen, H., Drozd, F.: Factors affecting perceived persuasiveness of a
behavior change support system. In: ICIS 2012 Proceedings of Human Behavior in IT
Adoption and Use: Paper 18 (2013)
20. Oinas-kukkonen, H., Harjumaa, M.: Persuasive systems design: key issues, process model,
and system features. Commun. Assoc Inf. Syst. 24(1), 28 (2009)
21. Ploderer, B., Reitberger, W., Oinas-Kukkonen, H., van Gemert-Pijnen, J.: Social interaction
and reflection for behaviour change. Pers. Ubiquit. Comput. 18(7), 1667–1676 (2014). doi:
10.1007/s00779-014-0779-y
22. Anhalt, J., Smailagic, A., Siewiorek, D.P., Gemperle, F., Salber, D., Weber, S. Jennings, J.:
Toward context-aware computing: experiences and lessons. IEEE Intell. Syst. 16(3), 38–46
(2001)
23. Beeftink, F., van Eerde, W., Rutte, C.G.: The effect of interruptions and breaks on insight and
impasses: do you need a break right now? Creativity Res. J. 20(4), 358–364 (2008)
24. Markus, H.: Self-schemata and processing information about the self. J. Pers. Soc. Psychol.
35(2), 63–78 (1977)
25. Sheeran, P., Orbell, S.: Self-schemas and the theory of planned behaviour. Eur. J. Soc.
Psychol. 30, 533–550 (2000)
26. Cacioppo, J.T., Petty, R.E., Sidera, J.A.: The effects of a salient self-schema on the evaluation
of proattitudinal editorials: top-down versus bottom-up message processing. J. Exp. Soc.
Psychol. 18, 324–338 (1982)
27. Wheeler, C.S., Petty, R.E., Bizer, G.Y.: Self-schema matching and attitude change: situational
and dispositional determinants of message elaboration. J. Consum. Res. 31, 787–797 (2005)
28. Faul, F., Erdfelder, E., Lang, A.-E., Buchner, A.: G*Power 3: a flexible statistical power
analysis program for the social, behavioral, and biomedical sciences. Behav. Res. Methods
39(2), 175–191 (2007)
29. DeLonzor, D.: Never Be Late Again, 7 Cures for the Punctually Challenged. Post Madison
Publishing, San Francisco (2003)
30. Kendzierski, D.: Self-schemata and exercise. Basic Appl. Soc. Psychol. 9(1), 45–59 (1988).
http://doi.org/10.1207/s15324834basp0901_4
31. Lehto, T. Oinas-Kukkonen, H.: Explaining and predicting perceived effectiveness and use
continuance intention of a behaviour change support system for weight loss. Behav. Inf.
Technol. 34(2), 176–189 (2015). http://dx.doi.org/10.1080/0144929X.2013.866162
32. Baddeley, A.D.: Working memory. Science 225, 556–559 (1992)
33. Rubinstein, J., Meyer, D., Evans, J.: Executive control of cognitive processes in task
switching. J. Exp. Psychol. Hum. Percept. Perform. 27(4), 763–797 (2001)
Persuasive Information Security:
Techniques to Help Employees Protect
Organizational Information Security
Marc Busch1(B) , Sameer Patil2,3 , Georg Regal1 , Christina Hochleitner1 ,

and Manfred Tscheligi1
1
AIT Austrian Institute of Technology GmbH, Giefinggasse 2,
1210 Vienna, Austria
{March.Busch,Georg.Regal,Christina.Hochleitner,
Manfred.Tscheligi}@ait.ac.at
2
Department of Computer Science and Engineering, New York University,
Brooklyn, NY 11201, USA
[email protected]
3
Helsinki Institute for Information Technology HIIT / Aalto University,
P.O. Box 15600, 00076 Aalto, Finland
Abstract. Digital information is an important corporate asset. Orga-

nizations typically devise policies and guidelines to help employees pro-
tect the security of such information. Complying with these rules can
often be confusing and difficult and may obstruct the task at hand,
leading employees to circumvent or ignore policies. Commercial tech-
nology and training programs to mitigate this issue suffer from various
shortcomings. To address these shortcomings, we designed six persuasive
features: Security Points, Security Quiz, Challenges, Statistics, Personal-
ization, and Risk Communication. A design probe that implemented the
features sheds light on how persuasive security features could influence
intentions to follow secure work practices. We apply the findings to offer
suggestions for further enhancing the six persuasive features.
Keywords: Usable security · Persuasive design · Information security

policy · Security compliance · Organizational information systems
1 Introduction
Breaches in organizational information security can lead to the loss or theft of

sensitive digital information, potentially costing significant financial amounts
and damaging the organization’s reputation. Studies show that a sizable propor-
tion of information security breaches are caused inadvertently in the course of
routine work of employees, despite absence of malicious intent. A recent study
This research was conducted when Dr. Patil was a researcher at the Helsinki Institute
for Information Technology HIIT / Aalto University, Finland.

DOI: 10.1007/978-3-319-31510-2 29
340 M. Busch et al.
that included more than 1400 employees of 277 companies in 9 countries found
that 35 percent of data breaches were caused by such “human factors” [5].
One of the mechanisms organizations use for the protection of information
is an information security policy, i.e., the rules and guidelines defining permit-
ted and forbidden actions related to information assets. Naturally, organiza-
tions expect employees to adhere to the policy. However, employees often find
it challenging to comply. Factors underlying these difficulties include individual
attitudes toward the policy, beliefs about organizational norms, and efficacy to
comply [1].
Solutions to ensure compliance with information security policies may be
technology- or human-centric. Enforcement via technical means (e.g., denying
access when the network connection is insecure) takes away employee autonomy
and may lead to potentially insecure workarounds, thus undermining, rather than
enhancing, security. Moreover, automated technological enforcement affects only
the corresponding work practices, not employee attitudes. In contrast, human-
centric approaches (such as training and awareness programs [12,17]) aim to
influence attitudes and motivate employees to follow secure information prac-
tices. However, such approaches are expensive and time consuming and need
repetition. The effectiveness of security education programs is further limited
because they are typically decontextualized from work practices and decoupled
from the situations under consideration (viz., actual policy breaches).
We believe that a key to enhancing the effectiveness of such tools and methods
is educating employees regarding the risks and rationale that form the basis of the
policy and engaging them by incentivizing secure information practices. Toward
this end, we designed six features aimed at increasing employee awareness and
knowledge of the organizational information security policy and changing atti-
tudes and behavioral intention toward compliance. Our goal was to identify
the effectiveness and promise of the features for raising employee awareness of
the organization’s information security policy and increasing policy compliance.
Specifically, we explored the following research question: Which persuasive fea-
tures are likely to effect a change in attitudes and behavior intention regarding
organizational information security?
To tackle the above question, we carried out a user study in which we
evaluated the perceived persuasiveness of these features via an interface imple-
mentation in the form of a design probe. We found that persuasive strategies
could indeed be beneficial for promoting secure work practices, albeit to varying
extents. We apply our findings to offer suggestions for further exploration of the
design space of persuasive security features.
Our contribution lies in the application of persuasive strategies for organiza-
tional security. Persuasive technology has not been systematically explored for
usable security in general and organizational security in particular. By extend-
ing persuasive design to this domain, we promote an approach to security policy
compliance that engages with and educates the human in the loop.
Persuasive Information Security: Techniques to Help Employees Protect 341
2 Related Work
Technological approaches to improve organizational information security pol-
icy compliance include commercial applications that manage endpoint security.
Such applications monitor company-owned devices and networks and enforce
access controls for information assets. For instance, the application could for-
bid access via a non-trusted and insecure network. The user interfaces of these
applications are typically geared toward information security managers and sys-
tem administrators, rather than end-users of the devices. Further, the systems
often enforce information security policies without helping end users under-
stand the importance of the rules and the consequences of violations. To mit-
igate this shortcoming, approaches such as training [12] and security aware-
ness campaigns [10] have been suggested to educate employees regarding the
importance, threats, risks, and compliance procedures associated with security
rules. Research shows that training and awareness programs can be useful and
effective [12,17], especially when personalized [11]. However, this utility incurs
high financial and time costs, and the program may need periodic repetition.
Moreover, these programs provide generic prescriptive guidance and, unlike our
approach, do not include incentives for incorporating the advice in work prac-
tice [15].
So far, only a few studies have applied persuasive technology for usable secu-
rity. Forget et al. [7–9] utilized persuasive technology to raise the strength of
user-chosen text passwords, while Chiasson et al. [3,4] achieved similar success
in helping users choose strong graphical passwords. Other researchers have used
persuasive technology to implement games targeting security related behavior.
Yeo et al. [19] demonstrated the effectiveness of a Web based multiple-choice quiz
to improve password management and virus protection, and Sheng et al. [14]
found that an anti-phishing game was effective in educating users about secure
Internet behavior. However, all of these studies utilized relatively small student
samples, not knowledge workers. Additionally, these explorations were limited
to specific actions, such as choosing passwords, rather than considering all work
practices that impact the security of an organization’s information resources. We
tackle this broader challenge and explore the effectiveness of several individual
persuasive strategies and their combinations.
3 Design Probe of Persuasive Features
To explore persuasive technology for organizational information security, we built

a Web based design probe in the form of an interactive front-end interface based on
several persuasive strategies. We treated compliance with the organizational secu-
rity policy as the target behavior; the security policy defines permissible actions,
thus setting the desired direction for a change in security-related behavior.
We selected 8 persuasive strategies from the collection of 28 described by
Oinas-Kukkonen and Harjumaa [13]. Our selection was based on a combination
of factors, viz., applicability to organizational information security, application
342 M. Busch et al.
in past usable security studies, amenability to implementation as a concrete and

distinct interface mechanism, and the necessity to limit strategies to a reasonable
number. Specifically, the probe covered the following eight persuasive strategies:
– Rewards: Rewarding the target behavior (with Security Points and Badges).
– Tailoring: Personalizing to one’s interests and personality (via a
questionnaire to determine features of potential interest to the employee).
– Competition: Promoting competition with others (through Challenges).
– Simulation: Providing a means for understanding the connection between
behavior and its consequences (by communicating potential risks).
– Social comparison: Comparing one’s performance with others (using Statistics
of past security behavior).
– Suggestions: Recommending appropriate behavior at opportune moments (by
suggesting interesting features determined using questionnaire responses).
– (Social) learning: Facilitating learning about target behavior (in the form of
a Security Quiz with questions about the information security policy).
– Self monitoring: Providing a means to track one’s performance and status
(using Statistics of past security behavior).
We term an operational system implementation of one or more persuasive strate-
gies as a persuasive feature. Our probe incorporated the above eight persuasive
strategies in the form of six persuasive features, viz., Security Points, Secu-
rity Quiz, Challenges, Statistics, Personalization, and Risk Communication. The
feature design followed Oinas-Kukkonen and Harjumaa’s [13] suggested process
model for designing persuasive technologies. Table 1 shows the mapping between
the strategies and the features. With the exception of Security Points, each fea-
ture operationalized a primary and a secondary strategy. We summarize each of
these features below. Busch et al. [2] provide further details.
Table 1. Mapping between probe features and persuasive strategies.
Probe feature Primary strategy Secondary strategy

Security points Rewards –
Security quiz (Social) learning Rewards
Challenges Competition Rewards
Statistics Self monitoring Social comparison
Personalization Tailoring Suggestion
Risk communication Simulation Rewards
Security Points: Users could collect virtual rewards in the form of Security
Points (see Fig. 1a). Security Points could be earned by taking a Security Quiz,
completing Challenges, or answering the Personalization questionnaire. Users
were awarded Badges corresponding to the progressive accumulation of Security
Points, viz., Beginner, Intermediate, Expert, Professional, and Master. Security
(a) Security Points (b) Security Quiz
(c) Challenges (d) Statistics
(e) Personalization (f) Risk Communication
Fig. 1. Screenshots of the six persuasive features implemented in the probe.
Points could be used to ‘buy’ perks, such as time to use social media and colors
to change the look-and-feel of the interface. Points were deducted if the user’s
actions were deemed insecure.
Security Quiz. In order to facilitate learning, users were presented with quizzes
on work scenarios related to the information security policy (see Fig. 1b).
Challenges. This feature provided motivation in the form of competition among
employees (see Fig. 1c). Users could accept Challenges that were either group
based (e.g., “Behave more securely than your colleagues for one week.”) or indi-
vidual (e.g., “Comply fully with the security policy for one week.”).
Statistics. The probe showed Statistics of security-related behavior, e.g., the
number of policy violations committed by the user per week as well as the average
number of violations for all employees across the organization (see Fig. 1d).
344 M. Busch et al.
Personalization. A user could answer a questionnaire that determined his or

her persuadability for each of the six persuasive features (see Fig. 1e). The Per-
sonalization questionnaire consisted of statements related to the persuasive fea-
tures (e.g., “I like to compete against others” related to the Challenges feature)
rated on a Likert-type scale.
Risk Communication. When a user might be engaging in risky information
security practices, the probe warned the user of the risk for the organization as
well as personal consequences for the user (simulation).
The first five persuasive features were included under a corresponding tab
in the probe. Since Risk Communication involved operating system dialogs, it
functioned outside the probe’s main interface (Fig. 1).
4 Method
We employed the design probe in a user study to evaluate the perceived per-
suasiveness of each of the implemented persuasive features. The study was con-
ducted in two modes – online and laboratory – with different participants. Both
modes followed the same procedure, except that laboratory participants were
asked to think aloud when performing study tasks. As the framing for the study,
participants read the following scenario: You are waiting at the airport to embark
on a business trip. While waiting, you wish to prepare a business document for
which you need sensitive information from your company’s servers. The security
policy of your employer states that you should access sensitive files only from
encrypted (secure) network connections. The wireless Internet at the airport is
unencrypted.
Participants were asked to imagine themselves in the scenario, open our
probe, and explore and interact with each feature one at a time by clicking
the corresponding tab. The features were presented in random order by random-
izing the tab sequence, with the exception of Risk Communication which was
excluded from randomization because it operated outside the probe interface; it
was always the final task. To evaluate Risk Communication, participants were
asked to open another application called ‘File Explorer.’ Within this applica-
tion, participants were instructed to (try to) open ‘Low-Sensitive Document.pdf,’
‘Medium-Sensitive Document.pdf,’ and ‘High-Sensitive Document.pdf.’ Clicking
on the ‘Low-Sensitive Document.pdf’ brought up a warning regarding a security
policy violation (see Fig. 1f). Given the low sensitivity of the file, the warning
allowed the user to proceed if he or she desired. The warning popup for the other
two files blocked opening the file with no user override.
4.1 Influence of Persuasive Features

After encountering each persuasive feature, participants rated it using the follow-
ing items in random order on a 7-point Likert-type scale from Strongly disagree
(1) to Strongly agree (7):
– Usefulness: I find this feature useful.
– Enjoyment: I enjoy using this feature.

– Increase in Awareness: This feature would make me more aware of
organizational information security.
– Attitude Change: This feature would have a positive influence on my attitude
toward organizational information security.
– Behavior Change: This feature would lead me to comply with organizational
information security.
The items measured the perceived persuasiveness of each feature and were
adapted from Drozd et al. [6] and Venkatesh and Bala [18]. We further allowed
participants to provide open-ended opinions regarding the feature.
Recruiting and Sample. Participants were recruited from a database of

research volunteers from Austria. We screened potential online participants to
include only those who were employed full- or part-time. For the laboratory
mode, we selected participants on the basis of the relevance of information secu-
rity policies for their jobs and recruited them by telephone. Nearly all partic-
ipants worked in professions such as consulting, management, administration,
programming, marketing, etc., requiring the handling of digital information, As
participation incentive, online participants were entered into a raffle for 20 gift
certificates of e30 for Amazon.com, while each laboratory participant received
a gift certificate of e30 redeemable at a variety of brick-and-mortar stores.
5 Findings
Of the 81 online participants, we retained the responses of the 64 who met our
screening criterion of the presence of explicit information security policies in their
organizations. Of these, 48 (75 %) were employed full-time and 16 (25 %) part-
time (33 females / 31 males) with ages ranging from 21 to 60 (median = 32).1
A further 15 participants completed the study in the laboratory (12 employed
full-time and 3 part-time; 8 females and 7 males) ranging in age from 27 to 43
(median = 36).
The lowest value of Cronbach’s alpha for the set of items for rating each
feature (see Sect. 4) was 0.94, indicating excellent internal consistency. There-
fore, we combined the ratings into a persuasiveness score by averaging the five
individual ratings. Figure 2 shows notched box plots of the persuasiveness scores
for each feature (higher values indicate greater persuasiveness). The notches in
the boxes indicate the 95 percent confidence interval of the median. The line at
4 on the y-axis marks the neutral mid-point of the 7-point Likert-type scale. We
found that Risk Communication, Statistics, and Security Quiz were rated more
persuasive compared to Security Points, Challenges, and Personalization.
We consulted the open-ended answers of the online and laboratory partici-
pants to the question “What is your opinion about this feature?” The first author
1
Four participants declined to provide their age.
346 M. Busch et al.
Fig. 2. Persuasiveness rating for each feature (1 = Strongly Disagree to 7 = Strongly

Agree).
and a researcher not connected with the study independently read all responses
and used open coding [16] to generate potential descriptors. Upon comparison
and discussion, the two independent sets of descriptors were consolidated into
a single set of 14 characteristics describing the features: instructive, easy to
understand, useful, playful, motivating, informative, innovative, efficient, pleas-
ing, context appropriate, appealing, competition inducing, privacy protecting, and
detailed.
The coding further indicated whether the answer expressed positive or neg-
ative views regarding the particular characteristic since this was identified as
an important dimension during open coding. For almost every feature, opinions
were split. For example, playfulness was seen positively (enjoyable) or negatively
(childish). Statistics received the greatest proportion of positive remarks (64.8 %)
and Security Points the most negative (65.1 %). For a nuanced understanding,
we examined salient comments. On the positive side, many participants found
each feature to be understandable and useful. Several participants expressed
positive opinions about the ability of the Security Quiz, Statistics, and Risk
Communication features to raise security awareness and promote learning. On
the negative side, a roughly equivalent number of participants deemed the fea-
tures useless, unhelpful, or confusing. Several commented that the playfulness
of features like Security Points was childish and unprofessional. (“This is like a
game. This will not be appropriate in my company. Employees should do impor-
tant stuff.”). Participants found Risk Communication disruptive to the ongoing
task. Further, several expressed displeasure that they were allowed to dismiss
the warning. (“It is weird that I can open a ‘Low sensitive’ document although
it’s against the security policy.”). Instead, some preferred being denied access
altogether. (“If I am not allowed to open it, it should not open.”). Further, partic-
ipants felt that the competition introduced by Challenges could have a negative
effect on relationships with colleagues. (“To be better than others means show-
ing that someone else is worse.”). Other negative responses were an artifact of
the prototype nature of the probe. For instance, participants complained that
the Statistics were not sufficiently detailed and nuanced and the purpose, and
subsequent use, of the Personalization questionnaire was unclear.
Verbalized thoughts of the laboratory participants while interacting with the
probe confirmed that the various annoyances and shortcomings arose chiefly
due to the missing detail and polish of the probe, rather than the underlying
persuasive strategies. For instance, laboratory participants expressed desires for
more practical benefits for Security Points, such as meals, breaks, vacation, etc.
(“Not an incentive for me. Better: new tablet, free meals in the cafeteria.”).
Similar to typed open-ended responses, the verbalized thoughts included demand
for more detail. For instance, participants wished that the Statistics included not
only the number but also the severity of security breaches. They also alluded
to various opportunities for pertinent system feedback, such as the extent of
correctness of each option in the Security Quiz and the secure way to deal with
the Risk Communication warning.
6 Discussion and Implications

It was beneficial to conduct the study in two modes. Although the number
of laboratory participants was small, having access to their thinking as they
interacted with the probe was valuable for understanding the reactions to the
features and illuminating the underlying reasons. It also helped confirm that
participants understood the scenario and were immersed in it.
Security is secondary to the task at hand, often obstructing the ongoing
activity. Users are inclined to view security requirements with frustration rather
than enjoyment. While the neutral ratings for some of our persuasive features
are reflective of this fact, the overall comments referring to user experience and
adoption illustrate the general positive reception for the persuasive features, sug-
gesting that the underlying strategies could contribute to reducing the annoy-
ance of an otherwise unenjoyable task. Our results thus indicate the promise of
persuasive security for helping employees understand and follow work practices
compliant with the organizational security policy.
At the same time, results also show that individuals could react to a par-
ticular feature in opposite ways. It might be impossible to resolve certain fun-
damental divergences, such as whether one views playfulness as enjoyable or
unprofessional. Ensuring coverage across a diversity of views may require com-
bining several features and strategies, instead of relying on a single approach.
Appropriate personalization could also mitigate the impact of differences by tai-
loring system operation to individual preferences.
The probe merely served as the means to achieving our goal of evaluating the
individual persuasive features. Rather than implementation as a separate system,
such as our probe, we envision and advocate that the features be appropriately
integrated within existing security applications (e.g., virus scanners, password
change systems, access control mechanisms, etc.) and integrated across all sys-
tems that comprise the organizational security infrastructure. In this regard, our
348 M. Busch et al.
findings point to specific design implications for each of the persuasive features
we evaluated:
Security Points. It is important to take into account whether playful and
game-like elements such as Security Points and Badges are suitable in an organi-
zational context where they might run counter to expected professionalism and
seriousness. Moreover, rewards like Security Points should be designed to pro-
vide tangible value. The utility of such features could be significantly increased if
users could convert them from a virtual state to a personally meaningful incen-
tive, such as meals, gift certificates, etc.
Security Quiz. We found that it is essential not only to reveal the correct
answer to the quiz but also to explain the rationale behind how it was obtained.
In the case of multiple possible answers, the quiz ought to support and explain
relative rankings of the choices. Toward this end, it might be helpful to adminis-
ter quizzes at regular intervals (e.g., ‘Quiz of the Day’) as well as at contextually
appropriate times (e.g., when the user engages in an insecure action).
Challenges. While competition caters to the inner human desire for rivalry and
accomplishment, it might have a negative impact on work dynamics and colle-
giality among colleagues. Some participants in our study suggested including
Challenges based on cooperation. Cooperative security tasks could help employ-
ees learn from each other and foster a positive group dynamic and team spirit.
Statistics. The Statistics feature was well-liked. We believe that the appeal
stemmed from the usefulness of the information for comparing one’s practices
with the larger picture as well as the visual presentation that made it easy to
comprehend. At the same time, findings indicated that the feature needs to incor-
porate a large range of possible factors to visualize and to support modifying
the visualizations to reveal important nuance. The former could be implemented
by incorporating a diversity of aspects covered by the information security pol-
icy, such as security updates, VPN, etc. To add nuance, the interface ought to
facilitate examination of relative as well as absolute measures. It can be bene-
ficial to allow additional sensemaking operations such as classifying, grouping,
sorting, and filtering the visualized information based on relevant aspects, such
as importance, severity, recency, etc.
Personalization. Participants found it difficult to understand the Personal-
ization feature and its connection with information security. These difficulties
appeared to be driven largely by the one-time nature of the study and the non-
functioning nature of this feature in the probe. Nonetheless, we obtained useful
insight from participant complaints regarding the length of the questionnaire
and the lack of clarity regarding how the answers were used. For instance, a
personalization questionnaire in a deployed system could be administered one
question at a time and completed over a longer period. Interaction data regard-
ing employee use of devices and information resources could further augment the
quality and the nature of personalization.
Risk Communication. While participants appreciated the contextual nature
of Risk Communication, they complained about disruption and obtrusiveness.
Moreover, they were frustrated that it did not offer guidance on achieving the
task securely. These findings illustrate that effective Risk Communication needs
to balance a variety of tensions, such as whether to allow a policy violation to
avoid interrupting the task and blocking user action. Regardless of the approach,
whenever Risk Communication is necessary, it is an opportunity for contextual
security training via effective information presentation.
7 Limitations
There were interdependencies among the persuasive features. For instance, Secu-
rity Points were incorporated in the Security Quiz, Challenges, Personalization,
and Risk Communication features. These interdependencies might have led to
overlapping effects. Additionally, randomization to avoid order effects may have
resulted in somewhat unnatural sequencing, thus hampering full understanding
of a feature. Risk Communication may have been affected by an order effect
because it was encountered at the end. We examined a single usage instance
using an interface probe lacking back-end functionality. Moreover, participants
had no prior exposure to the probe. A longitudinal study of user behavior within
a functioning system is needed to verify that our self-reported intentions extend
to actual user practices and to examine how they are affected by usage experi-
ence and learning. Given our self-selected and geographically-restricted sample,
replication is needed to verify generalizability.
8 Conclusion
Our goal was to investigate if persuasive features could raise employee awareness
of their organization’s information security policy and promote policy compli-
ant behavior. To achieve this objective, we applied eight persuasive strategies
to implement six persuasive features in a design probe. Online and laboratory
evaluation suggests that the features hold promise but could be enhanced by
additional feedback and concretely beneficial incentives. A functioning real-life
deployment is needed to study the longitudinal impact of the persuasive fea-
tures on contextual behavior. We hope that these findings spur explorations to
investigate how persuasive strategies could be employed to develop additional
persuasive security features.
Acknowledgments. We thank the study participants. We acknowledge Elke Matth-

eiss for help in coding open-ended responses. This work was partially funded by the
European Union Seventh Framework Programme (FP7/2007-2013) under grant agree-
ment 318508 (MUSES: Multiplatform USable Endpoint Security). Dr. Patil’s partici-
pation in the research was funded by the Helsinki Institute of Information Technology
(HIIT) Visitor Programme.
350 M. Busch et al.
References
1. Bulgurcu, B., Cavusoglu, H., Benbasat, I.: Information security policy compliance:
An empirical study of rationality-based beliefs and information security awareness.
MIS Q. 34(3), 523–548 (2010)
2. Busch, M., Wolkerstorfer, P., Hochleitner, C., Regal, G., Tscheligi, M.: Design-
ing a persuasive application to improve organizational information security policy
awareness, attitudes and behavior. In: Poster Abstract, 10th Symposium on Usable
Privacy and Security, SOUPS 2014 (2014)
3. Chiasson, S., Forget, A., Biddle, R., van Oorschot, P.C.: Influencing users towards
better passwords: Persuasive cued click-points. In: Proceedings of the 22nd British
HCI Group Annual Conference, BCS-HCI 2008, pp. 121–130, British Computer
Society, Swinton, UK (2008)
4. Chiasson, S., Stobert, E., Forget, A., Biddle, R., Van Oorschot, P.: Persuasive
cued click-points: Design, implementation, and evaluation of a knowledge-based
authentication mechanism. IEEE Trans. Dependable Secur. Comput. 9(2), 222–
235 (2012)
5. Department for Business, Innovation & Skills: Information security breaches survey
(2013). http://www.pwc.co.uk/assets/pdf/cyber-security-2013-exec-summary.pdf
6. Drozd, F., Lehto, T., Oinas-Kukkonen, H.: Exploring perceived persuasiveness of
a behavior change support system: a structural model. In: Bang, M., Ragnemalm,
E.L. (eds.) PERSUASIVE 2012. LNCS, vol. 7284, pp. 157–168. Springer, Heidel-
berg (2012)
7. Forget, A., Chiasson, S., Biddle, R.: Persuasion as education for computer secu-
rity. In: Bastiaens, T., Carliner, S. (eds.) Proceedings of World Conference on
E-Learning in Corporate, Government, Healthcare, and Higher Education 2007,
pp. 822–829. AACE, Quebec City, Canada, October 2007
8. Forget, A., Chiasson, S., Biddle, R.: Lessons from brain age on persuasion for com-
puter security. In: CHI 2009 Extended Abstracts on Human Factors in Computing
Systems, CHI EA 2009, pp. 4435–4440. ACM, NY, USA, New York (2009)
9. Forget, A., Chiasson, S., van Oorschot, P.C., Biddle, R.: Persuasion for stronger
passwords: Motivation and pilot study. In: Oinas-Kukkonen, H., Hasle, P., Har-
jumaa, M., Segerståhl, K., Øhrstrøm, P. (eds.) PERSUASIVE 2008. LNCS, vol.
10. Khan, B., Alghathbar, K.S., Khan, M.K.: Information security awareness cam-
paign: An alternate approach. In: Kim, T., Adeli, H., Robles, R.J., Balitanas, M.
(eds.) ISA 2011. CCIS, vol. 200, pp. 1–10. Springer, Heidelberg (2011)
11. Mangold, L.V.: Using ontologies for adaptive information security training. In: Pro-
ceedings of the 2012 Seventh International Conference on Availability, Reliability
and Security, ARES 2012, pp. 522–524. IEEE Computer Society (2012)
12. Merhi, M.I., Midha, V.: The impact of training and social norms on information
security compliance: A pilot study. In: International Conference on Information
Systems: Research in Progress, ICIS 2012 (2012)
13. Oinas-Kukkonen, H., Harjumaa, M.: A systematic framework for designing and
evaluating persuasive systems. In: Oinas-Kukkonen, H., Hasle, P., Harjumaa, M.,
Segerståhl, K., Øhrstrøm, P. (eds.) PERSUASIVE 2008. LNCS, vol. 5033, pp.
14. Sheng, S., Magnien, B., Kumaraguru, P., Acquisti, A., Cranor, L.F., Hong, J.,
Nunge, E.: Anti-phishing Phil: The design and evaluation of a game that teaches
people not to fall for phish. In: Proceedings of the 3rd Symposium on Usable
Privacy and Security, SOUPS 2007, pp. 88–99. ACM, NY, USA, New York (2007)
15. Siponen, M.T.: A conceptual foundation for organizational information security

awareness. Inf. Manage. Comput. Secur. 8(1), 31–41 (2000)
16. Strauss, A., Corbin, J.: Open coding. Basics Qual. Res.: Grounded Theor. Proc.
Tech. 2, 101–121 (1990)
17. Thomson, M.E., von Solms, R.: Information security awareness: Educating your
users effectively. Inf. Manage. Comput. Secur. 6(4), 167–173 (1998)
18. Venkatesh, V., Bala, H.: Technology acceptance model 3 and a research agenda on
interventions. Decis. Sci. 39(2), 273–315 (2008)
19. Yeo, A.C., Rahim, M.M., Ren, Y.Y.: Use of persuasive technology to change end
users’ IT security aware behavior: a pilot study. Int. J. Psychol. Behav. Sci. 1(1),
48–54 (2009)
Lock Up the Lighter: Experience Prototyping
of a Lively Reflective Design for Smoking
Habit Control
Kenny K.N. Chow(&)
School of Design, The Hong Kong Polytechnic University,

Hung Hom, Hong Kong
[email protected]
Abstract. Mobile apps have been employed in assisting people to control

various addictive habits including smoking. These apps track users’ behavior
mainly via self-reporting, which demands users’ patience in making explicit
input. This paper presents Lock Up, which requires users to lock their lighters in
a smart case connected to an app. It heuristically estimates the continuous time
of a user without smoking and metaphorically overshadows the lighter by
foregrounding the smoking consequence due to the accessibility of the lighter.
After reviewing major behavior models, we extend the notion of lively inter-
active systems grounded in cognitive science and phenomenology, and intro-
duce a design framework promoting behavior control via provoking imagination
and reflection. An experience prototype of Lock Up has been built for partici-
pants to use in the laboratory environment. The evaluation is exploratory and
generative, providing qualitative data to enrich the proposed user experiences
and informing orientation toward the next iteration.
Keywords: Reflective design Liveliness Embodied interaction Conceptual

blending Imagination Behavior change
1 Introduction
When searching Apple’s App Store or Google Play with the keyword ‘smoking’, one
would find many mobile apps aiming to assist users to control their smoking behavior
or even to quit smoking. These apps usually track how much time has passed since
their users smoked last time, or how many cigarettes their users has smoked each day,
by requiring one to self-report. Users have to remember to launch the apps and input
whenever they smoke. This is demanding even to self-motivated users. Although these
apps usually provide the users’ progress statistics (e.g., how much money saved, how
many days of life regained based on scientific estimation, etc.) as a hopeful encour-
agement, the factual data needs sparks for highlighting the immediate pleasure or pain
and the contingent hope or fear. We believe incorporating reflective design [1] and the
liveliness framework [2] can create those sparks.
Lock Up is a reflective, lively interactive system aiming to assist users in con-
trolling their smoking habit via provoking imagination, emotion, and so reflection in
them. The initial idea originated from a design student (Ms. Lui Yan Yan, School of
DOI: 10.1007/978-3-319-31510-2_30
Lock Up the Lighter: Experience Prototyping 353
Design, The Hong Kong Polytechnic University)’s graduation project, which we have
further developed into an experience prototype based on the design framework of
liveliness. Grounded in cognitive semantics in cognitive science, the psychology of
emotion, and phenomenological approaches to interaction design, the notion of live-
liness suggests drawing on analogies with everyday life experience in designing
interactive systems in order to provoke user imagination and emotion at multiple
cognitive levels. Users immediately understand how to operate the systems at initial
moments of use, and then make metaphorical, reflective meaning during the later
contingent moments. The two levels of meaning construction enabled by liveliness can
be employed to create sparks for highlighting the motivators for behavior change.
Lock Up requires users to put their lighters in a smart case and lock it up. This is a
heuristic way to estimate how much time a user has passed without smoking. After
closing the hinged cover tightly, the user sees the LED (light-emitting diode) lights on
the case moving fast back and forth (Fig. 1). The mobile app then shows a wireless
connection set up with the case. To create a spark for highlighting the immediate pain
as a motivator, the user is required to blow at the smartphone for a lung capacity test
(Fig. 2). The app responds with a default result: a skeleton graphic and a message
saying that the user has been seriously infected (Fig. 3a). The skeleton, together with
the fast moving lights on the case, seems to signal emergency and call for action. After
spending some time on sports activities, the user gets improvement in the lung capacity
test, from the skeleton, to a zombie (Fig. 3b), then a half-zombie-half-human (Fig. 3c),
a healthy human (Fig. 3d), and so on. An immediate understanding comes about in the
user’s mind that locking up the lighter and doing workout make progress, which also
project a hopeful future.
Yet, the future is contingent. The user is given a leeway to get around the strict
smoking control. One can unwrap the case and take out the lighter anytime, but the
lung capacity test result then quickly drops back to the skeleton, even though the user
keeps doing sports. During these contingent moments, the unexpected change in health
status makes the user wonder. The user may invoke different interpretive frames, for
instance the infection frame, to make sense of the contingency. The lighter is the source
of infection. This metaphorical, reflective meaning is another spark that highlights the
Fig. 1. Lock Up requires the user to put the lighter into the case, and the lights on the case start
moving.
354 K.K.N. Chow
Fig. 2. Lock Up allows the user to blow at the smartphone for the lung capacity test.
Fig. 3. a-d. The default result of the first test is the skeleton, followed by the zombie, the
half-zombie-half-human, the healthy human, and so on.
contingent fear (or hope) in the future. The user may see the lighter in a very different
way and want to lock it up again.
The design of Lock Up demonstrates an application of the liveliness framework in
reflective design for behavior control. During initial moments of use, the system’s
sensory feedback is a spark highlighting the immediate pain or pleasure, resulting in
immediate understanding of the action. During later contingent moments, the system’s
unexpected changes are another spark foregrounding the contingent fear or hope in the
future of which the user makes a metaphorical, reflective meaning that ‘the lighter is
evil’. An experience prototype has been built for participants to use in the laboratory.
This evaluation process is more exploratory and generative than validating. The design
framework provides guidelines for designers and researchers to propose possible,
intended meaning in terms of user imagination, emotion, and reflection, while the
evaluation assists in finding real and particular instances in order to support and enrich
the user experience narratives predicted via the framework. This paper first articulates
the theoretical framework, followed by the research design, experience prototyping,
and discussion of the findings. The conclusion is the design implication for next
iteration of prototypes to be deployed in the wild.
2 Theoretical Framework
2.1 Behavior Models
Fogg Behavior Model (FBM). The design of Lock Up refers to a few influential
models of human behavior. B. J. Fogg [3] introduces a behavior model for designing
persuasive systems. The three determinants for a person to perform a behavior include
motivation, ability, and triggers. Motivation includes immediate results (i.e., pleasure
vs. pain), foreseeable outcomes (i.e., hope vs. fear), and social conforming. Ability is
related to perceived probability of success, which can be improved through empowering
the users. With sufficient motivation and ability, people become determined but still
need triggers for taking action. Fogg introduces three types of triggers that are applicable
to the design of Lock Up. Spark is used to foreground the relevant motivators to the user
(e.g., using iconic representation to show immediate status and assist in projecting
foreseeable development). Facilitator is to make the behavior easier or harder to do (e.g.,
locking up the lighter). Signal is to provide timely and relevant reminder.
While motivation and ability are dependent on many internal and external factors that
are less controllable to designers, triggers have become more accessible means to
designers with today’s ubiquitous technologies. FBM’s triggers are an important
consideration in designing persuasive systems, especially when different stages of
behavior change are concerned.
Five Stages of Change. James O. Prochaska and his colleagues [4] summarize
research on self-initiated and professionally assisted change of addictive behaviors in
five stages. At the pre-contemplation stage, the individual does not see the problem and
so has no intention to change. At the contemplation stage, the individual is aware of the
problem and starts seriously thinking of overcoming it, without taking any action yet.
At the preparation stage, one is about to take action. At the action stage, the modifi-
cation is visible and even recognized by others. Finally the maintenance stage is about
whether the change can endure for more than six months without relapse, which is
generally regarded as a successful indicator. Based on this model, designing for
behavior change needs overview of the whole journey but also identification of a
focusing stage. The pre-contemplation stage requires persuasively informing people of
the problem. Spark can be added to bring hope or fear to the people (e.g., showing
exaggerated graphic images of current status). The contemplation and preparation
stages have to enable and even empower the people. Using facilitators to make the
behavior easier or harder to do is an effective intervention (e.g., hiding the lighter from
one’s sight). At the action and maintenance stages, the design has to provide reminder
or progress report (e.g., showing incremental changes of the status).
In summary, Lock Up has the following features based on the above models:
• Focusing on the contemplation and preparation stages, Lock Up helps the user
temporarily forget smoking by hiding the lighter from the user’s sight. It makes the
key to smoking less accessible.
356 K.K.N. Chow
• To foreground the relevant motivators, Lock Up shows the user’s lung capacity in
terms of iconic images like skeleton and zombie. Subsequent images assist the user
in projecting foreseeable improvement, or conversely bring fear to those still in the
pre-contemplation stage.
• Lock Up shows incremental signs of successful performance from skeleton to
healthy human and even superhero to strengthen the user’s determination. It is
important for those in the action and maintenance stages.
2.2 Lively Interactive Artifacts for Behavior Control

Lock Up demonstrates how lively interactive artifacts can be used for promoting
behavior change via provoking reflection in users. Designing lively artifacts [5] usually
draws on embodied interaction [6] and conceptual metaphors [7, 8], making the
interactions reminiscent of users’ past experiences in everyday life and stimulating
meaning construction at multiple cognitive levels during different moments of use.
During initial moments of use, the user acts upon the artifact and perceive quick
feedback. The interaction between the user and the lively artifact, represented by two
arrows in Don Norman’s conceptual model diagram [9], is so instantaneous that turns
into a loop, resulting in immediate understanding of the operation with a sense of
control (Fig. 4a). The operation is similar, but not exactly, to a common experience,
triggering an immediate blend of the current experience and the past. Blending, a
theory introduced by Gilles Fauconnier and Mark Turner [10], is a pervasive cognitive
operation that integrates concepts and outputs new one. At this immediate level, the
output is an embodied imagination between familiar (past experience) and unfamiliar
(new experience).
sensory
perception
lively
interactive user
system
motor
action
Fig. 4a. A sensorimotor feedback loop between a user and a lively interactive system during
initial moments of use, giving the user a sense of control.
During later contingent moments of use, the lively artifact shows unexpected
changes, bringing the user a sense of wonder (Fig. 4b). The user then invokes an
interpretive frame [11] to make sense of and account for the changes. The frame comes
with a past scenario in the user’s memory. The user analogically maps the current
experience of the changes with the past scenario, and elaborates a metaphorical blend,
resulting in an imaginative scenario with reflective meaning.
sensory ?
perception
lively
interactive user
system
motor
action
Fig. 4b. Unexpected changes seemingly unrelated to user input during contingent moments,
bringing the user a sense of wonder.
We illustrate the above cognitive processes based on Fauconnier and Turner’s

integration diagrams [10]. An integration diagram consists of circles representing
mental spaces, each of which contains conceptual elements of a scenario, such as
actors, objects, actions, or relations. The two horizontal spaces are input for the blend,
while the one below is the output. The horizontal solid lines between the two input
spaces are links connecting the counterparts respectively. These outer-space links are
compressed into inner-space relations inside the blend. Other elements are only
selectively projected from either input to the blend. Figure 5a shows a possible
immediate blend. What we add to the integration diagram includes the sensorimotor
feedback loop, which envelops motor action (e.g., blow at the phone, lock the lighter
up, etc.) and sensory perception (e.g., see the skeleton and the warning, see fast moving
lights, etc.) in the mental space. The left input is the current experience enabled by a
lively interactive object featuring a loop mobilized by one’s impulsive desire and
automatic emotional appraisal, which is analogical to a past experience (e.g., walk to
security check with a buzz) denoted by the right input. The result is an immediate blend
(e.g., “I’m forced to put it away”).
Fig. 5a. A possible immediate blend triggered by Lock Up at initial moments

358 K.K.N. Chow
Figure 5b illustrates a possible blend during the later contingent moments. The loop
is fading in the user’s mind, but unexpected changes (e.g., see the skeleton again
despite exercising) take place in the lively interactive object, which prompts the user to
invoke an interpretation frame (e.g., infection), and a metaphorical blend results. This
gives rise to an imagined narrative (e.g., the lighter is infectious) and elicits emotions
(e.g., fear).
Fig. 5b. A possible metaphorical blend taking place at Lock Up’s contingent moments
The two levels of meaning making are built on the sense of control and the sense of
wonder enabled by the artifact, which can be effective in assisting habit control or
promoting behavior change. Lively interactions can be designed for regulating certain
behavior (e.g., lock up the lighter to stop smoking), while still giving the user a leeway
to get around (e.g., unlock it), and so a sense of control. Yet, if the user keeps getting
around, the artifact shows unexpected changes, bringing a sense of wonder and
prompting the user to make reflective meaning. Lock Up requires the user to put the
lighter in the case and close it tight. The lights on the case then move back and forth at
a speed faster than the typical respiratory rate. The rhythm seems to signal emergency
and urge the user to do workout. If the user spends some time on workout (e.g.,
running), one will make a progress in the lung capacity test (e.g., from the zombie to
the healthy human). The user can take out the lighter from the case anytime, but his or
her lung capacity test result will then drop (e.g., from the healthy human back to the
skeleton). The user on one hand feels in control of own behavior. This is important that
one would not feel being tightly regulated. On the other hand, the user wonders what
makes the health status reverted. After knowing the culprit is the lighter, one may
invoke the infection frame and want to stay away from the lighter.
3 Experience Prototyping
Aligning the design framework of liveliness with major behavior models allows us to
predict the possible user experiences of Lock Up in terms of conceptual integration
diagrams. To substantiate the proposed blends in Figs. 5a and 5b (i.e., for those in italic
in the figures), experience prototyping has been conducted to generate real and par-
ticular instances. Experience prototyping conveys an experience, from the sensory to
the cognitive, with a design to designers or target users [12]. It can be in any kind of
representation or medium. In this project, we have built both physical and digital
artifacts and let participants experience in a simulated environment in the laboratory.
The experience prototype of Lock Up includes the lighter case with a hinged cover
and a series of LED lights on its side, together with the mobile app. When the case is
closed with something in it, the lights start to move in a line back and forth. When the case
is open, the lights move only toward the hinged cover direction for a while and then go
off. The mobile app allows the user to blow at the smartphone’s built-in microphone for
the lung capacity test, and then responds with results according to the script of the
experience test. The app can also forge a sign of wireless connection with the closed case.
The study was conducted in our usability study room with one participant (having
smoking habit) a time, including activities and semi-structured interviews. The activities
started with launching the app and asking the participant to blow at the smartphone. The
result was scripted to be the skeleton. The participant was then asked to put his or her
lighter into the case and close the hinged cover tight by wrapping it with medical tape.
The lights on the case started to move back and forth. The participant was led to a
treadmill and ran for three minutes, followed by the second lung capacity test. The result
was scripted to be the zombie. This time the participant was tempted whether he or she
would like to take the lighter out. Another three-minute run and then the third blow
followed. If the lighter had already been taken out, the result reverted to the skeleton;
otherwise, the half-zombie-half-human was shown. This concluded the activity session.
The in-depth semi-structured interview was divided into two sessions. The outline
roughly matched the timeline of the predicted user experience based on the liveliness
framework. The first session is about the initial moments, questioning about immediate
blends, impulsive desires, and automatic appraisals. After the first session, the partic-
ipant was asked to do the lung capacity test again. The result was scripted to be the
skeleton again, because the lighter was not in the case. The second session then
follows, questioning about the later contingent moments related to unexpected changes,
metaphorical blends, and extended appraisals. Table 1 summarizes the questions.
Six participants, all males, have taken the test. One of them (DJ) is at the age between
18 and 25 years. The other participants are at the age between 25 and above and having
smoking habit for more than three years. Three of them (XX, LL, and DJ) have back-
ground in Design or related disciplines, and three (CL, XR, and ZB) in Land Surveying.
The questionnaire results informed that all of them smoked at least four or five times a
day, with XR being the most frequent smoker (about 10 times a day). XX was at the
preparation stage (was focusing on sports activities to reduce smoking frequency). Others
were at the contemplation stage, being fully aware of the health problems associated with
smoking. Half of them (DJ, CL, XR) were more motivated to discontinue smoking.
360 K.K.N. Chow
Table 1. Questions asked during interviews refer to the theoretical framework

About immediate blends • When you blew the first time, what did you see?
• When you put the lighter into the case, what did
you see? What did it mean to you? Could you
recall any slice of life like this?
• When you blew after the first run, what did you
see? What about after the second run? What did it
mean to you?
• Could you recall anything similar in your life?
About impulsive desires and automatic • How did you feel after seeing zombie?
appraisals • How did you feel after seeing half-zombie?
• How did you feel when seeing the flash right after
putting lighter in?
About unexpected changes, • Do you notice any changes? Is it what you
interpretive frames, and metaphorical expected? What do you think? (If don’t know
blends why, show the app reminder)
• Could you recall any similar scenarios from your
past experience?
• How would you see the relationship between
smoking and the lighter?
• How would you like the lighter? How would you
describe it?
• What do you want to do now? What else (other
than exercising) in your mind?
• From skeleton to zombie, how would you project
your goal?
Extended appraisals • What do you think now? What did you feel about
it overall?
3.1 Qualitative Findings

The findings include qualitative data such as participants’ quotes during the interviews,
which reveal their thoughts and feelings. This paper only highlights (Table 2) some of
those used to fill in Figs. 5a and 5b.
Table 2. Significant responses from the 6 participants

About immediate blends To DJ, the flashing rate of the lights seemed like
charging or downloading.
XX, CL, and XR recalled the moments they could
not have the lighter with them (e.g., forgot to
bring, during swimming, being on plane, etc.).
About impulsive desires and automatic Putting the lighter into the case, most participants
appraisals (XX, CL, and XR) felt insecure, anxious.
CL said, “do not want to tape the lighter”, but then
conversely “felt negatively after removing the
tape.”
(Continued)
About unexpected changes, All participants said the reverted status was
interpretive frames, and metaphorical unexpected. Most of them tried to interpret with
blends logic or science, like “I smoked too much before
this test” (ZB), or “maybe exercise generates
some chemicals in my body and after some time
when there is no chemical, I feel like having a
cigarette again” (XX).
Only DJ remembered that putting the lighter in the
case was one condition.
Being reminded of the lighter, participants invoked
different interpretive frames to predict the next
move.
The infection frame:
“As it told me I was infected so maybe the lighter is
the infection, and when you put it in the box,
you may not get the infection.” (DJ)
“It doesn’t seem like my lighter anymore … it
causes me some disease, made me lose all my
effort.” (XR)
The partnership frame:
“Key and lock, the lighter is like a key to light a
cigarette.” (ZB)
The devil frame:
“Lighters and exercise may be like devil and
angel.” (LL)
Extended appraisals Satisfied:
“Overall satisfied, also realized if I don’t put lighter
inside, I am a skeleton, not living a healthy
life.” (DJ)
“If putting the lighter back, would like to try doing
something to stay away from smoking.” (CL)
Uneasy:
“Overall experience is like going to swimming pool
and putting things like money and cards in a
box. We can’t use it…” (ZB)
4 Discussion
4.1 The Immediate Level
The lights on the case are designed to move back and forth at a speed faster than the
respiratory rate to create an illusion that the lighter is resisting inside the case. This is to
anthropomorphize the lighter. Yet, this metaphor was not obvious to the participants,
who unconsciously assumed the case was a device and the lighter was a tool, and so the
362 K.K.N. Chow
light movement seemed like charging or downloading (DJ). On the other hand, half of
the participants felt tense after putting the lighter in the case. They closed the hinged
cover tightly and wrapped it by tape. Meanwhile, they saw the lights moving fast. The
frequency seemed to suggest alertness. The sensorimotor feedback loop evoked a slice
of life in them. At least two participants (XX, XR) mentioned the moments of having
no access to lighters on plane. Lighters are classified as dangerous goods and cannot be
carried. The immediate blend between putting the lighter into the case and the expe-
rience of boarding resulted in a message that the dangerous lighter should be put away
(Fig. 5a). To facilitate this blend, the lights should be made to flash faster.
After exercising and making progress in the app, all participants became increas-
ingly happy or satisfied. Participant CL even had a negative feeling when taking back
the lighter, because this made smoking possible to him again.
4.2 The Reflective Level

All participants felt shock when they saw the test results reverting back to the skeleton
in the middle of the interviews. Nearly all of them, except Participant DJ, attributed the
downgrade to the hiatus of exercising. They invoked different frames based on logic or
science to account for the unexpected change. For example, Participant ZB thought it
was because of heavy smoking just before coming to the test; Participant XX even
imagined the “good” chemicals generated from exercising had been fading away.
Only Participant DJ remembered the lighter was in the case during the workout, and
so figured out that was the reason for the skeleton result. He could even point out the
system knew how long the lighter was in the case and gave the scores. Other than
running, walking or watching movies would also do. The text messages of the app
evoked the infection frame in his mind. The lighter was the source of infection, and
putting it in the case could save him. This metaphorical blend is shown in Fig. 5b.
Other participants who were unaware of the lighter were then shown with the
instructional screen of the app they have seen earlier. They immediately realized the
key role played by the lighter. They invoked different interpretive frames. Partici-
pant XR, initially thought the lung capacity tests were not serious, turned to see the
lighter with the infection frame and feared of it. Three participants (XX, CL, ZB)
invoked the partnership frame to see the coupling of lighters and cigarettes. They
worked together like the key-and-lock relationship. They could do harm, yet either one
did not (cf. Fig. 5b). Participant ZB explicitly said lighters might remind him of
smoking. Hiding the lighter could be a good strategy for him to control the habit.
Participant LL was skeptical of the singularity of the lighter. He argued the system
could not know if he smoked by using another lighter. Yet, he still saw lighters in
general like the devil. In summary, the system has successfully cast a shadow on
lighters, from dangerous goods, infection, to evil partners. To facilitate these
metaphorical blends, the system should highlight whether the lighter is locked or not.
Designing systems assisting people to control or even change their addictive habits is a
major yet challenging mission. Behavior models, including FBM and Five Stages of
Change, inform important factors for predicting and anticipating behavior change.
Together with our design framework of liveliness, we have developed Lock Up, a
reflective lively system, which assists users to reduce or stop smoking via provoking
imagination, emotion, and reflection in them. The theoretical framework helps delin-
eate the intended and possible user perceptions, actions, and cognitive processes in
terms of conceptual integration diagrams. Experience prototyping generates real and
particular samples to substantiate the proposed narratives in terms of blends, emotions,
and interpretive frames.
Although the current empirical study consists of only six participants, useful
findings include many nuances in the perceptions of different participants. For instance,
the moving lights are supposed to arouse the sense of alertness or emergency, but not
all participants could recall similar moments in their past experiences. This suggests
stronger color (e.g., red) and higher frequency of the moving lights. Most participants
focused too much on the workout and forgot the relevance of the lighter during the
interviews. This suggests more reminders of the lighter in the mobile app. These
insights generated from the tests will be used to further develop Lock Up into the next
iteration of prototyping and testing not only in the laboratory but also in the field.
Acknowledgments. The grant GRF from Hong Kong Research Grants Council (PolyU
5412/13H).
References
1. Sengers, P., Boehner, K., David, S., Kaye, J.J.: Reflective design. In: Proceedings of 4th
Decennial Conference on Critical Computing, pp. 49–58, Aarhus, Denmark (2005)
2. Chow, K.K.N.: Animation, Embodiment, and Digital Media Human Experience of
Technological Liveliness. Palgrave Macmillan, Basingstoke (2013)
3. Fogg, B.J.: A behavior model for persuasive design. In: Persuasive 2009, Claremont,
California, USA (2009)
4. Prochaska, J.O., DiClemente, C.C., Norcross, J.C.: In search of how people change:
applications to addictive behaviors. Am. Psychol. 47, 1102–1114 (1992)
5. Chow, K.K.N., Harrell, D.F., Yan, W.K.: Designing and analyzing swing compass: a lively
interactive system provoking imagination and affect for persuasion. In: MacTavish, T.,
Basapur, S. (eds.) PERSUASIVE 2015. LNCS, vol. 9072, pp. 107–120. Springer,
Heidelberg (2015)
6. Dourish, P.: Where the Action Is: the Foundations of Embedded Interaction. MIT Press,
Cambridge (2001)
7. Lakoff, G., Johnson, M.: Metaphors We Live By. University of Chicago Press, Chicago
(2003)
8. Lakoff, G., Turner, M.: More than Cool Reason: a Field Guide to Poetic Metaphor.
University of Chicago Press, Chicago (1989)
364 K.K.N. Chow
9. Norman, D.A.: The Design of Everyday Things. Basic Books Inc., New York (2002)
10. Fauconnier, G., Turner, M.: The Way We Think: Conceptual Blending and the Mind’s
Hidden Complexities. Basic Books, New York (2002)
11. Fillmore, C.J.: Frames and the semantics of understanding. Quaderni di Semantica 6,
222–254 (1985)
12. Buchenau, M., Suri, J.F.: Experience prototyping. In: DIS 2000 Proceedings of the 3rd
Conference on Designing Interactive Systems, pp. 424–433, Brooklyn, New York (2000)
Erratum to: Self-Reflecting and Mindfulness:
Cultivating Curiosity and Decentering Situated
in Everyday Life
Ralph Vacca(&) and Christopher Hoadley
Educational Communication and Technology, New York University,

New York, USA
{ralph.vacca,tophe}@nyu.edu
Erratum to:
Chapter 8: A. Meschtscherjakov et al. (Eds.)
Persuasive Technology
DOI: 10.1007/978-3-319-31510-2_8
In the originally published version of the paper starting on page 87 of this volume, the
means and the F-values in Table 2 and the M and SD values in the first paragraph of
Sect. 5.3 were incorrect. These have been corrected.
The updated original online version for this chapter can be found at 10.1007/978-3-319-31510-2_8

A. Meschtscherjakov et al. (Eds.): PERSUASIVE 2016, LNCS 9638, p. E1, 2016.
DOI: 10.1007/978-3-319-31510-2_31
Author Index
Adaji, Ifeoma 189 Hochleitner, Christina 339

Ali, Raian 27, 288 Horsch, Corine 276
Almaliki, Malik 27
Alrobai, Amen 288 Ilhan, Ezgi 122
André, Elisabeth 315
Janowski, Kathrin 315
Bellavista, Paolo 137 Kabir, Muhammad Ashad 99
Beun, Robbert Jan 276 Kampf, Ronit 242
Bogomolov, Sergey 315 Krischkowsky, Alina 78
Bongers, Bert 251
Brinkman, Willem-Paul 276 Lancee, Jaap 276
Busch, Marc 219, 339 Leclerq, S. 230
Lee, Duckki 137
Chittaro, Luca 3 Lelli, Matteo 137
Chow, Kenny K.N. 352 Lu, Shengnan 303
Comes, Tino 176 Lugrin, Birgit 315
Cugelman, Brian 65
Maes, Pattie 211
D’Aloia, Antonello 137 Mahmud, Abdullah Al 99
de Vries, Roelof A.J. 41 Mattheiss, Elke 219
DeGiuli, G. 230 Maurer, Bernhard 78, 199
Dogangün, Aysegül 15 McAlaney, John 288
Dolata, Mateusz 176 Meschtscherjakov, Alexander 162, 199
Midden, Cees 303
Evers, Vanessa 41 Mirnig, Alexander 162
Monarca, V. 230
Farve, Niaja 211 Mubin, Omar 99
Fitrianie, Siska 276 Murray, Tylar 150
Fröhlich, Peter 219
Nucci, M. 230
Gamberini, L. 230
Gärtner, Magdalena 162, 199 Oinas-Kukkonen, Harri 327
Gouin-Vallerand, Charles 110 Orji, Rita 219
Gram-Hansen, Sandra Burri 53
Griffioen-Both, Fiemke 276 Patil, Sameer 339
Phalp, Keith 288
Hacihabiboğlu, Hüseyin 122 Pickrell, Michelle 251
Ham, Jaap 303 Privitera, A. 230
Hammer, Stephan 315
Hekler, Eric 150 Raij, Andrew 150
Helal, Sumi 137 Regal, Georg 339
Herrmanny, Katja 15 Reisinger, Michaela 219
Hoadley, Christopher 87 Rivera, Daniel E. 150
366 Author Index
Rödel, Christina 162 Vacca, Ralph 87

Ruer, Perrine 110 Vallières, Evelyne F. 110
van den Hoven, Elise 251
Schenk, Birgit 176 van Halteren, Aart 263
Schwabe, Gerhard 176 van Wissen, Arlette 263
Sener, Bahar 122 Vassileva, Julita 189
Spagnolli, A. 230 Villa, C. 230
Spruijt-Metz, Donna 150 Vinkers, Charlotte 263
Spruit, Sandor 276
Stibe, Agnis 65
Wuchse, Martin 199
Tikka, Piiastiina 327
Truong, Khiet P. 41 Zamboni, L. 230
Tscheligi, Manfred 78, 162, 199, 219, 339 Ziegler, Jürgen 15

Persuasive Technology 11th International Conference PERSUASIVE 2016 Salzburg Austria April 5-7-2016 Proceedings

Uploaded by

Copyright:

Available Formats

Persuasive Technology 11th International Conference PERSUASIVE 2016 Salzburg Austria April 5-7-2016 Proceedings

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Persuasive Technology 11th International Conference PERSUASIVE 2016 Salzburg Austria April 5-7-2016 Proceedings

Uploaded by

Copyright:

Available Formats

Alexander Meschtscherjakov

Martin Murer Manfred Tscheligi (Eds.)

ISSN 0302-9743 ISSN 1611-3349 (electronic)

Library of Congress Control Number: 2016933467

© Springer International Publishing Switzerland 2016

Printed on acid-free paper

This Springer imprint is published by Springer Nature

Persuasive Technology is a vibrant and highly interdisciplinary research ﬁeld that

April 2016 Alexander Meschtscherjakov

Demo and Showcase Chairs

Doctoral Consortium Chairs

Persuasive [Game] Design Jam Chairs

Social Media Chair

Curtis Haugtvedt Ohio State University, USA

Tailoring Web Pages for Persuasion on Prevention Topics: Message

Supporting Users in Setting Effective Goals in Activity Tracking . . . . . . . . . 15

Persuasive and Culture-Aware Feedback Acquisition . . . . . . . . . . . . . . . . . . 27

Crowd-Designed Motivation: Combining Personality and the

The EDIE Method – Towards an Approach to Collaboration-Based

Persuasive Backfiring: When Behavior Change Interventions Trigger

Captology and Technology Appropriation: Unintended Use as a Source for

Prevention and Motivation

Self-Reflecting and Mindfulness: Cultivating Curiosity and Decentering

Alcohol Behaviour Change: Lessons Learned from User Reviews

Persuasive Strategies to Improve Driving Behaviour of Elderly Drivers by a

Creating Awareness of Sleep-Wake Hours by Gamification . . . . . . . . . . . . . 122

Methods and Models

Cicero: Middleware for Developing Persuasive Mobile Applications . . . . . . . 137

Formalization of Computational Human Behavior Models for Contextual

The Persuasive Potential Questionnaire (PPQ): Challenges, Drawbacks,

Persuasive Practices: Learning from Home Security Advisory Services . . . . . 176

Persuasive Patterns in Q&A Social Networks . . . . . . . . . . . . . . . . . . . . . . . 189

Games and Gamification

Utilizing a Digital Game as a Mediatory Artifact for Social Persuasion to

A Gamified Solution to Brief Interventions for Nightlife Well-Being . . . . . . . 230

Long-Term Effects of Computerized Simulations in Protracted Conflicts:

Interventions for Behavior Change

Understanding Changes in the Motivation of Stroke Patients Undergoing

Developing a Virtual Coach for Chronic Patients: A User Study on the

Improving Adherence in Automated e-Coaching: A Case from Insomnia

Online Peer Groups as a Persuasive Tool to Combat Digital Addiction . . . . . 288

Design Strategies and Techniques

Red Radiators Versus Red Tulips: The Influence of Context on the

Investigating Politeness Strategies and Their Persuasiveness for a Robotic

RightOnTime: The Role of Timing and Unobtrusiveness in Behavior

Persuasive Information Security: Techniques to Help Employees Protect

Lock Up the Lighter: Experience Prototyping of a Lively Reflective Design

Erratum to: Self-Reflecting and Mindfulness: Cultivating Curiosity

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365

Keywords: Tailoring · Web pages · Color priming · Message framing · Gender

Individual characteristics of the message recipient often aﬀect the eﬀectiveness of

© Springer International Publishing Switzerland 2016

3.1 Design, Participants, and Hypotheses

Gain-Framed Version Loss-Framed Version

THE SOLUTION IS SIMPLE THE SOLUTION IS SIMPLE

• Attitudes towards the speciﬁc prevention solution (smoke alarms) recommended by

1. Berkovsky, S., Freyne, N.J., Oinas-Kukkonen, H.: Inﬂuencing individually: fusing

Katja Herrmanny, Jürgen Ziegler, and Aysegül Dogangün ✉

Personal Analytics, Interactive Systems Research Group,