Journal of Consumer Research, Inc.

A Metaanalysis of the Spacing Effect in Verbal Learning: Implications for Research on

Advertising Repetition and Consumer Memory
Author(s): ChrisJaniszewski, HaydenNoel, and AlanG.Sawyer
Reviewed work(s):
Source: Journal of Consumer Research, Vol. 30, No. 1 (June 2003), pp. 138-149
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/10.1086/374692 .
Accessed: 26/12/2012 13:38
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

The University of Chicago Press and Journal of Consumer Research, Inc. are collaborating with JSTOR to
digitize, preserve and extend access to Journal of Consumer Research.

http://www.jstor.org

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

Re-Inquiries
A Meta-analysis of the Spacing Effect in Verbal
Learning: Implications for Research on
Advertising Repetition and Consumer Memory
CHRIS JANISZEWSKI
HAYDEN NOEL
ALAN G. SAWYER*
The effects of repeated advertising exposures depend on the size of the interval,
or space, between ad exposures. A meta-analysis of 97 verbal learning studies
identified several stimulus characteristics and learning context factors that interact
with stimulus spacing to facilitate memory for repeated information. The majority
of the findings are consistent with the predictions of two enhanced processing
explanations of learningthe retrieval hypothesis and the reconstruction hypothesis. These two hypotheses predict that an effective repetition strategy should
encourage incidental processing during one presentation of the material and intentional processing during the other presentation of the material, but the hypotheses differ about the optimal order of these two types of processing. Thus, the
most effective repetition strategy may be a combination of spaced exposures that
alternate in terms of media that are involving (e.g., television commercials) and
less involving (e.g., billboards, product placements).

Formal research about how repetition schedules interact

with advertising content and context is sparse (for some
exceptions, see Ray and Sawyer 1971; Rethans, Swasy, and
Marks 1986; Schumann, Petty, and Clemons 1990). One
strategy for generating hypotheses when there is limited
empirical evidence on an issue is to review conceptually
related phenomena. For example, the verbal learning discipline has a long history of investigation into learning via
repetition. More specifically, the spacing literature has investigated the influence of increasing the time interval between two presentations of a stimulus on memory for different types of stimuli. Two general findings are that (1)
distributed presentation schedules are more effective than
massed presentation schedules and (2) the advantage of a
distributed presentation schedule is highly sensitive to both
the stimulus content and the presentation context. These
content and context differences have motivated a number
of theories about how people learn via repetition.
Although the verbal learning literature lacks the external

ecommendations about how to make advertising more

memorable often focus on characteristics of the ad material, competing ads, and the audience (Burke and Srull
1988; Keller 1987; Sawyer 1981; Unnava and Burnkrant
1991). Another common focus is media scheduling and the
repetition of advertising material. However, what is often
neglected is the potential for an interaction between repetition and various advertising factors. Understanding how
different types of repeated ad material are learned and remembered can provide insight into how to make advertising
more effective.

*Chris Janiszewski ([email protected]) is the Jack Faricy

Professor of Marketing, and Alan G. Sawyer ([email protected]) is
the J. C. Penney Professor of Marketing, Warrington College of Business,
University of Florida, P.O. Box 117155, Gainesville, FL 326117155.
Hayden Noel ([email protected]) is assistant professor of
marketing, Baruch College, P.O. Box B 12-240, One Bernard Baruch Way,
New York, NY 10010. The authors thank James Algina for his many helpful
comments on the analysis.

138
2003 by JOURNAL OF CONSUMER RESEARCH, Inc. Vol. 30 June 2003
All rights reserved. 0093-5301/2004/3001-0011$10.00

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

SPACING EFFECT

139

validity of advertising field studies, it can be a useful foundation for generating hypotheses about how different advertising content could differentially benefit from repeated
exposure. This premise encouraged us to perform a metaanalysis of the spacing effect. A meta-analysis should provide two benefits. First, it should motivate hypotheses about
relationships among ad content or context, ad repetition, and
memory. For example, we know that some manufacturers
use meaningful words as brand names, whereas others (e.g.,
Procter and Gamble) use meaningless words. It may be interesting to know how other literatures account for the initial
memory benefits of meaningful words and the potential
memory benefits of repeated exposure to different classes
of words.
The second benefit of the meta-analysis is the potential
insight about how people learn via repetition. In the marketing literature, the most popular explanation of learning
via repetition is encoding variability theory (cf. Singh et al.
1994; Unnava and Burnkrant 1991). Encoding variability
theory predicts that presenting a series of ads containing
slight variations on a theme (e.g., the Absolut Vodka ad
campaign) enhances memory for the ad material. The verbal
learning literature has proposed competing theories that can
account for the findings attributed to encoding variability
(Dempster 1996; Postman and Knecht 1983). These competing theories may provide further insight into the memory
processes responsible for learning via repetition.
This article is organized as follows. First, we discuss five
of the explanations that have been proposed to account for
the spacing effect. Next, we offer 10 hypotheses implied by
one or more of these theories, paying special attention to
stimulus and context features that differentiate among the
prediction of the explanations. Then we present the results
of 97 independent studies that provide 269 data points for
testing the hypotheses. Finally, we identify relationships
among processing goals, stimulus characteristics, and the
spacing of stimulus presentations and discuss how these
relationships may provide insight into how consumers form
memories of marketing information.

and treat it as a rest period, waiting for something new to

process. The attention hypothesis assumes that people use
recognition as a cue that the P2 material is not novel and
can safely be ignored. Recognition should decline as the
interval between presentations increases; hence, people
should pay more attention to P2, and spaced presentations
should result in better memory for the material.

SPACING AND LAG EFFECTS

Explanations of the Spacing Effect

Retrieval Hypothesis. The retrieval hypothesis predicts

that the spacing effect is a positive function of the difficulty
of successfully retrieving P1 at P2. When exposed to an
event, a person is automatically reminded of prior events
(Braun and Rubin 1998; Greene 1989). In the context of
spacing, this means that P2 will serve as a cue for the
involuntary retrieval of P1 and its associated context. If P1
is retrieved from long-term memory instead of from working
memory at P2, then the person has had the opportunity to
engage in retrieval practice. Retrieval practice enhances the
probability that an item can be retrieved at a subsequent
time.

Attention Hypothesis. The attention hypothesis (Hintzman 1974) attributes the poorer recall in a massed stimulus
presentation condition to people voluntarily paying less attention to P2 (presentation 2) when it occurs shortly after
P1 (presentation 1). People recognize that P2 is repetitive

Reconstruction (Accessibility) Hypothesis. The reconstruction hypothesis predicts that the spacing effect
strengthens to the degree that the stimulus is reconstructed
at P2 (Jacoby 1978). The hypothesis assumes that the act
of perception requires a construction of a representation of

Five explanations of the spacing effect are particularly

relevant to understanding memory formation: attention, rehearsal, encoding variability, retrieval, and reconstruction.
After a discussion of each explanation of the spacing effect,
we discuss a meta-analytic strategy for differentiating between the explanations. Serendipitously, these differentiating tests will also allow us to hypothesize about potentially
effective advertising practices.

Rehearsal Hypothesis. The rehearsal hypothesis predicts that a massed presentation schedule inhibits recall because it limits rehearsal of P1 in a P1-P2 presentation sequence (Rundus 1971). The hypothesis assumes short-term
memory always contains a set of rehearsal items. The probability of an item being in the rehearsal set decreases as the
time since its exposure increases. When P2 occurs shortly
after P1, it limits the rehearsal of P1 because it is more apt
to replace P1 in the rehearsal set. When P2 occurs much
later than P1, P1 has been rehearsed until it was naturally
dropped from the rehearsal set. Since memory is aided by
more rehearsal, distributed presentations should create better
memory for the material.
Encoding Variability. The encoding variability hypothesis predicts that spaced presentations enhance recall because they allow for the formation of more cue-target associations (Glenberg 1979; Melton 1970). Glenberg (1979)
posits that cues can be general (e.g., associations to the learning environment), contextual (e.g., associations to contingent
items), and descriptive (e.g., associations to the stimulus).
Increasing the amount of time between P1 and P2 creates a
greater opportunity for general, contextual, and descriptive
cues to change. To the extent that the processing of a stimulus
reinforces associations to available cues, a spaced presentation
schedule should result in more cue-target associations and a
greater likelihood that a retrieval cue will be available at test.
Retrieval cues have also been classified as semantic or structural (Challis 1993). Semantic cues can be strong or weak
associates of the target stimulus, whereas structural cues are
organizational (e.g., position in list) or contextual (e.g., learning episode, learning location, font, color) events.

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

JOURNAL OF CONSUMER RESEARCH

140

the event and it is easier to retrieve a previous representation

of a stimulus than to construct a new one. If an item is
repeated (P2) while the previous representation (P1) is still
accessible in short-term memory, then there is no need to
construct the event. If the repeated experience is delayed
and the P1 representation begins to fade, then it is necessary
to go through a reconstruction process at P2 (i.e., construct
a representation of the stimulus). The amount of reconstruction is directly related to the amount of P1 that has decayed.
To the extent reconstruction at P2 begins with some portion
of P1, that portion of P1 can act as a retrieval cue during
subsequent attempts at retrieval.

Spacing and Advertising

Differentiating the competing explanations of the spacing
effect is important because individual explanations have the
potential to make predictions that challenge common advertising wisdom about learning via repetition. Also, individual explanations can motivate hypotheses about how to
enhance the effectiveness of individual advertisements and
campaigns. For example, one of the oldest findings in memory research is that relearning is easier when the material
is presented in the same context (Ebbinghaus 1885). In contrast, both the attention and the encoding variability hypotheses predict learning will be maximized when the learning context is varied at different exposures. Similarly,
another common rule of thumb is that simple ad material
is remembered better than complex ad material. In contrast,
the retrieval and reconstruction hypotheses predict that
memory for repeated complex material will eventually surpass memory for repeated simple material. Even one of our
most fundamental learning principles, that involved learning
leads to better memory for material, is challenged by retrieval and reconstruction hypotheses predictions that the
best learning scenarios are sequences of high and low involvement processing. Thus, differentiating among these
competing explanations of memory formation has the potential to generate a number of novel hypotheses about learning ad material via repetition.
Our effort to differentiate between the competing explanations of the spacing effect involved a series of 10 tests
conducted using meta-analysis. An understanding of these
tests depends on an understanding of the coding of the content or context variables and the predictions of each individual theory. In the next section, we discuss the data used
in the meta-analysis. Then we discuss each explanations
predictions and the results on a hypothesis-by-hypothesis
basis.

ANALYSIS
Procedure
An extensive search of the PsychINFO database was performed for the years 18872000. The database was searched
using key terms: spacing effect, lag effect, distributed presentation, and spaced presentation. All referenced papers in

these published articles were included in the study, and the

more prolific authors of these published articles were contacted requesting working papers that had not been published.
The only inclusion criteria were that the studies investigated
a spacing effect and reported statistical information sufficient
to calculate an effect size. Eighty-one articles were collected
of which 61 articles reported sufficient statistical information
to calculate an effect size. Although many of these articles
investigated memory for meaningful words (e.g., Glenberg
1979; Johnston, Coots, and Flickinger 1972), the remaining
articles investigated a diverse set of people and stimuli. For
example, subject populations included children (e.g., Topinno
1993), older adults (e.g., Balota, Duchek, and Paullin 1989),
people in different mood states (e.g., Elmes, Chapman, and
Selig 1984), and bilinguals (e.g., Glanzer and Duarte 1971).
Stimulus sets included nonsense syllables (e.g., Tzeng 1973),
sentences (e.g., Challis 1993), pictures (e.g., Hintzman and
Rogers 1973), and aurally presented materials (e.g., Kraft and
Jenkins 1981).1
The 61 articles reported 484 tests of the spacing effect
when all lag tests from the same study were included. When
multiple lags within a single study were collapsed, 269 tests
of the spacing effect remained. These tests were coded for
the independent variables needed to perform the differentiating tests and then a combined effect size, combined z,
and fail-safe N were calculated. The effect size was the
product-moment correlation (r), which is the square root of
the variance explained by a given variable or combination
of variables (Rosenthal 1994). Explained variance (EV) can
be calculated from any x 2, F- or t-statistic with one degree
of freedom (EV p x 2/N; EV p F/[F + df within]; EV p
t 2/[t 2 + df ]). The correlation coefficient was chosen as the
measure of effect size because it is easy to compute from
a t- or F-statistic, insensitive to cell sample size differences
associated with between-subject tests, and easily interpreted.
It is well known that r-values are not normally distributed,
so the r-values were transformed into a Fishers Z r (e.g.,
Z r p .5[ln (1 + r) ln (1 r)]) prior to all statistical tests.
Means calculated using the Fishers Z r were transformed
back into r (e.g., r p [e 2Z 1]/[e 2Z + 1]) for reporting
purposes.
A combined z and fail-safe N were calculated for each
combined effect size (Rosenthal 1994). A combined z is
calculated by expressing t-values as one-tail p-values, expressing these p-values as z-scores from a standard normal
distribution, adding these z-scores, and then dividing by the
square root of the number of scores. We report a combined
z that is not weighted by respective sample sizes because it
is a more conservative statistic than the weighted combined
z. A fail-safe or file drawer N (e.g., N p
{[(Sz j ) 2/2.706] k} estimates the number of unpublished
studies with an effect size of zero that would have to exist
in order to render the effect insignificant at the alpha p .05
level. As expected, the meta-analysis of the spacing effect
1
A list of the studies used to perform the meta-analysis is available from
the first author.

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

SPACING EFFECT

141

(n p 269) was statistically significant (r p .339; combined

z p 36.83; fail-safe N p 148,979).
The test of the influence of the stimulus content and context factors used 484 observations, but the repeated observations associated with lags were nested within an individual
test of the spacing effect so that, in effect, the tests depended
on 269 observations. To perform these tests, we used a
Generalized Estimating Equations (GEE) procedure. The
GEE procedure is recommended when observations are correlated because they are either repeated or have a common
characteristic (Liang and Zeger 1986). In our data, we had
a lag variable that varied across experiments. The net effect
of using the GEE procedure is that spacing experiments with
more levels of lag in the design did not receive additional
weight in tests of the manipulated variables. Note that there
was no significant difference in effect size for the six different dependent measures used in the spacing studies: the
percentage correct in a recall task (n p 216), the percentage
correct in a frequency estimate (n p 13), the percentage
correct in a recognition task (n p 18), the amount recalled
(n p 8), the estimated frequency (n p 13), and the response latency (n p 1) (F(5, 263) p 1.94, p 1 .05). Thus,
we collapsed the data over the type of dependent measure
except for hypotheses that differentiated between free and
cued recall tests.

Hypothesis Tests
Ten hypothesis tests were performed. The hypotheses and
test results are summarized in table 1 and the meta-analysis
statistics and mean recall by variable are reported in table
2.

The Lag Effect. Investigations into the spacing effect

often test multiple spacing intervals in the same study. The
rehearsal, attention, encoding variability, and reconstruction
hypotheses predict a positive linear or logarithmic effect of
the amount of the lag (i.e., spacing interval). The retrieval
hypothesis predicts an inverted-U relationship of increasing
lag (Bjork 1988). Initially, as the lag increases, retrieval of
P1 becomes more difficult and subsequent memory for the
stimulus increases. At some point, the lag becomes long
enough to prohibit retrieval of P1 at P2, and subsequent
memory for the stimulus declines.
Twenty-nine articles provided 83 direct tests of multiple
spacing intervals (i.e., two to five) ranging from five seconds
to one week. Fifty-one tests showed a significant positive
influence of lag, 19 showed no influence of lag, five showed
a significant negative influence of lag, seven showed an inverted-U relationship, and one showed a U-shaped relationship. In an aggregate test, we set the intercept of all 83 tests
to have the same value at a lag of one. The effect size averaged
.298 at a lag of one, .440 at a lag of two, .543 at a lag of
three, .531 at a lag of four, and .565 at a lag of five. The tests
for a linear relationship (F(1, 478) p 49.23, p ! .01) and log
relationship (F(1, 478) p 25.69, p ! .01) were both statistically significant, but the test for the quadratic relationship
was not significant (F(1, 477) p 0.89, p 1 .10).

Verbal/Pictorial Stimuli. Explanations of the spacing

effect make different predictions for verbal and pictorial
stimuli. The rehearsal hypothesis predicts that pictures are
harder to rehearse than words; hence, pictures should benefit
less from increased rehearsal in the distributed condition
and should show a weaker spacing effect. The reconstruction
hypothesis predicts an interaction effect based on the type
of dependent measure. The reconstruction of pictorial stimuli relies on stimulus features, thus the spacing advantage
for pictorial information should be strongest when there is
a recall cue that incorporates stimulus features (i.e., cued
recall). The reconstruction of verbal information relies on
conceptual features, thus the spacing advantage for verbal
information should be strongest when there is no recall cue
(i.e., free recall). The remaining three theories make no
direct predictions.
The stimuli used in each spacing study were coded as
nonsense words, words, sentences, or pictures (see table 2
for mean memory levels and average effect sizes) with the
later three types of stimuli contributing to this analysis.
Planned contrasts showed that the effect size for the verbal
stimuli (Mword+sentence p .338) was not different from the effect size for pictorial stimuli (M p .303; x 2 (1) p 0.17,
p 1 .10), a finding inconsistent with the rehearsal hypothesis.
Predictions of the reconstruction hypothesis were investigated after splitting the studies into cued recall and free
recall groups. For cued recall, pictures (M p .430) showed
a stronger spacing effect than verbal stimuli
(Mword+sentence p .235; x 2 (1) p 4.75, p ! .05). For free recall, verbal stimuli (Mword+sentence p .397) showed a stronger
spacing effect than pictures (M p .174; x 2 (1) p 6.39,
p ! .05). These results are consistent with the reconstruction
hypothesis.
Meaningful/Meaningless Stimuli. The rehearsal hypothesis predicts that meaningful stimuli are easier to rehearse and should exhibit a stronger spacing effect. The
encoding variability hypothesis also predicts that meaningful
stimuli will exhibit a stronger spacing effect. Meaningful
and simple stimuli will have both general and contextual
associates, but meaningful stimuli will also have descriptive
associates. There will be opportunities for these descriptive
associates to vary and be reinforced at a distributed P1 and
P2; hence, there will be more potential retrieval cues for
meaningful stimuli. The retrieval hypothesis predicts that,
at P2, people attempt to retrieve their memory of P1, a
process that should be easier with meaningful stimuli.
Hence, there should be a weaker spacing effect for meaningful stimuli. The reconstruction hypothesis predicts that
people attempt to retrieve the representation formed at P1
in order to avoid construction of a representation of the
stimulus at P2. Access to P1 should be easier with meaningful stimuli; thus, there should be a weaker spacing effect.
The attention hypothesis makes no prediction.
Meaningful stimuli (M p .335) exhibited a significantly
weaker spacing effect than meaningless stimuli (M p
.509; x 2 (1) p 20.04, p ! .05). These results are consistent

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

TABLE 1
TESTING COMPETING ACCOUNTS OF THE SPACING EFFECT

Hypotheses

Encoding
Attention Rehearsal variability

1. There is a ______ relationship between lag and the magnitude of the

spacing effect.

Loga

2. The spacing effect for verbal stimuli will be ______ than for pictorial
stimuli.

Loga

Loga

Retrieval

Reconstruction

Inverted-U Loga

Level
Logb
Inverted-Ub

Cued: weakera Cued: verbal

Stronger

Cued: pictorial
Free: strongera Free: verbal

3. The spacing effect for meaningful stimuli will be ______ than for
meaningless stimuli.
4. The spacing effect for familiar stimuli will be ______ than for novel
stimuli.

Stronger

Stronger

5. The spacing effect for structurally/semantically complex stimuli will be

______ than for simple stimuli.

Stronger

Weaker

Weaker

6. The spacing effect for bi-modal stimuli will be ______ than for unimodal stimuli.

Weaker

7. The spacing effect for related cues will be ______ than for unrelated
cues.

Stronger

Stronger

Stronger
Strongera Strongera

9. The spacing effect for isolated stimuli will be ______ than for embedded stimuli.

Stronger

Stronger

Weaker

Weaker

Weakera
Weakera
Nulla

Strongera Strongera

Stronger

8. Intentional processing will result in a ______ spacing effect than incidental processing.

10. The spacing effect for structurally/semantically complex intervening

material will be ______ than for simple intervening material.

Stronger

Nulla

Weaker

Weaker

Strongera Strongera

Stronger

Stronger

Weaker
Nulla

Strongera Strongera

Weakera
Weakera
Nulla
Strongera

Weaker

Weaker
Strongera

Weaker
Nulla
Strongera

Prediction is supported by the data.

F-statistic reported.
Additional test data reported in the text.
*p ! .05.

b
c

x2
25.67*
.89

.235
.430
.397

Free: pictorial

.174

Meaningful

.335

Meaningless

.509

Familiar

.338

Novel

.424

Structurally complex

.330

Simple

.325

Semantically complex .586

Simple

.325

Bi-modal

.302

Uni-modal

.347

Related

.404

Unrelated

.297

Intentional

.352

Incidental

.236

Insolated

.333

Embedded

.371

Structurally complex

.327

Simple

.331

Semantically complex .419

Other

.330

4.75*

6.39*

20.04*
2.72c

1.56

19.95*
.51c
1.21c

9.24*

.26

2.30

5.55*

SPACING EFFECT

143
TABLE 2
INFLUENCE OF STIMULUS FACTORS ON SIZE OF SPACING EFFECT

Spacing
Stimulus form:
Nonsense word
Word
Sentence
Picture
Stimulus meaningfulness:
Meaningless
Meaningful
Stimulus familiarity:
Familiar
Novel
Stimulus complexity:
Simple
Structurally complex
Semantically complex
Stimulus variety:
Unimodal, visual
Unimodal, auditory
Bimodal
Cue relatedness:
Unrelated
Structurally related
Semantically related
Learning goal:
Intentional
Incidental
Presentation context:
Isolated
Embedded
Intervening material:
Simple
Structurally complex
Semantically complex

Number of
qualifying
cases

Combined
effect size (r)

Combined
Z

14
189
40
24

.454
.330
.378
.303

8.35
30.06
17.49
10.91

!
!
!
!

9
256

.509
.335

253
12

Fail-safe N

% recall
massed

% recall
spaced

.001
.001
.001
.001

347
62,909
4,480
1,031

.333
.405
.336
.551

.490
.513
.459
.651

6.93
36.44

! .001
! .001

151
125,378

.267
.411

.444
.518

.338
.424

37.25
5.83

! .001
! .001

129,469
139

.404
.411

.511
.575

200
57
12

.325
.330
.586

30.04
18.38
14.08

! .001
! .001
! .001

66,518
7,060
867

.406
.413
.341

.513
.512
.526

134
75
60

.290
.449
.302

21.76
26.00
16.42

! .001
! .001
! .001

23,311
18,656
5,920

.395
.391
.433

.485
.550
.532

19
4
15

.297
.455
.391

6.65
9.02
8.87

! .001
! .001
! .001

292
116
421

.455
.460
.415

.525
.665
.539

238
31

.352
.236

29.83
10.06

! .001
! .001

78,050
1,128

.406
.383

.524
.440

228
41

.333
.371

33.68
16.59

! .001
! .001

95,365
4,129

.410
.338

.521
.447

215
39
8

.331
.327
.419

32.39
16.82
4.72

! .001
! .001
! .001

8,3149
4,040
58

.397
.502
.277

.510
.587
.398

NOTE. The combined effect size (r), combined Z, p, and fail-safe N are based on listed qualifying cases. Average percentage recall scores are based on
qualifying cases from 216 studies using recall as a dependent measure.

with the retrieval and reconstruction hypotheses, but not

with the rehearsal and encoding variability hypotheses.

Familiar/Novel Stimuli. The attention, rehearsal, and

encoding variability hypotheses predict a stronger spacing
effect for familiar stimuli. With respect to the attention hypothesis, novel stimuli are known to be better at capturing
attention than familiar stimuli (Berlyne 1970). To the extent
that a novel stimulus can better sustain attention at P2 in the
massed condition, the attention hypothesis suggests a smaller
spacing effect. Familiar stimuli should be easier to rehearse
and, hence, benefit more from spaced presentations. Encoding
variability theory predicts a stronger spacing effect for familiar stimuli because familiar stimuli have more preexisting
associates. Spaced presentations create opportunities for a
wider set of cues to become more strongly associated with a
stimulus and, hence, improve recall.
In contrast to the above explanations, the retrieval and
reconstruction hypotheses predict that the spacing effect
should be stronger for novel stimuli. The retrieval hypothesis
prediction is based on the assumption that novel stimuli are

more difficult to retrieve at P2. The reconstruction hypothesis prediction is based on the assumption that novel stimuli
are harder to retrieve and are more likely to be reconstructed
at P2.
The effect size was larger for novel stimuli (M p
.424) than the familiar stimuli (M p .338) , but this difference was not statistically significant (x 2 (1) p 2.72, p 1
.10). We were concerned that this test was nondiagnostic
due to the lack of power created by the small number of
observations in the novel condition (e.g., n p 12). An additional test used the standardized mean performance data
reported in table 2. The difference in the average proportion
of correct responses was greater for novel stimuli
(Mmassed p .411, Mdistributed p .575) than familiar stimuli
(Mmassed p .404, Mdistributed p .511; (F(1, 214) p 4.78, p !
.05). The direction of the effect size differences in the two
tests and the statistical significance of the second test are
consistent with the predictions of the retrieval and reconstruction hypotheses.

Complex/Simple Stimuli. Stimuli can be simple (e.g.,

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

JOURNAL OF CONSUMER RESEARCH

144

the word cat), structurally complex (e.g., the sentence

The cat is on the red brick wall.), or semantically complex
(e.g., homographs; Challis 1993; Greene 1989). The rehearsal hypothesis predicts that structurally and semantically
complex stimuli should be more difficult to rehearse than
simple ones. The lower rehearsal rate in the spaced condition
should lead to a weaker spacing effect for complex stimuli.
In contrast, the encoding variability hypothesis predicts that
a complex stimulus has more preexisting associations and,
since spaced presentations allow for a greater variety of
these associations to be reinforced, there should be a stronger
spacing effect for complex stimuli. The retrieval and reconstruction hypotheses further differentiate between spacing effects for structurally and semantically complex stimuli.
It should be more difficult to retrieve or reconstruct a semantically complex P1 stimulus at P2, and this additional
processing should lead to a stronger spacing effect than for
simple stimuli. In general, retrieval does not depend on
structural cues, so variability in structural complexity should
not influence the spacing effect.
A test for differences in the average effect size across the
three different types of stimulus complexity was statistically
significant (x 2 (2) p 23.13, p ! .05). The average effect size
for structurally complex stimuli (M p .330) was not significantly different than the average effect size for simple
stimuli (M p .325; x 2 (1) p 1.56, p 1 .05), but semantically complex stimuli (M p .586) had a larger average effect size than simple stimuli (M p .325; x 2 (1) p 19.95,
p ! .05). These results are consistent with the retrieval and
reconstruction hypotheses, partially consistent with the encoding variability hypotheses, and inconsistent with the rehearsal hypothesis.

Unimodal/Bimodal Presentation Format. Stimuli

can be presented using a unimodal format (e.g., only visual
presentation, only auditory presentation) or a bimodal format
(combined visual and auditory presentation). The attention,
retrieval, and reconstruction hypotheses all predict a weaker
spacing effect for a bimodal format. The attention hypothesis
predicts a weaker spacing effect for a bimodal presentation
since these stimuli should sustain attention to P2 in a massed
presentation context. The retrieval hypothesis assumes that a
bimodal presentation of massed stimuli will encourage the
retrieval of P1 at P2; hence, there will be little retrieval benefit
to spacing. The reconstruction hypothesis assumes that a bimodal presentation format will encourage an alternative construction of P1 at P2 in the massed condition. Again, there
will be little benefit to spacing. In contrast, the encoding
variability hypothesis predicts that a bimodal presentation format should access more preexisting contextual and descriptive
associates and that a distributed presentation schedule should
create more opportunities for these cues to become more
strongly associated with a stimulus. The results should be a
stronger spacing effect.
The average effect sizes for visual (M p .290), auditory
(M p .449), and bimodal (M p .302) stimuli were significantly different (x 2 (2) p 30.22, p ! .05). However, with
the unimodal formats combined (M p .347), the average

effect size did not differ significantly from the average effect
size for bimodal stimuli (M p .302; x 2 (1) p 0.51, p 1
.10). The average effect size for the unimodal auditory stimuli (M p .449) did significantly differ from the average
effect size for bimodal stimuli (M p .302; x 2 (1) p
8.90, p ! .05). These results are partially consistent with the
predictions of the attention, retrieval, and reconstruction
hypotheses.

Unrelated/Related Cues in a Paired-Associate Learning Task. Stimuli can be learned in a paired associate task
in which the cue and target are unrelated (e.g., wood-snow),
structurally related (e.g., crow-snow), or semantically related (e.g., ice-snow). The rehearsal hypothesis predicts that
related cues should make cue-target associations easier to
rehearse. Sustaining rehearsal between P1 and P2 should
lead to a stronger spacing effect. The encoding variability
hypothesis predicts no influence of cue relatedness. Experiencing the same paired-associate at P1 and P2 limits the
degree to which other cues can become associated with the
target stimulus. The retrieval and reconstruction hypotheses
predict that it will be more difficult to retrieve or reconstruct
a stimulus as the target and cue become less related. This
additional processing should lead to a stronger spacing effect
for unrelated targets.
A test for differences in the average effect size across
different types of cues was not statistically significant
(x 2 (2) p 1.35, p 1 .10). Also, in a separate test, the average
effect size for unrelated cues (M p .297) did not differ from
the effect size for both types of related cues (M p
.404; x 2 (2) p 1.21, p 1 .10). The statistically insignificant
test of the spacing effect is consistent with the encoding
variability hypothesis. However, the low statistical power
due to the small number of studies investigating cue relatedness led us to run an additional test using the standardized
mean memory performance data reported in table 2. The
improvement in memory for structurally related
(Mmassed p .460, Mdistributed p .665) and semantically related
(Mmassed p .415, Mdistributed p .539) cues was significantly
greater than the memory improvement for unrelated cues
(Mmassed p .455, Mdistributed p .525; F(1, 28) p 7.53, p !
.05). This result is consistent with the rehearsal hypotheses.
However, it is noteworthy that the one direct test of this
hypothesis found no difference between semantically related
and unrelated cues (Johnston et al. 1972).

Intentional/Incidental Learning. People can process

material with the intent to learn it or be exposed incidentally.
All of the explanations predict a larger spacing effect when
material is processed intentionally, but for different reasons:
people engaged in intentional learning are more likely to
ignore P2 in the massed condition because they recognize
the presentation as redundant (attention hypothesis); varying
the spacing of the stimuli will have a larger influence when
the respondent is actively rehearsing the material (rehearsal
hypothesis); intentional learning should reinforce more associations at a distributed P1 and P2 than incidental learning
(encoding variability hypothesis); intentional learning en-

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

SPACING EFFECT

145

courages a person to actively attempt to retrieve P1 (retrieval

hypothesis); P2 will be elaborated in the attempt to retrieve
P1, and when not successful, will encourage construction
of a representation of P2 (reconstruction hypothesis).
The average effect size for intentional processing
(M p .352) was larger than for incidental processing
(M p .236; x 2 (1) p 9.24, p ! .05). These results are consistent with the predictions of all of the hypotheses. We also
note that three experiments have manipulated the learning
goal for a direct test of its influence on the spacing effect.
Greene (1989) found a larger spacing effect in intentional
learning situations for cued recall but no difference in the
size of the spacing effect across learning situations for free
recall. Challis (1993) found that a spacing schedule improved recall in cued-memory tests under intentional and
incidental semantic learning conditions but not in an incidental learning condition that encouraged graphemic (surface) processing of the stimuli. However, in a test of the
retrieval hypothesis, Braun and Rubin (1998) found a significant spacing effect in both cued and free recall tests but
no difference between intentional and incidental learning
conditions.

Isolated/Embedded Stimuli. Stimuli can be presented

in isolation (i.e., one word or image at a time) or as part of
a larger whole (e.g., a target word in a sentence or a brand
name in an ad). The attention, rehearsal, and encoding variability hypotheses predict that isolated stimuli should show
a stronger spacing effect than embedded stimuli, whereas
the retrieval and reconstruction hypotheses predict the opposite. The attention hypothesis posits that isolated stimuli
should be more noticeable at P1 (the Von Restorff effect),
receive less attention at a massed P2, and thus show a
stronger spacing effect. The rehearsal hypothesis predicts
that embedded stimuli will compete with material from the
embedding context for access to short-term rehearsal and,
hence, a weaker spacing effect for embedded stimuli. The
encoding variability hypothesis predicts that a stimulus surrounded by other items should have an opportunity for variable encoding but, to the extent this surrounding context
is the same at P1 and P2, this variable encoding should
remain constant. In other words, there is no advantage to
spacing the presentations. Thus, there should be greater opportunity for variable encoding when the stimulus is presented in isolation.
The retrieval and reconstruction hypotheses predict a
weaker spacing effect for isolated stimuli. The retrieval hypothesis prediction is based on the assumption that it is easier
to retrieve an isolated P1 stimulus at P2. The reconstruction
hypothesis also assumes that it is easier to retrieve an isolated P1 stimulus and avoid reconstruction of the stimulus
at P2.
The difference between the average effect size for the
isolated (M p .333) and embedded (M p .371) learning
contexts was not significant (x 2 (1) p 0.26, p 1 .10). These
results are inconsistent with all of the hypotheses.
Simple/Complex Intervening Material. The material

that is processed between repeated presentations can be simple (e.g., nothing, monosyllabic words), structurally complex (e.g., multisyllabic words, sentences), or semantically
complex (e.g., text passages, sentences in a story line). The
rehearsal hypothesis assumes that semantically and structurally complex intervening material is more difficult to
rehearse and, thus, interferes with the rehearsal of the target
material. As a consequence, the target material becomes less
sensitive to the spacing manipulation. In contrast, the encoding variability hypothesis assumes that the surrounding
material is a contextual cue and complex surrounding material will engender a stronger spacing effect. The retrieval
and reconstruction hypotheses differentiate among the type
of complexity represented in the intervening material. The
former predicts that semantically complex intervening material should make retrieval of P1 more difficult at a delayed
P2, thus resulting in a stronger spacing effect relative to
both simple and structurally complex intervening material
conditions. The reconstruction hypothesis also predicts a
stronger spacing effect for semantically complex intervening
material since it should be more difficult to retrieve P1 and
force reconstruction of the stimulus representation at a distributed P2.
A test for differences in the average effect size across
different types of intervening stimulus complexity was statistically significant (x 2 (2) p 6.78, p ! .05). The effect size
for semantically complex intervening stimuli (M p .419)
was significantly greater than when either simple intervening stimuli (M p .331) or structurally complex intervening
stimuli (M p .327) were used (x 2 (1) p 5.55, p ! .05), with
no significant differences between the latter two conditions
(x 2 (1) p 2.30, p 1 .10). These results are consistent with
the predictions of the retrieval and reconstruction
hypotheses.

Limitations of the Meta-analysis

Our results and any implied conclusions share limitations
common to all meta-analyses. First, although the data come
from experimental studies, our results are essentially based
on a meta correlation of the size of the spacing effect and
the independent variables coded across these experimental
studies. Except for the four noted experiments, the studies
used in the meta-analysis include no direct, nonconfounded
tests of the hypotheses (i.e., the independent variables are
correlated). Second, the results are based only on the data
available from a natural (i.e., nondesigned) experiment
(Farley, Lehmann, and Sawyer 1995). That is, although one
can conceive of a matrix of studies that includes all combinations of the nine factors, the 269 observations in our
analysis obviously fall far short of the 1,152 cells that would
be required to construct this matrix. Moreover, the observations are lumped in that they include redundant combinations of independent variables. Third, some of our tests,
such as those investigating semantic complexity, are underpowered. Fourth, this meta-analysis does not weight individual studies on the basis of their quality. Research like
ours has been labeled a bare-bones meta-analysis, since

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

JOURNAL OF CONSUMER RESEARCH

146

it considers only the quantitative factors of (1) levels of the

potential moderating factor, (2) sample size, and (3) sample
effect sizes. We preferred to confine our study to substantive
and procedural variables that could be coded with considerably less controversy, rather than to introduce our biases
about study quality.
One method of understanding the impact of confounded
independent variables in a meta-analysis is to (1) examine
the correlations among the independent variables, (2) run a
multiple regression using all of the independent variables,
and (3) compare these results in order to gain insight into
the underlying factors that drive responses (Farley et al.
1995). A regression was run using dummy codes of the
variables presented in table 2. There were significantly larger
effect sizes for nonsense syllables (b p .174, t p 2.00,
p ! .05), stimulus semantic complexity (b p .162, t p
2.44, p ! .05), and an auditory presentation format (b p
.258, t p 2.79, p ! .05). An examination of a correlation
matrix based on a nominal coding of all levels of each factor
showed mostly small and nonsignificant correlations but a
few noteworthy exceptions. The high correlations among
the nonsense word, meaningless word, and novel word observations led to the realization that these data came from
a common set of studies. Thus, hypotheses 2 and 3 in table
1 are redundant. We also observed that the semantically
complex stimuli were more likely to be presented auditorily.
However, the effect of the presentation format remained after
controlling for stimulus complexity. Similarly, the effects
of an auditory presentation format and stimulus complexity
remained after controlling for nonsense words. Analysis also
revealed that the nonsense syllable and auditory presentation
studies were responsible for the significant intentional learning result reported in table 1. When these studies were removed from the intentional learning analysis, the remaining
149 studies showed no influence of whether the learning
tasks were incidental or intentional.

DISCUSSION
There were two goals for this research study. The first
goal was to gain insight into the processes responsible for
the memory improvement that results from distributed presentations. Our meta-analysis, which is the first to examine
the spacing literature, was used to compare five learning
theories: the attention, rehearsal , encoding variability, retrieval, and reconstruction hypotheses. The results are most
consistent with the retrieval and reconstruction hypotheses.
The retrieval (five successful predictions) and reconstruction
(seven successful predictions) theories perform best, followed by encoding variability with three correct and two
partially correct predictions. This result is surprising given
that voluntary attention (e.g., Malaviya and Sternthal 1997)
and encoding variability (e.g., Schumann et al. 1990; Singh
et al. 1994; Unnava and Burnkrant 1991) have been the
dominant explanations of spacing effects within the consumer behavior literature.
Although we again note that almost all of the studies that
were available for our meta-analysis came from the verbal

learning literature, we think it is warranted to speculate and

offer propositions for consumer behavior and advertising research and practice. These results are the only ones available,
and the analyzed studies represent the best attempts to understand how people process, learn, and remember information.
First of all, the retrieval and reconstruction hypotheses
can be used to generate propositions about how to use repetition to improve memory of advertising material. The retrieval explanation emphasizes the importance of obtaining
a strong initial encoding of a stimulus at P1 but sufficiently
limiting retrieval cues at P2 such that the retrieval is difficult
but successful. Consistent with the often cited theorizing of
Krugman (1970) about how advertising benefits from three
repeated exposures, the reconstruction explanation stresses
the importance of a strong, partial encoding at P1 in order
to encourage elaboration at P2 as a representation of the
stimulus is constructed. Thus, both explanations argue that
an optimal repetition strategy should encourage incidental
processing during one presentation and intentional processing in the other, but the two differ about the optimal order
of these two types of processing. Conceptually, this means
that a combination of elaborated and incidental processing
of the same advertising material may result in better memory
for the material than two occasions of elaborated processing.
Practically, this suggests that a more effective repetition
strategy may include a combination of involving media (e.g.,
television commercials) and less involving media (e.g., billboards, product placements) or messages that vary in the
level of involvementfor example, complex versus simple,
long versus short, or hard sell versus soft sell (Ray and
Sawyer 1971). It is also worthwhile to note that this variability in involvement does not have to occur across ads.
Varying the depth and breadth of encoding at P1 and P2
within a specific ad may make the ad more effective at
promoting memory for the material (in addition to being
more consistent with the time intervals most commonly used
in the verbal learning studies).

Moderators of the Spacing Effect

The 10 hypotheses used to discriminate between the explanations of the spacing effect investigated a number of
content and context variables that are important in advertising. Some of the content and context variables have garnered a small amount of consumer behavior research but
have lacked the theoretical underpinnings needed to make
a conceptual advance. In this final section, we attempt to
integrate the findings of the meta-analysis, the existing consumer behavior literature, and the assumptions of the retrieval and reconstruction hypotheses to create some interesting insights and testable propositions about repetition and
memory.

Repetition and Memory. It is well known that repetition of advertising is an effective learning tool. Laboratory
(e.g., Singh et al. 1994) and field (Zielske 1959) evidence
shows that memory for repeated material improves as the

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

SPACING EFFECT

147

time between presentations of advertising material increases,

especially when there is a delay between the second presentation of the stimulus and the memory test (Singh et al.
1994). Yet, there is also evidence that repeated exposures
are ineffective, a somewhat counterintuitive finding. Appel
(1971) and Blair (2000) present field evidence that memory
for a second (or nth) exposure to an ad declines as the time
between the first and second exposure increases, even if the
recall measurement is administered soon after the second
exposure. They also find that the decline in recall is not as
severe if the initial exposure to the ad resulted in a more
memorable or persuasive experience. These results are consistent with the retrieval hypothesis. To the extent that the
initial exposure is engaging, a strong memory trace is created. As the time between P1 and P2 increases, the number
of people who can retrieve this trace at P2 declines; hence,
the benefit of the first exposure declines. Yet, the P1 memory
failure will be slower for initially engaging ads. Note that,
if engaging ads simply capture attention at P2, memory at
test should be a function of the time since P2, not the time
between P1 and P2, because everyone is assumed to process
the ad at P2. Our proposed explanation of a well-documented, counterintuitive field finding would benefit from
laboratory investigation.

Pictorial/Verbal Stimuli. The recall of pictures has

been found to be more enduring than the recall of words
(Gardner and Houston 1986). Our analysis indicates that
spacing should benefit pictorial stimuli more than verbal
stimuli when cued recall tests are used. Consistent with this
finding, Keller (1987) found that a picture in a point-ofpurchase display from a memorable ad campaign facilitated
the recall of ad claims. On the other hand, our analysis shows
that, in situations where consumers might engage in memory-based choice, verbal ads would be more sensitive to
repetition-based advertising strategies.
Ray, Sawyer, and Strong (1971) found that the repetition
of color ads resulted in greater increases in ad recall than
the repetition of otherwise equivalent black and white ads.
However, when the measure of memory was the depth of
recall, repetition of black and white ads was more effective
than the repetition of color ads. These data suggest that
repeated exposure might have facilitated recall for different
types of information given different retrieval/reconstruction
cues (e.g., color might have been an effective retrieval cue
for ad identity, whereas brand name and product category
information might have been effective retrieval cues for ad
content). These results are consistent with the retrieval and
reconstruction explanations of the spacing effect and are
worthy of further investigation.
Stimulus Meaning. Keller, Heckler, and Houston
(1998) suggest that it is better to use meaningful brand
names when selecting a brand name for a new product; yet
there is anecdotal evidence that some manufacturers prefer
initially meaningless brand names. If advertisers do use initially meaningless brand names or product logos, it is interesting to hypothesize about how repetition can encourage

the development of meaningful brand associations (e.g.,

benefits, usage situations). Conventional wisdom assumes
that repetition is better than no repetition and elaborated
processing is better than incidental processing. Yet, the retrieval and reconstruction explanations argue that associated
meanings should be primary in some exposure but secondary in other exposures and that presentations of these two
types of ads should be spaced. The hypothesis that primary
information will encourage reconstruction of secondary information in previous ads or that secondary information will
encourage retrieval of primary information from previous
ads is novel. The proposition that either of these ad presentation strategies will result in stronger long-term memory
than a strategy that relies on elaborated processing of an ad
at each presentation is intriguing and worthy of empirical
investigation.

Stimulus Complexity. The results of the meta-analysis

show semantically complex stimuli benefit from distributed
presentations but that structurally complex stimuli do not.
These results suggest it may be useful to investigate how
different types of semantic ad complexity might respond to
massed or distributed presentations. For example, semantically complex ads could consist of multiple competing
meanings (e.g., unrelated product benefits), multiple synergistic meanings (e.g., related product benefits), a set of
overlapping associations (i.e., a stimulus with rich associations), or a clever and insightful interpretation (i.e., resonant ads). Stimuli consisting of each type of complexity
might respond differently to massed or distributed exposures. For example, Janiszewski and Meyvis (2001) find that
a pair of stimuli with a common meaning benefit more from
a massed repetition schedule and that a pair of stimuli with
competing meanings benefit more from a distributed presentation schedule. This may imply that memory for resonant ads (McQuarrie and Mick 1992) should benefit more
from distributed exposures because the text of the ad and
the text /picture combination create different meanings. Yet,
solving the pun of the ad does lead to a single, interesting
interpretation of the ad. Thus, resonant ads represent stimuli
with multiple initial meanings, a subsequent unidimensional
interpretation, and the potential to continually reengage the
reader. We have little insight into how material having this
level of complexity might interact with repetition schedules
to influence memory.

Conclusion
Our meta-analysis of the spacing effect has produced
some interesting results. We have offered several propositions based on these results and the two theories most consistent with the data. Clearly, such propositions must be
tested in a marketing context and in the field. We hope that
these results and inferred propositions intrigue other consumer researchers, and we encourage them to join us in this
future research.

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

JOURNAL OF CONSUMER RESEARCH

148

[David Glen Mick served as editor and Donald R. Lehmann served as associate editor for this article.]

REFERENCES
Appel, Valentine (1971), On Advertising Wear Out, Journal of
Advertising Research, 11 (February), 1113.
Balota, David A., Janet M. Duchek, and Ronda Paullin (1989),
Age-Related Differences in the Impact of Spacing, Lag, and
Retention Interval, Psychology and Aging, 4 (March), 39.
Berlyne, Donald E. (1970), Novelty, Complexity, and Hedonic
Value, Perception and Psychophysics, 8 (November),
279286.
Bjork, Robert A. (1988), Retrieval Practice and the Maintenance
of Knowledge, in Practical Aspects of Memory II, ed. Michael M. Gruneberg, Peter E. Morris, and Robert N. Sykes,
London: Wiley, 396401.
Blair, Margaret Henderson (2000), An Empirical Investigation of
Advertising Wearin and Wearout, Journal of Advertising Research, 40 (November), 95100.
Braun, Kelly Ann and David C. Rubin (1998), A Retrieval Model
of the Spacing Effect, Memory, 6 (January), 3765.
Burke, Raymond R. and Thomas K. Srull (1988), Competitive
Interference and Consumer Memory for Advertising, Journal
of Consumer Research, 15 (June), 5568.
Challis, Bradford H. (1993), Spacing Effects on Cued-Memory
Tests Depend on Level of Processing, Journal of Experimental Psychology: Learning, Memory, and Cognition, 19
(March), 389396.
Dempster, Frank N. (1996), Distributing and Managing the Conditions of Encoding and Practice, in Handbook of Perception
and Cognition: Memory, ed. Elizabeth Lignon Bjork and Robert A. Bjork, San Diego, CA: Academic Press, 317344.
Ebbinghaus, H. ([1885] 1913), Memory, trans. H. A. Ruger and
C. E. Bussenius, New York: Teachers College.
Elmes, David G., Paul F. Chapman, and Camden W. Selig (1984),
Role of Mood and Connotation in the Spacing Effect, Bulletin of the Psychonomic Society, 22 (May), 186188.
Farley, John U., Donald R. Lehmann, and Alan G. Sawyer (1995),
Empirical Marketing Generalization Using Meta-analysis,
Marketing Science, 14 (September), G36G46.
Gardner, Meryl P. and Michael J. Houston (1986), The Effects
of Verbal and Visual Components of Retail Communications,
Journal of Retailing, 62 (Spring), 6478.
Glanzer, Murray and Anibal Duarte (1971), Repetition between
and within Languages in Free Recall, Journal of Verbal
Learning and Verbal Behavior, 10 (December), 625630.
Glenberg, Arthur M. (1979), Component-Levels Theory of the
Effects of Spacing of Repetitions on Recall and Recognition,
Memory and Cognition, 7 (March), 95112.
Greene, Robert L. (1989), Spacing Effects in Memory: Evidence
for a Two Process Account, Journal of Experimental Psychology: Learning, Memory, and Cognition, 15 (May),
371377.
Hintzman, Douglas L. (1974), Theoretical Implications of the
Spacing Effect, in Theories in Cognitive Psychology: The
Loyola Symposium, ed. Robert L. Solso, Hillsdale, NJ: Erlbaum, 7799.
Hintzman, Douglas L. and Miriam K. Rogers (1973), Spacing
Effects in Picture Memory, Memory and Cognition, 1 (October), 430434.
Jacoby, Larry L. (1978), On Interpreting the Effects of Repetition:
Solving a Problem versus Remembering a Solution, Journal

of Verbal Learning and Verbal Behavior, 17 (December),

649667.
Janiszewski, Chris and Tom Meyvis (2001), Effects of Brand
Logo Complexity, Repetition, and Spacing on Processing Fluency and Judgment, Journal of Consumer Research, 28
(June), 1832.
Johnston, William A., James H. Coots, and Robert G. Flickinger
(1972), Controlled Semantic Encoding and the Effect of Repetition Lag on Free Recall, Journal of Verbal Learning and
Verbal Behavior, 11 (December), 784788.
Keller, Kevin L. (1987), Memory Factors in Advertising: The
Effect of Advertising Retrieval Cues on Brand Evaluations,
Journal of Consumer Research, 14 (December), 316333.
Keller, Kevin L., Susan E. Heckler, and Michael J. Houston (1998),
The Effects of Brand Name Suggestiveness on Advertising
Recall, Journal of Marketing, 62 (January), 4859.
Kraft, Robert N. and James J. Jenkins (1981), The Lag Effect
with Aurally Presented Passages, Bulletin of the Psychonomic Society, 17 (March), 132134.
Krugman, Herbert (1970), Why Three Exposures May Be
Enough, Journal of Advertising Research, 12 (NovemberDecember), 1114.
Liang, K. Y. and S. L. Zeger (1986), Longitudinal Data Analysis
Using Generalized Linear Models, Biometrika, 73, 1322.
Malaviya, Prashant and Brian Sternthal (1997), The Persuasive
Impact of Message Spacing, Journal of Consumer Psychology, 6 (3), 233255.
McQuarrie, Edward F. and David G. Mick (1992), On Resonance:
A Critical Pluralistic Inquiry into Advertising Rhetoric, Journal of Consumer Research, 19 (September), 180197.
Melton, Arthur W. (1970), The Situation with Respect to the
Spacing of Repetitions and Memory, Journal of Verbal
Learning and Verbal Behavior, 9 (October), 596606.
Postman, Leo and Kitty Knecht (1983), Encoding Variability and
Retention, Journal of Verbal Learning and Verbal Behavior,
22 (April), 133152.
Ray, Michael L. and Alan G. Sawyer (1971), Repetition in Media
Models: A Laboratory Technique, Journal of Marketing Research, 8 (February), 2029.
Ray, Michael L., Alan G. Sawyer, and Edward C. Strong (1971),
Frequency Effects Revisited, Journal of Advertising Research, 11 (JanuaryFebruary), 1420.
Rethans, Arno, John L. Swasy, and Lawrence J. Marks (1986),
Effects of Television, Commercial Repetition, Receiver
Knowledge, and Commercial Length: A Test of the TwoFactor Model, Journal of Marketing Research, 23 (February), 5061.
Rosenthal, Robert (1994), Parametric Measures of Effect Size,
in The Handbook of Research Synthesis, ed. Harris Cooper
and Larry V. Hedges, New York: Sage, 231244.
Rundus, Dewey (1971), Analysis of Rehearsal Processes in Free
Recall, Journal of Experimental Psychology, 89 (July),
6377.
Sawyer, Alan G. (1981), Repetition, Cognitive Response, and
Persuasion, in Cognitive Responses to Persuasion, ed. Richard Petty, Tom Ostrum, and Timothy Brock, Hillsdale, NJ:
Earlbaum, 237262.
Schumann, David W., Richard E. Petty, and D. Scott Clemons
(1990), Predicting the Effectiveness of Different Strategies
of Advertising Variation: A Test of the Repetition Variation
Hypothesis, Journal of Consumer Research, 17 (September),
192202.
Singh, Surendra N., Sanjay Mishra, Neeli Bendapudi, and Denise

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions

SPACING EFFECT

149

Linville (1994), Enhancing Memory of Television Commercials through Message Spacing, Journal of Marketing
Research, 31 (August), 384392.
Toppino, Thomas C. (1993), The Spacing Effect in Preschool
Childrens Free Recall of Pictures and Words, Bulletin of the
Psychonomic Society, 31 (January), 2730.
Tzeng, Ovid J. (1973), Stimulus Meaningfulness, Encoding Var-

iability, and the Spacing Effect, Journal of Experimental

Psychology, 99 (July), 162166.
Unnava, H. Rao and Robert E. Burnkrant (1991), Effects of Repeating Varied Ad Executions on Brand Name Memory,
Journal of Marketing Research, 28 (November), 406416.
Zielske, Hubert A. (1959), The Remembering and Forgetting of
Advertising, Journal of Marketing, 23, 239243.

This content downloaded on Wed, 26 Dec 2012 13:38:37 PM

All use subject to JSTOR Terms and Conditions