Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 1

The Likert scale revisited: an alternate version.(product preference testing)

by Gerald Albaum

An investigation of the effect of alternative scale formats on intensity of attitudes report based on
the Likert scales of agreement was conducted. The study utilized both a one-stage format and a
two-stage alternate format in three separate experiments on sample population of college
students from three different nations. The two-stage format, which generated the greatest amount
of extreme-position response, was determined to be a better gauge of product preferences.

© COPYRIGHT 1997 NTC Publications Ltd.

The Likert scale revisited: an alternate version

Marketing researchers have available for their use in measuring attitudes or components of attitudes a number of different
types of scales or scale formats, including the semantic differential, Stapel scale, Likert scale, Thurstone differential scale,
and direct rating scales. One of the most widely used formats is the Likert scale.

The purpose of this article is to examine the Likert scale in terms of its ability to detect the intensity of feeling that
respondents have about their attitudes. An alternative format is proposed and tested against the standard approach.

Nature of the Likert scale

Attitude is a complex thing. Recent theoretical models of attitudes suggest that there are two dimensions - direction (e.g. a
positive or negative predisposition toward an object) and strength (Petty & Krosnick in press; Raden 1985). For example,
a person may like or dislike a product and may hold this attitude with a varying degree of strength or intensity - interpreted
as either confidence, certainty, accessibility or conviction (Berger & Alwitt 1996). Abelson (1988) distinguishes between
firmly held attitudes and those that are more superficial by utilising the idea of attitude conviction. Berger & Alwitt (1996)
propose that conviction is a subjective component of an attitude which gauges the extent to which the attitude is ’owned’
or firmly held and reflects the functions the attitude serves. Our concern in this paper is with the extent to which an attitude
is firmly held. Consequently, we equate conviction with strength and intensity.

When a Likert scale is used to measure attitude, its usual or standard format consists of a series of statements to which a
respondent is to indicate a degree of agreement or disagreement using the following options: strongly agree, agree,
neither agree nor disagree, disagree, strongly disagree. As such the scale purports to measure direction (by
’agree/disagree’) and intensity (by ’strongly’ or not) of attitude. The scale, per se, was intended as a summated scale,
which was then assumed to have interval scale properties (Likert 1932). This level-of-measurement characteristic together
with ease of administration and response explains its popularity in marketing research applications. The individual scale is
not assumed to be intervally measured although it usually is treated as such. In practice, the scale is often used by
researchers in marketing as individual scale items or as a summated scale based on a small number of scale items (as
few as two or three).

According to Tourangeau (1984; 1992) with some attitude questions a person must compute an evaluative judgement
whereas with others such a judgement is simply retrieved. The form of the attitude question may also influence whether a
respondent employs one process or the other. A Likert-type item, because it requires a person to rate extent of
agreement, may encourage the retrieval and integration of more detailed information from memory than do items calling
for a simple evaluation. In a real sense we can view an opinion as a verbal expression of an attitude which means that
opinions are the means we have for measuring attitudes.

The standard Likert scale tends to confound the direction and intensity dimensions of attitude so there may be an
under-reporting of the most intense agreement or disagreement (i.e. the extreme position of the scale). When looking at
extreme positions, we are dealing with the furthermost categories of the response set alternatives and not extremeness in
amount of direction. In the Likert scheme the furthermost categories represent the most intensity of response. In an
analysis-of-variance context the standard Likert scale measures directly the interaction and indirectly, the main effects of
direction and intensity. In short, main effects are inferred from the interaction measure.

An alternative way of presenting the Likert scale is as a two-stage scale (Mager & Kluge 1987). The first stage asks
- Reprinted with permission. Additional copying is prohibited. - GALE GROUP
Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 2

The Likert scale revisited: an alternate version.(product preference testing)

whether there is agreement or disagreement with the statement. The second stage then asks how strongly the person
feels about the answer provided in the first stage. In a more general sense, the split question technique has been called
’unfolding’ and has been used in telephone surveys as an alternative to the use of ’show cards’ in face-to-face interviews
(Groves 1979; Groves & Kahn 1979; Miller 1984; Sykes & Collins 1988). The end result of this split question process for
the Likert scale is that the direction dimension of attitude is being addressed by the first stage and the intensity (or
strength) dimension is measured directly by the second stage. Interaction effects can then be measured. This process
comes closer to the usual ANOVA approach. In addition, Berger & Alwitt (1996) present evidence that attitude strength
and direction are unique separable dimensions of an attitude. It is this two-stage approach that is examined by the study
reported in this paper (see figure 1).

What is known about the methodological characteristics of the Likert scale?(*) Jacoby & Matell (1971) examined the issue
of number of response alternatives to be provided. Since there was no effect on the use of categories these researchers
concluded that as little as two or three categories could be used. Such scales are sufficient to meet criteria of test-retest
reliability, concurrent validity, and predictive validity (Jacoby & Matell 1971). A major result of reducing the categories in
this manner is that only direction is provided. Matell & Jacoby (1972) raised a broader question about optimal number of
categories while Hulbert & Lehmann (1972) argued that the goals or objectives underlying use of scales should dictate the
number of scale categories. A related issue is that of what response categories should be used. Spector (1976) was
concerned with the interval-level measurement issue and attempted to derive scale values for alternative response
categories. In contrast, Wyatt & Meyers (1987) examined four different ways to label five-point response scales. They
found no difference in mean values but the scales did differ on variability. A similar finding emerged from a study by Dixon
et al (1984).

Another aspect studied was that of the neutral point (Komorita 1963; Guy & Norvell 1977; Garland 1991). There could be
a sensitising effect if a respondent expects to find a neutral position and does not find one. Since composite scores tend
not to be affected by inclusion, it has been argued that a neutral position should always be provided. In another study,
Armstrong (1987) compared different ways to express the midpoint - undecided or neutral. Analysis showed differences
were negligible and little if any erosion of score appeared to result. A neutral position is one that is neither agree nor
disagree, and does not include the categories ’don’t know’ and ’no opinion.’

As with any scale and scale format there is concern for systematic errors, one type of which is form-related (Bardo,
Yeager & Klingsporn 1982; Bardo & Yeager 1982; Phelps et al 1986; Greenleaf 1992b). Form-related errors concern
psychological orientation towards responding to different item formats and include the following types:

* Leniency: tendency to rate something too high or too low (i.e. rate in an extreme way).

* Central tendency: reluctance to give extreme scores.

* Proximity: give similar responses to items that occur close to one another.

The present study will allow for an appraisal of whether the form-related potential errors might be ’generic’ for the Likert
scale and Likert-type data.

Directly related to the present study is the issue of direction and extremeness. Indeed, a question can be raised as to
whether there is a difference between extremeness and intensity in using a Likert scale. Converse & Presser (1986) have
argued that the Likert scale confounds extremity (a dimension of attitudinal position) with intensity (how strongly a position
is felt). Although intensity and extremity may frequently covary, a person may hold an extreme position with little feeling,
or have a middle of the road position with considerable passion; without separate questions for positions and intensity it is
difficult, if ever possible, to separate these dimensions. Peabody (1962) has argued that there are both theoretical and
practical reasons for considering the components of direction and extremeness (i.e. intensity) separately. Their research
did not deal specifically with the Likert scale. Similarly, Compeau & Franke (1990) explored direction and intensity in the
broader category rating scale. Both sets of researchers agree that composite scores (e.g. that from a Likert scale) do not
reflect the intensity dimension well, and separating the direction and intensity components may allow the researcher to
explain more variance than the composite score where the two dimensions are combined. Furthermore, separating the
components may assist the researcher in interpreting how the respondents used the measure (Compeau & Franke 1990).
Applying the technique to dual scaling - an approach to correspondence analysis - has been suggested as a way to

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 3

The Likert scale revisited: an alternate version.(product preference testing)

separate intensity and direction (Compeau & Franke 1990; Franke 1985). The present paper presents another
approach.(**) It must be remembered that in the Likert Scale format extremity and positions on the scale measure
intensity of feeling about direction and not amount of direction. Amount of direction in attitude has to be captured by the
statement itself.

More recent research dealing specifically with the Likert scale and Likert-type data involves adaptive survey designs
based on a graded-response Latent, Trait Theory model (Singh, Howell & Rhoads 1990). Operating like an expert system,
an adaptive survey design adapts the questions asked to each respondent based upon responses to previous questions.
In short, an adaptive survey asks different questions of different respondents, but is able to estimate all respondents’
locations, along the same attitude continuum. Because Likert-type data provide much information, a smaller number of
items may be used for a design based on the graded-response model than for other survey designs using the scale.
These researchers suggest efficiencies of the magnitude of 33% compared with conventional survey designs (Singh,
Howell & Rhoads 1990, p. 320). Finally, Russell & Bobko (1992) examined how characteristics of a Likert-type scale
affected the power of moderated regression analysis. The response scale was not a true Likert-scale as the categories
were of motivation towards some future behaviour not categories of agreement with statements.

As mentioned earlier, the present study looks primarily at the issue of confounding of attitude components and the
reporting of intensity (or extreme positions available) by the scale. It is hypothesised that the standard one-stage Likert
scale tends to under-report the most intense position (thus, enhancing central leniency errors), compared with a two-stage
format. The specific hypothesis was suggested by the results of a small exploratory study of these alternative scale
formats for the Likert scale (Albaum & Murphy 1988). In most cases, a two-stage approach generated a greater
percentage of most intense position values than did standard and modified one-stage versions.

Method

Three distinct studies were conducted to generate the data needed to test the overall hypothesis. Each study was set up
as a simple experiment using a completely randomised design. Two treatment levels were used as shown previously in
figure 1. Data were collected in different countries at different time periods over several months.

The topic of Study 1 was economic systems. Separate convenience samples of business students were drawn from two
different universities in the United States ([N.SUB.I] = 64, [N.SUB.II] = 97), one university in New Zealand (N = 121), and
one university in Denmark (N = 98). The sample in Denmark consisted of part-time students, each of whom is employed
full time. This study was done in three countries to explore whether each scale type tends to be culture-bound (an emic)
or culture-free (an etic). Independent of the absolute values that emerge, the Likert scale will possess eric properties as a
scale type if results of one-stage and two-stage comparison are similar within each country.

The Likert-scale instrument for the United States sample included 11 statements about capitalism and different economic
systems. For both the Denmark and New Zealand samples, two additional statements were added giving a total of 13
statements. These two statements were added because of the interest in them by associates in the countries who
assisted in data collection. One statement dealt with communism and resource allocation efficiency; the other statement
was about size of firm and private enterprise as capitalism. For the most part, the final measurement instruments were the
same in each country. The instrument used in Denmark was presented to respondents in Danish after two rounds of
translation and back translation.

To remove doubt that might arise about a topic/scale-format interaction Study 2 was designed around a topic that is quite
different from that used in Study 1. The overall topic of Study 2 was alienation. A convenience sample of 176 students in
New Zealand responded to a set of 37 statements about consumer alienation from the marketplace (Allison 1978). As in
Study 1, a completely randomised experimental design with two treatment levels (one-stage and two-stage Likert scales)
was used in Study 2.

The results of the first two studies were such that a third study was needed to assess which of the formats was ’best’.
Study 3 was designed around the topic of consumer ethnocentrism. A sample of 50 students in the United States
responded to the 17 statements that comprise the CETSCALE (Shimp & Sharma 1987). This scale is designed to
measure consumers’ ethnocentric tendencies related to purchasing foreign- versus American-made products. Although
originally developed and validated with samples of US consumers only, other research has shown that the CETSCALE is

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 4

The Likert scale revisited: an alternate version.(product preference testing)

a reliable measure across other countries and affords some evidence of validity as well (Netemeyer, Durvasula &
Lichtenstein 1991). In addition to the CETSCALE items, a question was included asking for preferences regarding
purchasing nine consumer durable products ranging from electronics to athletic shoes. A nine-category scale was used,
anchored with ’American-made’ and ’foreign-made.’ As in the previous two studies, a completely randomised
experimental design with two treatment levels was used in Study 3.

Results

For all three studies responses to the one-stage and two-stage Likert scales were compared at the level of the individual
scale. Although the Likert Scale is supposed to be used as a summated scale, it often is not used that way by academic
and practitioner marketing researchers. Individual scale item scores are used for analysis, as are average scale values for
the scales comprising a construct. In addition, many reported summated scores consist of as few as two or three scale
items. As we pointed out earlier there is evidence in the literature that composite scores do not reflect the intensity
dimension well. On a more practical level, there are times when researchers are interested in groups - such as market
segments - having intense beliefs about an object of concern. Often, this is best picked up at the individual scale level.
The two-stage version was scored to be comparable with the one-stage version. That is, a respondent who answered the
first stage with ’agree’ or ’disagree’ and the second stage with ’not very strong’ was scored the same as one who
indicated ’agree’ or ’disagree’ to the one-stage version. Similarly, a respondent who answered the second stage with ’very
strong’ was scored the same as one who answered ’strongly agree’ or [TABULAR DATA FOR TABLE 1 OMITTED]
[TABULAR DATA FOR TABLE 2 OMITTED] ’strongly disagree’ to the one-stage version. The analysis was of most
intense (i.e. extreme scale position), values only, i.e. ’strongly agree’ and ’strongly disagree’.

For Study 1 the absolute values of the proportion of subjects in each sample who responded ’strongly agree’ or ’strongly
disagree’ varied across scales and samples, as shown in tables 1 and 2. For both United States samples and that in New
Zealand, on an absolute basis without exception there were a greater number of extreme position (most intense)
responses (agree and disagree) from the two-stage format. Another possible type of analysis would be to compute mean
values for each version and then test for significant differences using the Student t test. Mean values tend to ’hide’
differences such as those for extreme intensity values. In fact, for Study 1 the number of differences within three of the
four sample groups were within chance expectations based on the binomial distribution. Similar results emerged for the
Mann-Whitney nonparametic test for distribution. Our concern is with a scale format’s ability to detect people with extreme
intensity responses. In Denmark, there were only two instances, one each for ’agree’ and ’disagree’, where the one-stage
generated a greater response frequency than did the two-stage format.

Turning now to Study 2, as stated earlier the general topic of this study was consumer and social alienation. A total of 37
Likert scales (see table 3) were presented to the sample of business students in New Zealand. The basic findings are
presented in table 4. As shown, the absolute values varied across scales and scale types. For all but one of the 74
possible comparisons there were a greater number of extreme position (intensity) responses from the two-stage format.

For Study 3 the absolute values of the proportion of subjects who responded ’strongly disagree’ varied widely across
scales, as shown in table 5. Similar variation occurred for subjects responding ’strongly agree’ in the two-stage treatment
condition. However, there was little variation for ’strongly agree’ in the one-stage condition. Of the 34 possible
comparisons there was only one where the one-stage version elicited a greater extreme intensity response than did the
two-stage format.

Discussion and conclusions

On the basis of the three studies reported in this paper an overwhelming conclusion is that the two-stage Likert scale
elicited greater extreme position values than did the one-stage format. Although not all proportion differences were
statistically significant the vast majority were at p [less than] .05 and the number far exceeded chance binomial
probabilities. In light of this finding, it appears that Likert scales as generally used tend to underestimate the extreme
positions held by people, and that a central tendency forms-related error exists. Perhaps when faced with the standard
one-stage format respondents are reluctant to express an extreme position even though they have it. The two-stage
would give them the flexibility to express their true opinions. It is always possible that what has been observed is a
response style artifact, i.e. choosing a response category a disproportionate number of times independently of the
question content. This would mean that extreme response style behaviour would be operating for the two-stage format.

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 5

The Likert scale revisited: an alternate version.(product preference testing)

Since the absolute proportions vary so much across scale items as do the differences between the stages, this is not a
problem. Had it been, response frequencies would have been more or less uniformly the same across scales and
countries. On the basis of these findings, particularly those from Study 1, it seems reasonable to suggest that the Likert
scale format tends to have etic properties.

Table 4

Extreme responses to alienation scales (Percent of respondents)

Strongly disagree Strongly agree

One- Two- One- Two-
Scale(a) stage(b) stage(c) p stage(b) stage(c) p

1. 7.8 20.6 .01 4.9 12.3 .02

2. 22.3 50.7 .01 3.9 6.9 .22
3. 3.9 5.5 .48 19.4 34.2 .01
4. 2.9 8.2 .03 13.6 41.1 .01
5. 6.8 19.2 .01 8.7 27.4 .01
6. 13.6 28.8 .01 3.9 15.1 .01
7. 12.6 30.1 .01 4.9 15.1 .01
8. 1.9 8.2 .01 10.7 21.9 .01
9. 12.6 26.0 .01 9.7 11.0 .69
10. 21.4 56.1 .01 5.8 11.0 .08
11. 2.9 5.5 .23 12.6 42.5 .01
12. 25.2 32.9 .11 7.8 19.2 .01
13. 40.8 60.3 .01 1.9 4.1 .23
14. 3.9 1.4 .15 18.4 50.7 .01
15. 2.9 4.1 .55 27.2 49.3 .01
16. 5.8 28.8 .01 2.9 8.2 .03
17. 14.6 19.2 .25 3.9 12.3 .01
18. 6.8 12.3 .08 4.8 24.7 .01
19. 12.6 26.0 .01 1.9 13.7 .01
20. 13.6 32.9 .01 3.9 11.0 .02
21. 7.8 26.0 .01 2.9 4.1 .54
22. 40.8 71.2 .01 1.9 13.7 .01
23. 3.9 9.6 .04 11.6 34.2 .01
24. 6.8 11.0 .17 16.5 34.2 .01
25. 1.0 26.2 .01 26.2 67.1 .01
26. 15.5 30.1 .01 2.9 11.0 .01
27. 9.7 15.1 .13 10.7 37.0 .01
28. 4.8 6.8 .43 7.8 26.0 .01
29. 33.0 52.0 .01 1.9 11.0 .01
30. 9.7 28.8 .01 5.8 11.0 .08
31. 4.8 11.0 .03 10.7 24.7 .01
32. 10.7 20.5 .02 1.0 9.6 .01
33. 5.8 23.2 .01 2.9 9.6 .01
34. 10.7 13.7 .39 5.8 27.4 .01
35. 13.6 27.4 .01 0.0 1.4 .12
36. 13.6 20.5 .09 3.9 24.7 .01
37. 9.7 13.7 .25 1.9 15.1 .01

a See table 3 b N = 103 c N = 73

[TABULAR DATA FOR TABLE 5 OMITTED]

Another interpretation of the one-way direction of the results is that the one-stage format is satisfactory and it is the
two-stage format that over-reports extreme values and that the Likert scale tends to show leniency error. This seems
unlikely as the judgements made about direction and intensity are independent judgements. Confounding of the direction
and intensity states does not occur. It is this confounding and the interaction of direction and intensity that leads to error.

This issue can be resolved by assessing which format is ’best’ in the sense of predicting preferences and/or behaviour.
Study 3, the CETSCALE, included a question asking subjects to indicate the extent of their preference for foreign-made or

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 6

The Likert scale revisited: an alternate version.(product preference testing)

American-made products if they were to be purchasing nine products. The products, which included athletics shoes,
automobiles, and a number of consumer electronic products were available from both American and foreign sources. A
product preference score was created for each subject by summating their responses to a nine-point preference scale for
the nine products. The higher the score the greater the preference for foreign-made products. Similarly, an ethnocentrism
score was computed for each subject by summing responses to the 17 items of the CETSCALE. The greater the score
the more ethnocentric a person is. Correlations were run between ethnocentrism and preference scores with the following
results:

One-stage: r = -.02 Two-stage: R = -.55

For the reduced version of the CETSCALE (only 10 items used), the correlations are:

One-stage: r = -.21 Two-stage: R = -.70

All correlations are in the expected direction as there should be a negative relationship between GETS GALE score and
preference as scored. It is quite clear that the two-stage format was a better predictor in Study 3.

How does the proposed two-stage Likert scale affect other uses of the data, such as in a multivariate analysis? To
examine this question the data for each study were factor analysed using an eigenvalue = 1 as the criterion for extraction
of factors. Varimax rotation was used. For Study 1 the two USA sample groups were combined. The number of factors
extracted, the percentage of total variance these factors explain, and the reliability of this analysis are shown for all
sample groups in both studies in table 6. There is no pattern to these results, and it appears that the one-stage and
two-stage formats are not noticeably different. Reliability was assessed by coefficient theta (0), which is based on the
number of items factor analysed and the largest (i.e. the first) eigenvalue. In effect, theta is a special case of Cronbach’s
alpha (Carmines & Zeller 1979, pp 60-1).

Different ways of measuring intensity may lead to different results. However, to be comparable, the measure should be as
similar as possible to that of the standard Likert scale. The approach used in this study meets this criterion. Of course, it
may be that the two-stage format measures extremeness of direction not intensity. This is an issue brought up by
Converse & Presser (1986), who argue that extremity is confounded with intensity, and that extremity is a dimension of
attitudinal position. This issue applies to the standard Likert scale as well. However, the wording used in the two-stage
format would suggest that an intensity of attitude was being provided. In contrast, it is possible that respondents could
interpret the confounded categories in the one-stage format (strongly agree, agree, etc.) as amounts of agreement. This
would mean that ’strongly agree’ and ’strongly disagree’ were measuring extremeness of direction, not intensity.

Table 6

Factor analysis characteristics

Number of Percent variance Reliability

Study factors explained ([Theta])

Study 1

USA:

One-stage 4 66.0 .73

Two-stage 5 71.8 .78

Denmark:

One-stage 5 66.3 .64

Two-stage 5 75.4 .70

New Zealand:

One-stage 5 71.1 .75

Two-stage 5 67.3 .75

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 7

The Likert scale revisited: an alternate version.(product preference testing)

Study 2

One-stage 13 73.3 .76

Two stage 12 79.6 .75

Study 3

One-stage 5 79.6 .91

Two stage 5 74.9 .89

This study has been limited in that students were used as experimental subjects. There are no inherent reasons why
students should react differently from the different formats than other groups. Consequently, in order to assess more
accurately whether the scaling methods are eric- or emic-bound, further research using non-students, more countries and
different topics would seem to be in order.

The data in this study suggest that the two-stage format is a better predictor of preferences. In addition, we have issued a
warning that researchers need to be aware that reporting extreme positions on a Likert scale can vary by scale format.
The next step would seem to be testing the two-stage format on actual behaviour. This would represent a so-called
’building-block’ approach to research. As such then the present study represents a foundation of the knowledge-building
process, albeit a major foundation. The study also raises questions about whether the phenomenon observed exists for
other scale formats as well, such as the semantic differential and rating scales in general.

One inherent characteristic of the two-stage approach to Likert-scale use is that the length of the measurement instrument
will be increased, leading perhaps to greater costs in implementing research projects. Respondents may have to commit
more time to responding. This is most likely to be true for surveys using mail questionnaires. For personal interview
surveys it may or may not be true. For a telephone interview, we suggest that it would be easier for an interviewer to use
the two-stage Likert format. Indeed, as we mentioned earlier two-stage questioning (i.e. unfolding) is used in telephone
surveys for various types of question and scale formats. Studies that have compared one-stage and two-stage versions of
telephone interviewing have been inconclusive regarding differences between the two. This is probably due in part to the
fact that the studies were not comparable in that different types of questions and formats were studied. It is interesting to
note, however, that in one of the experiments conducted by Sykes & Collins (1988), the average response to 10 Likert
scales showed that the unfolded version generated a greater response percentage in the two ’strongly’ categories than
did the standard one-step version. However, the main justification for using the two-stage, rather than the one-stage,
format is that for those whose major interest is in most intense (i.e. extreme position) views the two-stage provides higher
data quality! After all, is it not data quality that should be guiding researchers?

Figure 1

Formats for the Likert-type scale

I. One-stage (the traditional format)

Please indicate your degree of agreement with each of the following statements by circling one of the symbols beneath
the statement where

SA = strongly agree A = agree N = neither agree nor disagree D = disagree SD = strongly disagree NO = no opinion

’A product’s price will usually reflect its level of quality.’

SA A N D SD NO

II. Two-Stage

For each of the statements listed below indicate first the extent of your agreement and second how strongly you feel about
your agreement.

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 8

The Likert scale revisited: an alternate version.(product preference testing)

’A product’s price will usually reflect its level of quality.’

_____agree

_____neither agree nor disagree

_____disagree

_____no opinion

How strongly do you feel about your response?

_____very strong

_____not very strong

Table 3

Scales used for alienation study

1. Most companies are responsive to the demands of the consumer.

2. It seems wasteful for so many companies to produce the same basic product.

3. Unethical practices are widespread throughout business.

4. Retail stores do not care why people buy their products just as long as they make a profit.

5. Shopping is usually a pleasant experience.

6. People are unable to help determine what products will be sold in the store.

7. Advertising and promotional costs unnecessarily raise the price the consumer has to pay for a product.

8. What a product claims to do and what it actually does are two different things.

9. Mass production has done away with unique products.

10. Misrepresentation of product features is just something we have to live with.

11. Harmful characteristics of a product are often kept from the consumer.

12. It is embarrassing to bring a purchase back to a retail store.

13. I tend to spend more than I should just to impress my friends with how much I have.

14. Even with so much advertising it is difficult to know what brand is the best.

15. A sale is not really a bargain but a way to draw people into a retail store.

16. It is difficult to identify with current trends and fads in fashion.

17. I often feel guilty for buying so many unnecessary products.

18. Most brands are the same with just different names and labels.

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 9

The Likert scale revisited: an alternate version.(product preference testing)

19. A product will usually break down as soon as the warranty is up.

20. Business is responsible for unnecessarily depleting our natural resources.

21. It is difficult to identify with business practices today.

22. One must be willing to tolerate poor service from most retail stores.

23. It is difficult to know what store has the best buy.

24. Business’s prime objective is to make money rather than satisfy the consumer.

25. I often feel frustrated when I fail to find what I want in the store.

26. After making a purchase I often find myself wondering ’why.’

27. It is hard to understand why some brands are twice as expensive as others.

28. It is not unusual to find out that business had lied to the public.

29. Buying beyond one’s means is justifiable through the use of credit.

30. It is often difficult to understand the real meaning of most advertisements.

31. Products are designed to wear out long before they should.

32. Most claims of product quality are true.

33. I am often dissatisfied with a recent purchase.

34. The wide variety of competing products makes intelligent buying decisions more difficult.

35. Advertisements usually present a true picture of the product.

36. A product’s price will usually reflect its level of quality.

37. What other people think will often influence what I buy.

* Our concern is with specific characteristics of the Likert scale. There is a body of research that has compared the Likert
scale with other types of scales such as Semantic scales (Ofir et al 1987), direct ranking scales (Conrath, Montazemi &
Higgins 1987), Thurstone scales (Flamer 1983) and the Guttman scale (Byrne, 1987).

** On a more general level, Greenleaf (1992a) proposes a method for creating, validating and scoring extreme response
style measures.

The author thanks Jesper Strandskov, Aarhus School of Business (Denmark) and Brian Murphy, Auckland Institute of
Studies (New Zealand) for their assistance in data collection. The comments made by Paul Green, Wharton School, Frank
Carmone, Wright State University, Robert Peterson, University of Texas at Austin, and the anonymous referee are greatly
appreciated.

References

ABELSON, ROBERT P. (1988). Conviction. American Psychologist, 43, 4, pp 267-75.

ALBAUM, GERALD & MURPHY, BRIAN D. (1988). Extreme response on a Likert scale. Psychological Reports, 63, pp

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 10

The Likert scale revisited: an alternate version.(product preference testing)

501-2.

ALLISON, NEIL K. (1978). A psychometric development of a test for consumer alienation from the marketplace. Journal of
Marketing Research, 15, pp 565-75.

ARMSTRONG, R. L. (1987). The midpoint on a five-point Likert-type scale. Perceptual and Motor Skills, 64, pp 359-62.

BARDO, J. W. & YEAGER, S. J. (1982). Consistency of response style across types of response formats. Perceptual and
Motor Skills, 55, 1, pp 307-10.

BARDO, J. W., YEAGER, S. J. & KLINGSPORN, M. J. (1982). Preliminary assessment of format-specific central
tendency and leniency error in summated rating scales. Perceptual and Motor Skills, 54, 1, pp 227-34.

BERGER, IDA E. & ALWITT, LINDA F. (1996). Attitude conviction: a measure of strength and function. Unpublished
paper.

BYRNE, BARBARA M. (1987). Multitrait-multimethod analyses of three self-concept scales: testing equivalencies of
construct validity across ability. Paper presented at the annual meeting of the American Educational Research
Association.

CARMINES, EDWARD G. & ZELLER, RICHARD A. (1979). Reliability and validity assessment. Beverly Hills, CA: Sage
Publications.

COMPEAU, L. D. & FRANKE, G. R. (1990). Isolating and quantifying the intensity dimension of category rating scales.
Proceedings of the Annual Educators’ Conference of the American Marketing Association, pp 293-98.

CONRATH, D. W., MONTAZEMI, A. R. & HIGGINS, C. R. (1987). Evaluating information in ill-structured decision
environments. Journal of the Operational Research Society, 38, pp 375-85.

CONVERSE, JEAN & PRESSER, STANLEY (1986). Survey questions. Beverly Hills, CA: Sage Publications.

DIXON, PAUL N. et al. (1984). Response differences and preferences for all-category-defined and end-defined Likert
formats. Educational and Psychological Measurement, 44, pp 61-6.

FLAMER, STEPHEN (1983). Assessment of the multitrait-multimethod matrix validity of Likert scales via confirmatory
factor analysis. Multivariate Behavior Research, 18, pp 275-308.

FRANKE, G. R. (1985). Evaluating measures through data quantification: applying dual scaling to an advertising copytest.
Journal of Business Research, 13, pp 61-9.

GARLAND, RON (1991). The mid-point on a rating scale: is it desirable? Marketing Bulletin, 2, pp 66-70.

GREENLEAF, ERIC (1992a). Measuring extreme response style. Public Opinion Quarterly, 56, Summer, pp 328-51.

GREENLEAF, ERIC (1992b). Improving rating scale measures by detecting and correcting bias components in some
response styles. Journal of Marketing Research, 29, May, pp 176-88.

GROVES, ROBERT M. 1979). Actors and questions in telephone and personal interview surveys. Public Opinion
Quarterly, 43, pp 190-205.

GROVES, ROBERT M. & KAHN, ROBERT L. (1979). Surveys by telephone: a national comparison with personal
interviews. New York: Academic Press.

GUY, REBECCA F. & NORVELL, MELISSA (1977). The neutral point on a Likert scale. The Journal of Psychology, 95, pp
199-204.

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 11

The Likert scale revisited: an alternate version.(product preference testing)

HULBERT, JAMES & LEHMANN, DONALD R. (1972). Are three-point Likert scales always good enough? Journal of
Marketing Research, 9, pp 444-6.

JACOBY, JACOB & MATELL, MICHAEL S. (1971). Three-point Likert scales are good enough. Journal of Marketing
Research, 8, pp 495-506.

KOMORITA, S. S. (1963). Attitude content, intensity, and the neutral point on a Likert scale. Journal of Social Psychology,
61, pp 327-34.

LIKERT, RENSIS (1932). A technique for the measurement of attitudes. Archives of Psychology, 140, pp 44-53.

MAGER, JOHN & KLUGE, E. ALAN (1987). One-stage versus two-stage attitude measurement: separating the
dimensions of direction and intensity. Paper presented at the Academy of Marketing Science World Marketing Congress,
Barcelona, Spain.

MATELL, MICHAEL S. & JACOBY, JACOB (1972). IS there an optimal number of alternatives for Likert-scale items?
Journal of Applied Psychology, 56, pp 506-9.

MILLER, PETER V. (1984). Alternative question forms for attitude scale questions in telephone interviews. Public Opinion
Quarterly, 48, pp 766-78.

NETEMEYER, R. G., DURVASULA, S. & LICHTENSTEIN, D. R. (1991). A cross-national assessment of the reliability and
validity of the CETSCALE. Journal of Marketing Research, 28, August, pp 320-7.

OFIR, CHEZY et al (1987). Are semantic response scales equivalent? Multivariate Behavior Research, 22, pp 21-38.

PEABODY, D. (1962). Two components in bipolar scales: direction and intensity. Psychological Review, 69, pp 65-73.

PETTY, RICHARD E. & KROSNICK, JON A. (in press). Attitude strength: antecedents and consequences. Hillside, NJ:
LEA.

PHELPS, LEADELLE et al (1986). The effects of halo and leniency on co-operating teacher reports using Likert-type
rating scales. Journal of Educational Research, 79, Jan-Feb, pp 151-4.

RADEN, DAVID (1960). Strength related attitude dimensions. Social Psychology Quarterly, 48, pp 312-30.

RUSSELL, CRAIG J. & BOBKO, PHILIP (1992). Moderated regression analysis and Likert scale: too coarse for comfort.
Journal of Applied Psychology, 77, 3, pp 336-42.

SHIMP, T. A. & SHARMA, S. (1987). Consumer ethnocentrism: construction and validation of the CETSCALE. Journal of
Marketing Research, 24, August, pp 280-9.

SINGH, JAGDIP, HOWELL, ROY D. & RHOADS, GARY K. (1990). Adaptive designs for Likert-type data: an approach for
implementing marketing surveys. Journal of Marketing Research, 27, pp 304-21.

SPECTOR, PAUL E. (1976). Choosing response categories for summated rating scales. Journal of Applied Psychology,
61, pp 374-5.

SYKES, WENDY & COLLINS, MARTIN (1988). Effects of mode of interview: experiments in the UK. In R. Groves et al.
(Eds.), Telephone survey methodology. New York: Academic Press.

TOURANGEAU, ROGER (1984). Cognitive sciences and survey methods. In T. B. Jabine, M. L. Straf, J. M. Tanur & R.
Tourangeau (Eds.), Cognitive aspects of survey methodology: building a bridge between disciplines. Washington, DC:
National Academy Press, pp 73-100.

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity
Journal of the Market Research Society April 1997 v39 i2 p331(1) Page 12

The Likert scale revisited: an alternate version.(product preference testing)

TOURANGEAU, ROGER (1992). Context effects on responses to attitude questions: attitudes or memory structures. In N.
Schwarz and S. Sudman (Eds), Context effects in social and psychological research. New York: Springer-Verlag, pp
35-47.

WYATT, RANDALL C. & MEYERS, LAWRENCE S. (1987). Psychometric properties of four 5-point Likert-type response
scales. Educational and Psychological Measurement, 47, pp 27-35.

Professor Gerald Albaum

Gerald Albaum is a distinguished Professor of Marketing at the University of Oregon, where he has been on the faculty
since 1969. In addition, he holds an appointment as a Senior Research Fellow at the IC2 Institute at The University of
Texas at Austin. He is also a Visiting Professor in the Department of Marketing School of Business and Management at
the Hong Kong University of Science and Technology.

He is the author of seven books on marketing and marketing research and eight monographs, as well as an active
contributor to professional journals and a frequent speaker at international and national conferences.

- Reprinted with permission. Additional copying is prohibited. - GALE GROUP

Information Integrity