174 Research Methodology ttitude Measurement and Scaling 175

an erample of the affective component. The other example could be that the
because each of the item forms some part of the construct (satisfaction) which the
products produced by Pearl Pet are cheaper than upperware products; how
the quality of Tupperware products is better than that of Pearl Pet researcher is trying to measure. As an example, some of the following questions may
be asked in a multiple item scale.
Intention component: This component
or action attitude, also called.
ol an
the How satisfied are you with the pay you are getting on your current job?
behavioural component, reflects a predisposition to an action by reflecting
he Very dissatisfied
consumer's buying or purchase intention. It also retlects a person's expectationeof
The behavioural future behaviour towards an object. How likely a person 1s to buy a designer carpet Dissatisfied

component of an attitude may range from most likely to not at all likely, reflecting the purchase intentions Neutral
reilects a predispesition However, when one is talking about the purchase intentions, a time horizon has to Satisfied
to an action by retlecting Very satisied
be kept in mind as the intentions may undergo a change over tüme. The intentions
the consumers buying or How satisfied are you with the rules and regulations of your organization?
incorporate information regarding the respondents Wilingness to pay for the
purchase intention.
product. Very dissatisfied

a favourable attitude
and behaviour. If
There is a relationship between atitude a consumer
he/she will certainly not
does Dissatisfied
Neutral
not
have towards the product,
the product. However, having a favourable attitude
buy
Satisfied
does not mean that it would be Very satisfied
retected in the purchase behaviour. This is becau_e
intention to buy a product has
to be by power of a favourable
backed the purchasing the consumet.Having attitude
towards Mercedes Benz does not mean that a person is going to purchase it even
How satisied are you with the job security in your current job?
Very dissatisfied
if he does not have the ability to buy a product. Therefore, the relationship between Dissatisfied
the attitude and the purchase behaviour is a necessary condition for the purchase of Neutral
the product but it is not a sufficient condition. This relationship could hold true at Satisfied
the aggregate level but not at the individual level. Very satisfied

Define atitude. Comparative vs Non-comparative Scaies

2. What is meant by the tem 'affective component'?
The scaling techniques used in research can also be classified into comparative and
non-comparative scales (Figure 7.1).
CLASSIFICATION OF sCALES
FIGURE 7.1
ypes of scalirg Scaling Techniques
LEARNING OBJECTIVE 4
Discuss the various
One of the ways ofthis,
scale. Based
classifications
the
of scales is in terms of the number of items in the
techniques
upon classification
following be may proposed:
classificaftions of scales.

Single ltem vs Multiple Item Scale Comparative Scales Non-comparative Scales

Single item scale: In the single item scale, there is only one item to measure a given
construct. For example:
Paired Comparison
Consider the following question: Graphic Rating Scale Itemized Raing Scaie
(Continuous Rating Scale)
How satisfied are you with your current job?
-

Constant sum
Very Dissatisfied
LIKert
Dissatisfied
Neutral Rank Order
Satisfied Semantic Differential

Very satisfied
Q-Sort and Other
'he problem with the above question is that there are many aspects to a job, Ik Procedures
pay, work environment, rules and regulations, security oi job and communicanon Stapel
with the seniors. The
respondent may be satisfied on some of the factors but
in nutiple item scale,
a not on others. By asking a question as stated above, it wili be difficult to analyse ne Comparative Scales
eack item forms some problem areas. To overcome this problem, a multiple itenm scale is In comparative scales it is assumed that respondents make use of a standard frame
paft of the construct that
proposea.
Multiple item scale: In multiple item scale, there are items that of reference before answering the question. For example:
the researdher s trying to 1n
1ormin8 the
underlying construct that the researcher is
many play a ro Barista in comparison to Cate Coffee Day on
measure. trying to measure. i A question like "How do you rate

of is example of the comparative rating scale. It involves the

quality beverages? an
 176 Research Methodo1oOy
Attitude Measurement and Scating 177

direct comparison of stimulus objects. For exaiupl p o n d e n t s may be acbo.

1n a coinparative scale, it is TABLE 7.2
assumed that a respondent whether they prefer Chinese in comparison to i 0d. Consider the followina Paired comparici data
A B D
E
set of questions generally used to compare various duibutes of Domino's Pizza a n
makes use of a standard frame 0.60 0.30 0.60 0.35
of reference before answering Pizza Hut.
the questio Please rate Domino's in comparison to izza u t on the basis of you 0.40 0.28 0.70 0.40
satisfaction level on an 11-point scale, based on the following parameters:
(1 = Extremely p o o r , 6 Average, 1l = Exremely good). Circle your 0.70 0.72 0.65 . 10

response: D 0.40 U.i30 0.35 0.42

a. Variety ofmenu options 1 2 3 45 6 7 8 9 10 11 0.65 0.60 0.90 0.58

b. value tor money 1 2 34 5 67 8 9 10 11
C.Speedof service (deliverytime) 1 2 3 4 56 7 8 9 10 11 The above table may be interpreted by assuming that the cell entry in the matrix
d. | Promotional offers 1 2 345 6 7 8 9 10 11 represents the proportion ot respondents who believe that "the column brand is
preferred over the row brand. For example:
e. Food quality 1234 6|7 8| 9 10 11
f.Brandname 1 2 34 56 7 89 10 11 In brand A versus brand B comparison it can be said that 60 per cent of the
9.|Quality of service 1 23 45|6 78 9 10 11 respondents preter brand B to brand A. Similarly, 30 per cent of the respondents

h. Convenience in terms oftakeaway 1 2 3 4 5 6 7

8 9 10 11 prefer brand C to brand A and so on.
To develop the ordinal scale from the given paired comparison data in the above
| location
2 34|5 6 7 8 9 1 0 11 table, we can convert the entries in the table to 0- 1 scores. This is to show whether the
i.Friendlinessof the salesperson on the column brand dominates the row brand and vice versa. if the proportion is greater
phone
than 0.5 in the above table, a number of T'is assigned to that cell, which means that
i. Quality of packaging 1 2 34 5 6 7 8 9 10 11
the column brand is preferred over the row brand. Whenever the proportion is less
k.AdaptationofIndian taste 1 2 34 5 67 8 910 11 than 0.5 in above table, a number of '0' is assigned to that cel, which means column
I. Side orders/appetizers 1 2 34 5 6 7 8 9 10 11 brand does not dominate the row brand. The results are in Table 7.3.
Comparative scale data is interpreted generally in a relative kind. The
comparative scale includes paired comparison, rank order, constant sum scale and TABLE 7.3
Q-sort technique to mention a few. Conversion of pairedd
will discuss below each of the scales under comparative ráting scales in comparison dats inta
detail:
We O to 1 torm
0
in a paired omparison
scale, a respondent is
Paired comparison scales: Here a respondent is presented with two objects and
is asked to select one according to whatever criterion he or she wants to use. The
presented with two objects
resulting from this scale is ordinal in nature. As an
and is asked to select one data example, suppose a parent
wants to offer one of the four items to a child-chocolhte, burger, ice cream and
according to vwhatever
riterion he/she wants to pizza. The child is offered to choose one out of the two from the six possible pairs, E

uSe. i.e,chocolate or burger, chocolate or ice cream, chocolate or

pizza, burger or ice Total
Cream, burger or pizza and ice cream or pizza. In general, if there are n items, the

number of paired comparison would be (n(n 1)/2). Paired comparison technique

-

To get the ordinal relationship among the brands, we total the columns. Here
is items because itrequires a direct
useful whenthe number of is limited
and overt choice. In case the number of items to be compared is large (say 10), it
comparison the ordinal scale of brands is D > B>A>C> E. This means brand D is the most

preferred brand, followed by B, A, C and E.

would result in 45paired comparisons which would further result in fatigue for the
does not make the choice from two
In order to obtain the interval scale data from the paired comparison data
respondents.
items
Further, in reality a respondent
time-there presented above, the entries in the table can be analysed by using a technique called
at a are multiple
alternatives available to him. Thurston's law of comparative judgement, which converts the ordinal judgements
nere are many ways of analysing the paired comparison data. The analysis into the interval data. Here the proportions are assumed as probabilities and using
of paired comparison data would result in an ordinal s ale and also in an interval
scalemeasurement. This will be shown with the help ofan example. Let us assume
the assumption of normality, Z-scores can
be computed. Z-value has symmetric
of "0' and variance of 1: If the proportion is less than 0.5,
distribution with a mean
that there are five brands-A, B,
C, D and E-and a paired comparison with two a negative sign and lor the proportion that is greater
Z-value has
brands at a time is presented to the the corresponding
respondent with the option to choose one o than 0.5, the Z-score takes a positive value. The Z-scores for the paired comparison
them. As there are five brands, it will
result in 10 paired comparisons. Suppose this data is given in Table 7.4.
is administered to a
sample of 250
respondents with the results as presented
Table 7.2.
 178 Research Methodology
ATttu
tutude
Measuremert and staling179
TABLE 7.4 B C E Allocate a total of 100 points among the various schools into
Z-sCoras for paired which you
COnparISOn Cat3
0.255 -0.525 0.255 38 would like to admit your child. The more the
points you allocate to a school,
-0.255 0 -0.58 0.5525 -0.255 more preferred it is to be considered. The
points should be allocated in
0.525 0.58 .385 1.28 such a way that the sum total of the points allocated to various schools adds
-0.255 -0.525 -0.385 up to 100.
-0.2
0.38 0.255 1.28 0.2 0
Schools
1.365 Points
Total Distance
0.395 0.565 -0.21 -2.115
0.079 -0.042 0.273 DPS
Average Distance 0.113 -0.423
Brand B A Modern School
Interval scale value with 0.696 0.536 0.502 0.381 Mother's International
changeof origin APEEJAY

DAV Public School

The in show
he average distance is
computed by dividing the the
entries Table 7.4
scores can
distance between two brands. Assuming that
the
be added, the total distance is computed. The Laxman Public School
average distance is
total score by the number of computed by dividing the total score by the number of brands. This way one obtains Tagore International
brands. Ihis way one obtains the absolute position of each brand. Nowthe highest negative values among all the TOTAL POINTS 100
the absolute position ot each column is added to each entry corresponding to the average value so that
by change Suppose Mother's International is awarded 30 points, whereas Laxman Public
brand. of origin, interval scale values can be obtained. This is shown in the last row and
the
values are of interval scale, indicating the difference between brands. Brand D is the School is awarded 15 points, one can make a statement that the respondent rates
most preferred brand and E is the least Mother's International twice as high as Laxman Public School. This type of data is
preferred brand and the distance between
the two is 0.696. The distance between brand C and E not only comparative in nature but could also result in ratio scale measurement. This
equals 0.381.
In the rank order scaling, Rank order scaling: In the rank order scaling, type of scale is widely used in allocating weights which the consumer may assign to
respondents are presented
respondents are presented with the various attributes of a product.
several objects simultaneously and asked to order or rank them
with several objects
according to some InaQ-sorttechnique, a Q-sort technique: The Q-sort technique was developed to discriminate among
criterion. Consider, for example the following question:
SiImuitaneously and asked to rank order procedure is used a large number of objects quickly. This technique makes use of the rank order
Rankthe following soft drinks in order ofyour preference, the most preferred
order of rank them according in which objects are sorted procedure in which objects are sorted into different piles based on their similarity
soft drink should be ranked one, the second nost preferred should be
to some riterion. into diferent piles based on
ranked two and so on. with respect to certain criterion. Suppose there are 100 statements and an individual
their similarnty ith respect
is asked to pile them into five groups, in such a way, the
to certain criterion. that strongly agreed
Soft Drinks statements could be put in one pile, agreed statements could be put
Rank in another pile,
Coke neutral statements form the third pile, disagreed statements come in the fourth pile
and strongly disagreed statements form the fifth pile, and so on. The data generated
Pepsi in this way would be ordinal in nature. The distribution of the
imca number of statement
in each pile should be such that the resulting data may follow a normal distribuion.
Sprite The number of piles need not be restricted to 5. It could be as large as 10 or more as

Mirinda the large number increases the reliability or precision of the

Seven Up
results
anta Non-comparative Scales
frame of
Like n thenon-comparative scales, the respondents do not make use of any
paired comparison,this approach is also
comparative in nature. 1ne
in the non-comparative
assumed to
problem with this scale is that if a respondent does not like Scales, the respondents do
reterence betore answering the questions. The resulting data is generally
mentioned soft drink and is forced to rank them in the order any of the above not make use of any frame of be interval or ratio scale. For example:
of his choice, then, tne The respondent may be asked to evaluate the quality of food in a restaurant on
soft drink which is ranked one should be reference before answering
treated as the least disliked soft drink and a five pont scale (1 = very poor, 2 = poor and 5 = very good). The non-comparative
the (questions.
similarly, the other rankings can be interpreted. This scale is very commonly used into two categories, namely, the graphic rating Scales and
the
measure preferences for brands as well
as attributes. The rank order
to Scales are divided
Likert
in the ordinal data. scaling resuits itemized ratingscales. The itemized rating scales are further divided into
these come under the category
In constant sum
rating Scale, semantic differential scale and Stapel scale. All
Constant sum rating scaling: In constant
Scale, the respondents are sum rating scale, the
respondents ae of the multiple item scales.
asked to allocate a total of 100
asked to allocate a total of points between various objects arid brands. 1
100 points betvieen var'pus respondent distributes the points to the various objects in the order of his pretere
Consider the following
objects and brands, example:
 de225urement anc Scaling 181
180 Research Methodology

Graphic rating scale are certain issues that should be kept in mind while designing the itemized rating
calied graphic rating Scale. In the graphic rating scale scale. These issues are:
This is a continuous scale, also
on a grapn. Lonsider for example the Number of categories to be used: There is no hard and fast rule as to how many
the respondent is asked to tick his preference
categories should be used in an itemized rating scale. However, it is a practice to
following question:
Please put a tick mark ( ) on the following i n e to indicate your preference use five or six categories. Some researches are of the opinion that more than five
in attitudes are to be
for fast food. categories should be used in situations where small changes
would find it difficult to
7 Most measured. There are others that argue that the respondents
Least 1 It is, however, a fact that the additional
Preferred distinguish between more than five categories.
Preferred with the attitude being measured. It is
categories need not increase the precision
seen that researchers use five-category
scales and in special cases, may
To measure the preference of an individual towards fast food one has to measure generally
increase or decrease the number of categories.
the distance from the extreme left to the positüon where a tick mark has been put.
the
Odd number of categories: It has been a matter of debate among
Higher the distance, higher would be the individual preference for fast food. This or even
used in survey
or even number ot categories are to be
scale suffers from two limitations-one, if a respondent has put a tick mark at a researchers as to whether odd
research. By using even number of categories the
scale would not have a neutral
particular position and after ten minutes, he or she is given another form to put a or the
be forced to choose either the positive
tick mark, it will virtually be impossible to put a tick at the same position as was done category and the respondent will
numbers of categories are used, the respondent
negative side of the attitude. If odd
earlier. Does itmean that the respondent's
in this
preference for fast food has undergone
that has the freedom to be neutral if he wants
to be so. The Likert scale (to be discussed
a change inl0 minutes? The basic assumption scale is the respondents
can distinguish the fine shade in differences between the preference/attitude which later) is a balanced rating scale with an odd
number of categories and a neutral
that ifa respondent is not aware of the subject matter being
need not be the case. Further, the coding, editing and tabulation ofe a generated point. It is generally seen
neutral. However, if we have selected
try to avoid using it.
measured by the scale, he would preter to be
through such a procedure is a tedious task and researchers
Another version of graphic scale could be the following: ourunit of analysis to be one who is knowledgeable about
the study being conducted
him from this opportunity.
Please put a tick mark ( ) on the following line to indicate your preference and if he prefers to be neutral, we should not debar
is the one which has equal
for fast food. A balanced scala has equat Balanced versus unbalanced scales: A balanced scale
number of favouable and number of favourable and untavourable Examples of balanced and
categories.
Least 7 Most unbalanced scale are given below.
unfavourable categories.
Preferred Preferred The following is the example of a balanced scale:

car?
.Howimportant is price to you in buying
a new
This is a slightly better version than the one discussed earlier.
the limitaion ot the scale to some extent. For example, if a
It will overcome
respondent had earlier Very important
ticked between 5 and 6, it is likely that he would remember the same and Relatively important
time, he would tick very close to where he did earlier. This means that the difiference
thesecond Neither important nor unimportant

in the two responses could be negligible. Relatively unimportant

Very unimportant
way of
Another graphic rating scale is
presenting the
The following example would illustrate the same.
through smiling face scale. of which emphasize the
In this question, there are five response categories, two
Please indicate how much do you like fast food by pointing to the face that show its importance. The middle
importance of price and two others that do not
best shows your attitude and taste. If you do not prefer it at all, you would category is neutral.
oint to face one. In case you prefer it the most, you would poirnt to 1ace unbalanced scale.
The following is the example of the
seven.
car?
How important is price to you in buying a new

More important than any other factor

Extremely important
Important
Somewhat important
In the itemized roting Unimportant
scale, the respondents that are skewed towards the
dre provided Witn 3 scale question there are four response categories
In this
Itemized rating scale is for the unimportant side.
that has a number of brief importance given to the price, whereas one category
In the itemized In the unbalanced scale, the
desciptions 3ssociated rating scale, the respondents are provided with a scale that nas Therefore, this question is an unbalanced question.
vwith each of the response
number of
brief descriptions associated with each of the response categories numbers of favourable and unfavourable categories are not the same. One could
response use an unbalanced scale depending upon the nature
of attitude distribution to be
tategories. categories are ordered in terms of the
scale position and the respoue
measured. If the distribution is dominantly favourable, an unbalanced scale with
are

an
supposed to select the specified category that describes in the best posS1D
object is rated. Itemized rating scales widely used in survey research. 1 more favourable categories than untavourable categories
should be appropriate. If
are
ne
 C e 1easirement anC Scaling 183

182 Research VethodoicSy

an unbalanced scale is used, the nature and degree ot the unbalance in the scale To construct a Likert scale to measure a particular construct, a large number
of statements pertaining to the construct are listed. These statements could range
data analysis.
should be taken into account during the
from 80 to 120. The identification of the statements
is done through exploratory
Nature and degree of verbal description: Many researchers believe that each locus group, unstructured interviews
Verbai descriptions must
category must have a verbal, numerical or pictorial description. Verbal description
research which is carried out by conducting a
of studies and so on.
with knowledgeable people, literature survey, analysis
case
be ciearly ond precisely
should beclearly and preciselyworded so that the respondents are able to differentiate
worded so Suppose image of a company. As a first step, an exploratory
we want to assess the
between them. Further, the researcher must decide whether to label every scale interview with the customers, and
that the respondents i
research may be conducted by having an intormal
scale categories, or oniy extreme scale categories. It is argued that a also be contacted. A survey of
employees of the company. The general public may
able to d.fferentiate betv:een category, some

the. clearly defined response category increases the reliability of the measurement. also give a set ot intormation that could be useful
the literature on the subject may
Forced versus non-forced scales: An imporiant issue concerning the construction
for constructing the statements. Suppose
the numr of statements to measure the
of reptesentative respondents are asked
An important 1ssue
itemize rating scale is the use of a forced scale versus non-forced scale. In constructs is 100 in number. Now samples
concerning tne construction the forced scale, the respondent is forced to take a stand, whereas in the non-forced on those statements. Table 7.5 gives
to state their degree of agreement/disagreement
cf an iremzed ratng scale of the company.
scale, the respondent can be neutral if he/she so desires. The argument for a forced a statements to assess the image
few
is the use of a forced scale to do so and no numerical values are assigned
scale is that those who are reluctant to reveal their attirude
are
encouraged It may be noted that only anchor labels
versus non-forced scale.
scale, rank order scale and constant sum Once the scale is administered, numerical values are
with the forced scale. Paired comparison to the response categories.
of which
The scale contains statements' some
rating scales are examples of forced scales. assigned to the response categories.
and some are uniavourable
the construct we are trying to measure
form: There are many options that are available for the presentation of are favourable to
Physical
the scales. It could be presented vertically or horizontally. The categories could be to it.
not statements numbering 1, 2, 4, 6 and
expressed in boxes, discrete lines or as units on a continuum. They may may For example, out of the ten statements given,
or

have numbers assigned to them. The numeric a! values, it used, may be positive, whereas the remaining are untavourable
9 in Table 7.5 are favourable statements,
negative or both. statements. The reason for having a
mixture of favourable and unfavourable
to measure the perception about Jet Airways using a multi-
Statements in a Likert scale is that the responses by
the respondent should not
Suppose we want
members. Given in a Likert scale,
questions is about the behaviour of the become monotonous while answering the questions. Generally,
crew
item scale. One of the
their behaviour. favourable and unfavourable statements.
below is a set of scale configurations that may be used to measure
there is an approximately equal number of
ot presenting the scales assigned to the responses. The
The following are some of the examples where various forms Once the scale is administered, numerical values are
statement should get the same
are showm: rule is that a 'strongly agree' response for a favourable
statement.
The behaviour of the crew members of Jet Airways iS:
numericalvalue as the disagree' response of the unfavourable
'strongly

1. Verybad. Very good TABLE 7.5 No. Statement

Strongly Disagree Neither Agreestrongly
agree
Likert c:le disagree agree no
2. Very bad 1 3 Very good Statemerits to disagree
measurein3ga 1. The company makes
3. Verybad of the cormpany quality products
It is a leader in technology
Neither bad nor good It doesn t care about the

general public
The company leads in R&D
Verygood to improve products

4. Very bad Bad Neither bad nor good Good Very good . The company is not a good
paymaster
. -2 The products of the
.

Very bad Neither bad nor good Very good

company g0 through
Below we will describe some of tie itemized rating scales which are very stringent quality tests
commonly used in survey research. The company has not done
Likert scale s alsc celed s Likert scale: This is a ive-point
mulúple item agree-disagr ee scale.
The respondents anything to Curb pollution
It does not care
SemmaTed scale bec3use he a
are given certain number oi items (statements) on which ihey are asked
to express
about the
community near its plant
Scores on rdie3ual tems their degree of agreement/disagreement.
Can be acd-d together t
is also
This caled summated Scale
a

because the scores on individual itens can be added together to produce a total The company's stocks are
produce a otaB sxere for the scale is Uiat each of the items good to buy or own
responde1. Score lor ne respondent. An assumption of the Likert The company does not
(statements) measures some aspect of a single common factor, otherwise the scores 0.
on the items cannot legitimately be summed up. In a typical research study, there are have good labour relations
generally 25 to 30 items on a Likert scale.
 Atitude Measurement
and Scaling 185
184 Research vetnodoiogy

favourable statement the numbeng

is done as Strongly disagree. semantic differential
Semantic differential secale: This scale is widely used to compare the images of
Suppose for a In a
competing brands, companies or services. Here the respondent required
is to rate

Disagree 2 , Neither agree

nor disagree 3, Agree
4 and
Strongly agree =5. =

scale, a respondent is
scale is
1,
statement would get
the numerical values a to rate eacn atitude each attitude or object on a number of five-or seven-point rating scales. This
Accordingly, an unfavourable
4, Neither agree nor disagree = 3, Agree 2 and Strong
required
number of bounded at each end by bipolar adjectives or phrases. The difference between
Likert
disagree = 5, Disagree orobjezt on a a number of statements (items)
the image thattune iepondenthas about the company, fve-or-seven point rating and Semantic differential scale is that in Likert scale,
= 1. In order to measure
to express their degree of agreement/disagreement
agree are presented to the respondents
scores are added. SCales. or phrases are used. As
the
However, in the semantic diferential scale, bipolar adjectives
For example, if a respondent has tickea staements numbering from one to in the case of Likert scale, the information on the phrases and adjectives is obtained
total score would
be 3 +5+4+4 +5 + 4 +4+5+4+*
ten as shown in Table 7.5, his are 100 respondents and 100
statements, the score on through exploratory research. At times
there
may be a favourable or unfavourable
42 out of 50. Now if there side and on certain occasions these may be
4
descriptor (adjectives) on the right-hand
the image of the company can be
worked out ror
each respondent by adding his/her side. This rotation becomes necessary to avoid the halo
The minimum score for each respondent will be 100, presented on the left-hand
scores on the 100 statements. on the scale may iniuence
effect. This is because the location of previous judgments
whereas the maximum score would be 500. carelessness of the respondents. The mid
scale comprises about 25-30 statements. the subsequent judgements be cause of the
As mentioned earlier, a typical Likert of a bipolar scale is a neutral point.
In the Likert scale, ten statements were used
the 100 statements, we need to discard some point
In order to select 25 statements from where respondents were asked to express their degree
of agreement/disagreement
is that those items that are non-
of them. The rule behind discarding the
statements
same example further, the semantic
for choosing 25 (say number of regarding the image of the company. Taking the
discriminating should be removed. The procedure differential scale corresponding to those ten statements in Likert scale is
shown
statements) is show. seven points. These
each of the below where the bipolar adjectives/phrases are separated by
As mentioned earlier, the score for each of the respondents on
7 or +3, +2, +1, 0, -1, -2, -3 for a favourable
his/her total about the image of the points can be numbered as 1, 2, 3, .,

side. For an unfavourable descriptor the

score
statements can be used to measure
descriptor positioned on the left hand
company. The data may look as given in Table 7.6. semantic differential scale where bipolar
numberings would be reversed. A typical 7.7.
for 1 is 410, whereas for at the two extreme ends is given in Table
positioned
Table 7.6 shows that the total score respondent no.
adjectives/phrasesare
favourable
respondent no. 2 it is 209. This means that respondent no. 1 has
a more
Does not make quality
2. in order to select 25 TABLE 7.7 1 Makes quality products
image for the company as compared to respondent no. Now, products
Select bipolar
statements, let us consider statements numberingi and j. We note that the statement
i. This is because thhe adjectives/phrases of o C ao 5aCkward in technoiogy
no. j is more discriminating as compared to statement no. 2 Leader in technology
to semantic differential
sCore on statement j is very highly correlated with the total score as compared 3 Does not care about general Cares about generai public
we will Scale O
the scores on Therefore, if we have to choose between i and j,
statement i.
public
statements will be
choose statement no. j. From this we can conclude that only those in R&D
the 100 4 Leads in R &D Laggingbehind
selected which have a very high correlation with the total score. Therefore,
correlations are to be arranged in the ascending order of magnitudes corresponding 5Not a good paymaster A good paymaster
Products don't go through
correlation with the total Products go through
to each
statement and only top 25 statements having a high stringent
quality test quality test
score need to be selected.
Does a remarkable job in
Another method of selecting the number of statements from a relatively large Does nothing to curb
This aspect will be covered at curbing pollution
number of them is the use of
through factor analysis. pollution
Does not care about Cares about community
the appropriate stage in the chapter on factor analysis.
community near plants near plants
in
Not advisable to invest
TABLE 7.6 Scores of Statements 9Company stocks good to company stock
Total score and 100 Total Score Duy
Resp.No. 12 3 ***| ****
**********

10 Does not have good labour oooooo a Has good labour relations
individua SCOre i 410
*********
*********J relations
each respondent On
209
vafious Ststeme TS
. . 4 ...nee. 2 ***** Once the scale is constructed and administered to the representative respondents,
***********|
********.
****
the mean score for each of the descriptor is calculated. The scale is administered
to the response categories
********** | *******" under the assumption that the numerical values assigned
* *** ARssn**
***
are of interval scale in nature. This is generally the practice adopted
by many
***** researchers. However, if the response categories are treated as
ordinal scale, instead
*********| ae**

of computing the arithmetic mean, median may be computed.

In this example, we
********

**********.
are treating the responses as the interval scale
and hence the mean is computed.
Once the mean for all the bipolar adjectives/phrases is computed
we
put the result
****** ** ** ****
At this time, all
in the form of a pictorial profile so as to make the comparison easy.
100 unfavourable descriptors
the favourable descriptors are kept on one side and all the
***********
* ****"
 186 Research Methodology
Attitude Measurement and Scaling 187

TABLE 7.8 Makes quality products

Does not make The Stapel scale overComes this problem by using only single adjectives. This scale
PiCtorial profile based quality
products generally has 10 categories involving numbering -5 to +5 without a neuiral point and
on semantiCiterential
2 Leader in technology is usually presented in a vertical form. The job of the respondent is to indicate how
1atiny Backwardin accurately or inaccurately each term describes the object by selecting an appropriate
echnology numerical response category. Ifa positive higher number is selected by the respondent,
Cares about general
Does not care about it means the respondent is able to describe it more favourably. Suppose a restaurant is
public
jeneral public to be evaluated on quality of food and quality of service, then the Stapel scale would
4 Leads in R & D be presented as shown on the previous page:
Lagging behind in R&D
Inthe above scale, the respondents are asked to evaluate how accurately each
5 A good paymaster word or phrase describes the restaurant in question. They
will choose a value of +5 if
Not a
good paymaster describe at
the restaurant very accurately describes the attribute and -5 if it does not
chosen his options as
6 Products go through all correctly the word in question. Suppose a respondent has
Products do not go the respondent slightly prefers the quality of food and
stringent quality test indicated by". This shows that
through quality test useless.
Done remarkable job is of the opinion that the quality of service is totally
Done nothing to curb
in curbing pollution semantic diferential scale.
pollution |1. Distinguish between the Likert scale and
8 Cares about List the various forms of presenting the scales.
Does not care about
community near plants When is a Stapel scale used?
community near plants
9Companystocks good Not advisable to invest
to buy
in company stock MEASUREMENT ERROR
10 Has good labour
Does not have good when the observed m e a s u r e m e n t on a construct
relations labour relations
Measurement error occurs
of
LEARNING O8JECTIVE 5 true values. The following is a list of the
sources
or concept deviates rom its
CompanyA --- - - - . CompanyB Define m e a s u r e m e n t
measurement errors.
eror and explain
There are factors like mood, fatigue
and health of the respondent which
are positioned at the other. In
our example, we have
positioned all the favourable fhe criteria for good
may influence the
observed response while the instrument is being
descriptors for the two companies whose image we want to compare on the left hand measurement.
administered.
side. This is shown in Table 7.8. m e a s u r e m e n t s are taken may
As per the results presented in the The variations in the environment in which
pictorial profile, Company A is better than also result in a departure from the true value.
Company B in the sense that it makes quality products, leads in R&D, its products not understand the question
There are situations when a respondent may
80 through stringent quality tests, its stocks are good to buy and it has good labour have to rephrase the same. While
relations. Company B is ahead of Company A as it cares about general public and is being asked and the interviewer may
interviewer's bias may get into the responses.
a good paymaster. Company A is a better than Company B as it is leads in technology rephrasing the question the
survey, personal
whereas Also how the questionnaire is administered (telephone
Company B is better than Company A as it has done a remarkable job in or mail survey) will have its own impact on

curbing pollution. However, these differences are not statistically signiticant. interview with questionnaire
Stapel scale is used to Stapel scale: The Stapel scale is used to measure the direction and intensity of an the responses.
may be ambiguous
measure the direction ànd attitude. At times, it may be difficult to use semantic differential scales because of the At times, some
of the questions in the questionnaire Both of
for the respondents to understand.
intensity of an attitude. problem in creating bipolar adjectives. and some may be very difficult rise to
correct response, thereby giving
them can cause deviation from the

measurement error.
RESTAURANT of
at the time of coding, entering
5 +5 A t times, the errors may be committed
on the computer and at
the
-4 +4 data from questionnaire to the spreadsheet
3 +3 tabulation stage. be equal to the true
2 The observed measurement in any research need not
+1 measurement can be written as
measurement. The observed
Quality of Food Quality ofService O=T+S+R
2 O = Observed measurement
where,
3 3 T = True score
4 S = Systematic error
5 5* R= Random error
188 Research Methodology
Attitue
ude Measurement and Scaling189

It may be noted that the total error Cousisis Ot two

components-systematic average of all possible split-half reliabilities for a multiple item scale. This coeficient
error and random error. Systematic error causes a constant bias in the demonstrates whether the average score of all split-half of reliabilities converge to a
measurement
Suppose there is a weighing scale that weigils du gm less for every one kg of certain point or not.
product
being weighed. The error would consisternuy remain the same irrespective of the kind The coefficient alpha does not address validity. However, many researchers
use

can take values between 0

of product and the time at which productlS wegnea. Kandom error on the other hand this as a sole indicator of validity. The alpha coefticient
their interpretations are suggested below
involves influences that bias the values of alpha with
The randon error on measlrements but are not systematic. Suppose we and 1. The following
use different weighing scales to weigh One kg
the other hand involves Ota product and if systematic error is fhere is no consistency between the various
assumed to be absent, we may find that recorded weights may fall within a = 0 means
influences that bias the range items of a multiple item scale
around the true value of the weight, thereby causing random error. There is complete consistency between
measurements but are not = I means

systematic. various items of a mulüple item scale

Criteria for Good Measurement 0.80 s a s 0.95 implies There is very good reliability between the

There are three criteria for evaluating measurements: reliability, validity and various items of a multiple item scale
There is good reliability between the various
0.70 a s 0.80 implies
sensitivity. items of a multiple item scale
There is fair reliability between the various
0.60 sas0.70 implies
Reliability items of a muitiple item scale
Reliability is concerned with consistency, accuracy and predictability of the scale. It There is poor reliability between the various
a< 0.60 means
refers to the extent to which a measurement process is free from random errors. The items of a multiple item scale
reliability of a scale can be measured using the following methods:
Test-retest reliability: In this method, repeated measurements of the same person
In the test-retest
or group using the same scale under similar conditions are taken. A very high Validity
reliability, repeated what we
correlation between the two scores indicates that the scale is reliable. However, the
The validity of a scale refers The validity scale refers to the question whether we are measuring
of a
measurements ot the same
to which the measurement
person or group using the following issues should be kept in mind before arriving at such a conclusion. to the question whether we want to measure. Validity of the scale refers to the extent
of scale is a
What should be the appropriate time difference between the two both systematic and randon errors. The validity a
same scale under the simiar are measuring what we want process is free from
observations is a question which requires attention. If the time difference
more serious issue than reliability.
There are different ways to measure validity.
Condition are taken. to measure
between two consecutive observations is very small(saytwo or three weeks) involves subjective judgement by
remember the Contentvalidity: This is also called face validity. It
it is very likely that the respondents would previous answer
of the construct. For example, to measure

the same answer when the instrument is administered the

Content val dity is also an expert for assessing the appropriateness
and may give called face validity in vhich the perception of a customer towards Jet Airways,
a muliple item scale is developed.
second time. This will make the instrument reliable, which may not actually the
These items when combined in an index
measure
an expert provides subjective A set of 15 items is proposed.
be the case. However, if the difference between the two observations is very
15 items, a set
judgement to assess the
perception of Jet Airways. In order to judge the
content validity of these
more than a year) it is quite likely that the respondent's
answers
large (say appropriateness of the of the 15 items. The
to the various questions of the instrument might have actually undergone of experts may be requested to examine the representativeness
construct.
items covered may be lacking in the content validity
if we have omitted behaviour of
a change, resulting in poor reliability of the scale. Therefore, the
researcher
the crew, food quality, and food quantity, etc.,
from the list. In fact, conducting the
the two
has to be very careful in deciding upon the time difference between perception of the airline
of about five to exploratory research to exhaust the list of items measuring
observations. it is thought that a time difference
Generally, would be of immense help in such a case.
six months is an ideal period.
of the new measuring
Another problem in this test is that the first measurement may change the Concurrent validity: It is used to m e a s u r e the validity
established techniques. It involves
response of the subject to the second measurement. techniques by correlating them with the
m e a s u r e s of the same phenomena (for
The situational factors working on two different timeperiods may not be computing the correlation coefficient of two
of a company) which are administered
the same, which may result in different measurement in the two periods.
example, perception of an airline and image ot Jet
the perception
T h e second reading on the same instrument from the same subject may at the time. We prepare a 15 item scale to measure
same
a researcher proposes an
produce boredom, anger or attempt to remember the answers given in
an
Airways, which is assumed to be a valid one. Suppose
of the new technique
inítial measurement. alternative and shorter technique. The concurrent validity
between the two techniques when
A favourable response with a brand during the period between the two tests would be established if there is a high correlation
identical conditions.
might cause a shift in the individual rating by the subject. administered at the same time under similar or
a measured phenomena at
one point
Ahigh correlation Split-half reliability method: This method is used in the case of multiple item Predictive validity: This involves the ability of
at a future point of time. If the
correlation
indicates that the internal scales. Here the number of items is randomly divided into two parts and a correlation of time to predict another phenomenon
initial measure is said to have a high
consistency of the construct coefficient between the two is obtained. A high correlation indicates that the internal coefficient between the two is high, the
consider the use of the common admission
eads to greater reliability. test
consistency of the construct leads to greater reliability. Another measure which
predictive ability. As an example, in a business
is used to test the internal consistency of a multiple item scale is the coefficient candidates for admission to the MBA programme
to shortlist
(CAT)
alpha (a) commonly known as cronbach alpha. The cronbach alpha computes the