10
AN EXPERIMENTAL STUDY OF RELATIONSHIPS BETWEEN ATTITUDES, BRAND PREFERENCE, AND CHOICE’ by Frank Ar. Bass,2 Edgar A. €’essemier,2 arid Bonald R. Lehmann3 This paper examines the effectiveness of models of preference and choice based on specific product attributes. An experiment with soft drink brands produced data that reflect preference rankings, choice, and evaluation of specific attributes at different points in time. Deterministic and stochastic choice models are examined. 0+3 N THIS St~Udy \VC extelld the SCqUCIlCC Of I studies (Bass CE Talarzyk, 1972; Bass, l’essmiier, Teach, ~fi Talarzyk, 1969; Leli- mann, 1971) in which we have been cngagcd where rclat,ioriships between prcfercnce for similar alternativcs and attitude measures based 011 specific attributes of thc altcma- t,ivev are explored. We rcplicatc certain fcatm-cs of our previous studies and intro- duce the new olements of dynamirs and act,ual choice. The basic proposition which motivates this research is tlie idea that choice behavior is influenced by pcrceptioris of and values for specific attribut,cs of t.he choice objects. The theoretical foundation for the basic propositmion lies in tJhc:i,ttitjudo theories of social psychologists. Lancastcbr (1966) in the context of economic theory has suggested similar ideas t’o those which stcm from social psychology in indicating that thc utility of a good is bascd on tlie utility of thc propertics which the good dd‘ 1 ivers. lrishbein (1967) has suggested :I gclner:il theoretical (and operational) model for attitude measurement that scrves as the point of departurc for our studies Fishbein’s iiiodel, consistent with the att,itude theory tradition from which it derives, \\-as not formulitted with the measurement of relative at,titudes toward similar objccts in mind, but with the view toward the measurement of an attitude toward a given object. Houwcr, it swms quite natural to extend the theory. Since an attitudc is defined as “:I prcdis- position to respond in a particular wag * This research was supported by :L grant from Kraririert Graduate School of Tiitliistrial the A. h. A. A. Edricational Foundation. Administration, Pnrdue University. 582 Behavioral Science, Volume 17, 1972 to.l\-ard a spccfied class of objects” the argu- ment presented subsequently seems coii- sisterit with this definition. If one’s attitude t,o\vard object A is more favorablc than his attitude toward object R, then object A is prefcrred to object B, arid if object A is preferred to object R, then, other things being equal, it is more likely that object 9 will be chosen than object R. The central underlying proposition in at- titude theory is that attitudes are composed of beliefs about the attributes of objects and the evaluative aspect of thosc beliefs. Thus, for cxamplc, if we were measuring attitudes toward diffwent brands of t’oothpaste we would determine the beliefs which consumers have about the extent t’o which each of scveral brands possess properties such as decay prevention, teet’h whitening, taste, and br&h control and tlie importance which consumers attach to each of these properties. The measurements, then, are specific to the choice object rnthcr t,han general measurements such as personality or generalized attitudes and intentions. The cognitive and affective components of att.it,udc are included in the measurement,^. CHOICE OF THE MODEL AND MEASUREMENT METHOD The basic underlying t:heory concerning the determinants of choice is capable of analysis by direct (obtrusive) measures or by indirect (unobtrusive) measures such as iionmctric multidimensional scaling. Zim- bardo and Ebbcson (1969, p. 62) havc noted I‘. . . thc paradox of two secmingly 3 (;mtIuatc School of Biisiness, Columbia Uiiiversit y.

Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

Embed Size (px)

DESCRIPTION

Studiu experimental cu privire la atitudini si preferinte ale brandului

Citation preview

Page 1: Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

AN EXPERIMENTAL STUDY OF RELATIONSHIPS BETWEEN ATTITUDES, BRAND PREFERENCE, AND CHOICE’

by Frank Ar. Bass,2 Edgar A . €’essemier,2 arid Bonald R. Lehmann3

This paper examines the effectiveness of models of preference and choice based on specific product attributes. An experiment with soft drink brands produced data that reflect preference rankings, choice, and evaluation of specific attributes at different points in time. Deterministic and stochastic choice models are examined.

0+3

N THIS St~Udy \VC extelld the SCqUCIlCC Of I studies (Bass CE Talarzyk, 1972; Bass, l’essmiier, Teach, ~fi Talarzyk, 1969; Leli- mann, 1971) in which we have been cngagcd where rclat,ioriships between prcfercnce for similar alternativcs and attitude measures based 011 specific attributes of thc altcma- t,ivev are explored. We rcplicatc certain fcatm-cs of our previous studies and intro- duce the new olements of dynamirs and act,ual choice. The basic proposition which motivates this research is tlie idea that choice behavior is influenced by pcrceptioris of and values for specific attribut,cs of t.he choice objects. The theoretical foundation for the basic propositmion lies in tJhc :i,ttitjudo theories of social psychologists. Lancastcbr (1966) in the context of economic theory has suggested similar ideas t’o those which stcm from social psychology in indicating that thc utility of a good is bascd on tlie utility of thc propertics which the good dd‘ 1 ivers.

lrishbein (1967) has suggested :I gclner:il theoretical (and operational) model for attitude measurement that scrves as the point of departurc for our studies Fishbein’s iiiodel, consistent with the att,itude theory tradition from which i t derives, \\-as not formulitted with the measurement of relative at,titudes toward similar objccts in mind, but with the view toward the measurement of an attitude toward a given object. Houwcr, it swms quite natural to extend the theory. Since an attitudc is defined as “:I prcdis- position to respond in a particular wag

* This research was supported by :L grant f r o m

Kraririert Graduate School of Tiitliistrial the A . h. A. A. Edricational Foundation.

Administration, Pnrdue University.

582

Behavioral Science, Volume 17, 1972

to.l\-ard a spccfied class of objects” the argu- ment presented subsequently seems coii- sisterit with this definition. If one’s attitude t,o\vard object A is more favorablc than his attitude toward object R, then object A is prefcrred to object B, arid if object A is preferred to object R, then, other things being equal, i t is more likely that object 9 will be chosen than object R.

The central underlying proposition in at- titude theory is that attitudes are composed of beliefs about the attributes of objects and the evaluative aspect of thosc beliefs. Thus, for cxamplc, if we were measuring attitudes toward diff went brands of t’oothpaste we would determine the beliefs which consumers have about the extent t’o which each of scveral brands possess properties such as decay prevention, teet’h whitening, taste, and br&h control and tlie importance which consumers attach to each of these properties. The measurements, then, are specific to the choice object rnthcr t,han general measurements such as personality or generalized attitudes and intentions. The cognitive and affective components of att.it,udc are included in the measurement,^.

CHOICE OF THE MODEL AND MEASUREMENT METHOD

The basic underlying t:heory concerning the determinants of choice is capable of analysis by direct (obtrusive) measures or by indirect (unobtrusive) measures such as iionmctric multidimensional scaling. Zim- bardo and Ebbcson (1969, p. 62) havc noted I ‘ . . . thc paradox of two secmingly

3 (;mtIuatc School of Biisiness, Columbia Uiiiversit y.

Page 2: Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

ATTITUDES, BRAKD PREFERENCE, AND CHOICE 5 33

interdependent areas of investigation- psychological scaling (psychometrics) and attitude change-existing independently and even in ignorance of each other.” To this might be added the recent thrust from economics as suggested by Lancaster (1966).

Several exploratory studies employing nonmetric scaling procedures (Green & Carmone, 1970; Niedell & Teach, 1969; Noinpour, Engel, & Talarzyk, 1970) have been published. In our experiment we have utilized measures which permit both the direct and the indirect approach. In this paper, however, we focus on the analysis of tjhe direct measures.

The Fishbein model for attitude is:

whertb

A , = the attitude toward object o, B , = the strength of belief i about 0,

i.e., the probability that o is related to some other object x,,

at = the evaluative aspect of B , , i.e., the evaluation of 2%. Fishbein’s theory suggests that a, is to be measured in terms of the goodness or badness of attribute i. In adapting the model to the brand preference context, we have taken ai to mean the relative importance of attribute i.

N = the number of attributes.

A deterministic application of this model to the prediction of brand preference would suggest that the rank order of a consumer’s preference for brands should be predicted by the rank order of the attitudes toward the brands and a deterministic application to the prediction of choice would imply that the consumer would choose his most pre- ferred brand. This implies that consumer choice is entirely deterministic and that changes in choice are necessarily associated with changes in attitude. A stochastic application of the model would imply that the probability that a consumer would choose a brand on a given occasion would bc governed by his attitude toward the brand. Attitudes should then imply something about relative frequency of choice of brands

and something about the nature of brand switching. We examine both deterministic and stochastic models of choice and, in addition, study the variety seeking dimen- sion of choice behavior.

THE EXPERIMENT

The experiment was conducted in Lafay- ette, Indiana, between July 7 and July 25, 1969. The subjects were 280 students and secretaries, of whom 264 remained a t the end of the experiment. These subjects were required to select a 12-ounce can of soft drink four days a week for three weeks from the following set of alternatives:

NON - DIET DIET I I I

Diet Pepsi P e p s i Tab I L i k e

S p r i t e Fresco

Soft drinks were chosen as the set of alter- natives for a variety of reasons:

1. They are a low cost item. 2. They are familiar to the subjects. 3. Once opened, they are perishable.

This reduces the probability of stockpiling. 4. It is a product which is normal for

college students to buy daily. In order to keep the selection as natural

as poPsible and to control the effect of specific purchase and use contexts, subjects were allowed to make their choice any time between 9:OO A.M. and 12:30 I’.RI. in a room which adjoined the student lourigc where qoft drinks, candy, etc., are available in vending machines. All the subjects had a reason to be in the building daily betn-cm those times.

In addition to making choices, partici- pants were required to fill out three differcnt questionnaires a t various times:

(1) QUESTIONNAIRE 1 (filled out both a t the beginning and at the end of the experi- ment)

(a) Similarity judgments. Paired com- parison similarity judgments for each pair of brands.

(b) Actual consumption. Subject esti-

Behavioral Science, Volume 17, 1972

Page 3: Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

534 E’. M. BASS, E. A. PESSE~LIIER, AND D. It. LEHVANN

mates of their usual consumpt,ion rates for soft drinks.

(c) General attitudes. Rleasures of at- titude toward such general ideas as health, dieting, and dependence on others.

(d) Flavor prefeience. l’refcrcnce for cola, orange, lemon-lime, etc., on a 1-6 scale.

(e) Ideal position on five attributes: carbonation, calories, sweetness, thirst quenching, and popularity with others. Subjects were asked to indicate ideal soft drink brand by marking on a 1-6 scale the ideal amounts of attributes.

(f) Importance of specific attributes : The five in (c) plus flavor and price. Subjects were asked to assign a weight on a 1-6 scale indicating thc relative importance of each attribute.

(2) QUESTIONNAIRE 2 (filled out a t the beginning of the experiment and a t the end of each of the three weeks, constituting four waves of measurement of preference measurement and belief componcnts of thc attitude model)

(a) Preference for the eight brands. Sub- jects were asked to provide a rank ordering of their preferences for the brands.

(b) Beliefs about thc brands on five at- tributes in (le) above. Subjects were asked to assign a value on a 1-6 scale indicating belief about the brand’s possession of that attribute where 1 indicated that the brand was high in the characteristic and 6 indicated that the brand was low in the characteristic.

(3) QUESTIONNAIRE 3 (filled out a t the

(a) Dollar metric preference measurc. A measure of relative preference for each of the brands developed by asking subjects to indicate how much more in cents they would be milling to pay for a six pack of their favorite brand in a paired compari- son with each of the ot’hcr sevcn brands.

end of the experiment)

(b) Beliefs about the brands on attributes:

(c) Some general questions. packaging and after taste.

For participating in the experiment, sub- jects reccivcd $3 in addition to 12 cans of soft drink. Subjects were told that they wer(’ participating in a choice c.xpcv-iment arid that they wcrv free to make their choices in any uay which seemed appropriate to them.

THE MODEL

Among the primal y issuw to bc caxplored is the choice of a model from among a general class of models. This choice will hinge not only on the selection of parameters but also upon the measurement mc.thod for certain variables. If we define:

number of dimensions, or at- tributes, distance from object j to thc ideal point at time t , weight attached by the individual to attributr i a t time t , belief about object j on attiibute i a t time t , ideal position on attribute i a t time t , error associated with the measure- ment of the distance of object j to the ideal point a t time t ,

then Y’,, , V t t t , ZtJz, I’,,, and W t J are the corresponding measurements of these vari- ables.

Postulate: il’j - Y , , < Y t J Now let us write Y’, in terms of a paiticu-

lar type of distance measure:

Y’tJ = CL V’t, 1 B’,,, - l ’ t L /K

where K is an integer. In order to select a particular form of the model, therefore, different values of K , I , % , and V t L can be used to calculate Y’,. The preferred model will be the one which produces a I”, that is closest to being in- versely monotonic with preference.

SELECTION OF PARAMETERS AND MEASUREMENTS

Exploration of paramrters and measure- ments was limitcid to the following varia-

Behavioral Science, Volume 17, 1972

Page 4: Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

r‘ r ATTITUDES, BRAND PREFERENCE, AND CHOICE d < L I

tions : lines 1, 2, 3 and 4, the City block distance

(a) K = 1 (City block distance) = 2 (Euclidean distance)

(b) I , i = 1 (More is better) = 5’ (Subject’s stated ideal point)

(c) V t i = 1 (Equal weighting of at-

= I (Importance weight assigned

Ccrtain theoretical implications associatcd with various forms of the model are worth noting. If one views the model in an economic t,heory context (Lancaster), K = 1 (City block) implies constant marginal utility for the attributes, while K = 2 (Euclidean) implies one form of diminishing marginal utility. When I , i is a constant for all i, such as 1, the ideal point measurement does not contribute information about preference. If K = 1 and I , i = 1 the model is equivalent to the Fishbein model. A description of the ability of different forms of the model to predict an individual’s stated profercnce ranking is summarized in Table 1.

Shown in Table 1 is a comparison of the attitude models in each of the four waves (belief measures from questionnaire 2) with statod preferences of the subjects. For each subject a Spearman rank correlation coeffi- rient has been computed between that sub- ject’s attitude toward each brand and t,hc subject’s preference rank assigned tho brand. Tho average Spcarmati coefficient and the number of subjects for which thc computed coefficient exceeded .8 is shown for each wave and model. In addition, t’he percentage of subjects whose most prc- fcrred and least preferred brand was cor- rcctlv predicted for each wave and each modcl is indicatcd. For comparison purposes w also show the number of Spearman cor- relation coefficients which exceed the value which results when the stated preference rank for each subject is correlated with a predicted rank based on average preference order, or market share, i.e., Coke first, Pepsi second, 7-Up third ctc.

Direct comparisons of City block versus Ehclidean distances can be found on lines 5 and 6, and 7 and 8. In these two cases, City block is the preferred metric. As indicated by

tributes)

by subject)

measure with an ideal point of all ones and unequal weighting seems best, as it did in an earlier study (Lehmann, 1971) where tele- vision shows were the preference set. It results in an average Spearman 1’ which varies from .75 t,o .81 over the four waves. The first preference brand is predicted cor- rectly about 58 percent of the time over the four waves and the best model provides noticeably better predictions than prcdic- tions based on the average preference order.

A comparison of lines 1 and 15 and lines 5 and 7 in TabIe 1 suggests that the stated ideal points diminish predictions slightly when a.11 of the attributes are used and sig- nificantly when a reduced set of attributes is used. These results are also consistent with those from our earlier studies. While we cannot be certain about the reasons for the variability of predictive consistency with diff crent weighting and ideal point measures, it is our belief that consumers understand the question concerning the importance of attributes, but tend to be confused about thc ideal point question. Also, me are concerned about a possible change in the psychological metric which may apply when one moves from attribute to attribute. Thus, we arc not yet ready to abandon the ideal point con- cept. l ka l ly , lines 15 and 16 indicate a declinc in predictive power that occurs when thc number of attributes is rcsduccd.

Table 2 shows the mean distanccs from ideal points and the standard deviation for each of thc eight brands arid each of the four ivaves. The mean distances arc: r e markably stable over time as indicatcd in Table 2, wen though the diffcrenccs in the means are statistically significant in somc cases.

PREFERENCE ORDER PREDICTION AND CAUSALITY

Table 3 is a corifusion matrix shoji ing th(L conditional probabilities of actual prrfercw-c. rank given the predicted preference rank for the Fishbein model in wave 1. The numbers in the diagonal in Table 3 are the fractions of subjects whose rank ordering of the brands in terms of preference were correctly prc- dieted by thv Fishbein model. For each

Behavioral Science, Volume 17. 1972

Page 5: Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

536 I;. RI. BASS, E. A. PESSEMIER, AND D. R. LEHNANN

TABLE: 1 SUM31 \RY T ~ E S U L T S FOR A L T E R N \ T I Y E MOI)ELS O F P R E F E R E N C E O R D E R P R E D I C T I O N S

Averaxe Spearman v r

S o . of for market ' y , I'irst ':, Last Spearman share Ijreferencc preference I ai question- \'ariables

i's >.8 prediction2 correct correct naire wavezl used

1 1 2 2 3 3 4 4 5 1 6 1 7 1 8 1 0 1

10 1 I1 1 12 1 13 1 14 1 15 1 16 1

I 1 I I I 1 1 1 1 1 2 1 1 s 2 s 1 1 2 s 1 1 I 1 1 1 1 1 1 s 1 s

I .75 I .70 I .7R I .81 1 .70 1 .68 1 .35 I .27 I .42

I .75 I .70 I . ti6 I .76 I .64 I .42

I . 80

146 208 21 1 222 121 113 53 45 69

102 140 127 lo!) 158 113

G I

174 l'J3 1518 204 155 151 70 71

102 186 174 154 143 188 148 102

62.5 58.3 57.8 54.8 4R.3 47.9 23.4 18.8 37.9 42.0 62.5 54.5 52.7 63.8 16.6 37.9

52.7 51.0 57.1 58.3 45.2 47.1 33.3 34.9 33.7 47.0 52.7 44.3 39.0 51.5 45.5 33.7

*

** I t

** ** l"i

t ** 1

* -i.B

* Carbonation, calories, sweetness, thirst r(nenching, popuiarity, flavllr, packaging and aftrlr taste. ** Carbonation, calories, sweetness, thirst quenching and popularity. f Carbonation, calories, sweetness, thirst quenching, popularity and flavor. 68 CarlJi>natiou, calories, sweetness, thirst i~uencliiug and flavor. 1 This involved tnking beliefs from the initial questionnnire and predirting preference on the second queationnaire. Due to the

a Eaoli individual was assomcd to prefer Coke first, I'cpsi second, 7-1Trj third, and so on according t o average popularity. The

3 The wave during wliich IJelieC was rnensured comes before tlic comma and tlic wive during which preference wits rneasured follows

stnliility of beliefs and preferences, this is essentially the sainc rcsult its using beliefs and preferences from the same period.

number of individuals better described I,y the model than 1)y average preference is summarized here.

the comma.

JVh(~n t'he Fishboin attitude scorc for a subjwt implies that a brand is most pro- f w ~ c : t l by a subject thc brarid is, in fact, the oiio tho subject indicates lie prefers most 62.5 percent of thc time. The probabilit,ics :~Ioiig tlic: di:tgon:tl (italics) are largest and thr: prcdictions are strongest for first prcf- (wJi(:c arid lcast jmfcrcncc prcdie tions. The rtmlts arc similar to those of our previous studicts (Bass Q. Tdarzyk, 1972). The prob:tbility that a soft drink brarid will be preferred first or s(:cond, given that tho riiodel predicts that it is preferred first, is about .S and tjhc probability that the brand will be preferred last or next t'o last, given that the prediction is that it \vill be pro- forred last,, is about .7.

I?'estingcr (1957) arid ot'hers havc: raiscd questions conccrning tho naturc: and dircc- tion of causation in relations brt\vc.cn at- titudes arid behavior. Thr striving for cognitive consistency may be a factor to consider in examining relat,ions b(hvc1cri p r d c rmw and attitudc arid thc causal c o ~ i - rioc:t8ic ~i b(lt,wecn thc. t,\vo. A rcspondmt may ratc :L brand highly on a given att,ributc

______

Wave 2 ____ Wave 3

-~

h a v e 4

Coke

7-IJp

'rail

Like

I'epsi

Sprite

Uict I'qisi

Frescn

34. nr (lcJ.50)

46.46 (19.43)

82.21 (21.04)

79.01 (22.66)

35.58 (19.41)

56.15 (21.43)

x2.49 (21.78)

78.81 (26.90)

~~~

31.54 i I 8.37)

43.01 (20.05)

83.74 (21.88)

76.84 (23.06)

34.20 (19.04 j

57.27 (22.08)

83.91 (23.33)

81.69 (27.51 j

31.52 (18.11)

45.95 i l9.2G)

R5.14 (23. Gti 1

82.18 (25.20)

33.50 (18.29)

56.17 (22.76)

85.39 (24.07 j

83.63 (27.83)

31.51 (18.84)

48.46 (46.94j

R!).211 (3!). 05 )

85.33 (28.75)

34.50 (xi. 8!)) 5!i.73

(42.32)

X8.82 (49.6!i)

85.15 (3ti.85)

wbjcct, a lcihbcin attitudr SCOI c is dcvcbpcd for each brand. Tht. rank ordrring of thvw attitudv woreb then predicts the statcd rank ordcting of prefrrcncc for that subjclct.

Behavioral Science, Volume 17, 1972

Page 6: Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

ATTITUDES, BHAND PREFERENCE, AND CHOICE

I 2 3

Predicted 4 Rank 5

537

.6%5 ,189 ,102 ,042 ,030 .OO8 ,004 ,000 I . 00 ,182 ,499 - 1 4 4 .O‘J1 ,030 ,019 ,027 ,015 1.00 .091 ,167 . I@ , 174 ,129 . O l i l ,027 ,008 1.00 ,045 .068 .I52 .371 . I 0 7 ,106 .061 .030 1 .C0 ,027 ,027 ,091 ,140 ..?.JO ,208 . I44 ,125 I . 00 ,011 ,042 .095 ,011 ,011 .053 ,004 .008 ,023

087 ,239 .ZGf ,159 ,106 1.00 057 ,110 .205 .Jfi7 ,186 1.00 04s ,057 .m . 2 1 2 ,527 1.00

because the brand is preferred. WP haw, in fact, observed a slight halo effect. S omc rwpondrnts do, apparently, bias their rating of attributos becausc they prefer the brand. Howcwlr, insofar as 11 c can dctrrminc, this t endency is riot especially strong or pc’r- vasive. Correlations between beliefs about a givcri brand on two dimensions arc in goncml cluitc low and almost always less than . 3 . In fact, tlie correlation bet\\ (’en prcfcrenrc for the brand and belief about the brand ncvcr cbxcecds .3 , although for important attributes the correlations are fairly strong. For example, the avorage cor- relation for prcferencc. for a brarid with bc1ic.f about that brand on the attributc thirst qucmhing is 3 7 .

BRAND CHOICE AND BRAND SWITCHING

While. the F’ishbcin attitudv modd pro- vidvs fairly good predictions of thc prcf- (wnce order for brands, a much hardcr test is it5 ability to provide predictions of choicc. 1 1 1 an cwlicr study Pesscmier, Burg r , Teach, and Tigert (1971) used laboratory tlcrived, scaled preference data to predict rthtive frequency of purchase of brands mvasurcd ovw a scvm-month pwiod. On thc basis of this work we know that (a) pr vfwcnce docs provide significant predic- tions of choice behavior, but that (b) b(v.a,usc of noise in tlir: measuremmts arid the passage of time, high predictive p o ~ e r is dificult to achieve. Clearly, w qhould vxpect stated prrfcrence to bo a stronger predictor of choice behavior than the more remote attitude index based on beliefs about and values for specific attributes of brands.

Table 4 provides a comparison of the.

TABLE 4 A(:c’rrIt.\c.v OF CHOICE PREDICT~ONS BASED ON

I)IFFERENT MODELS ..... __~ .~ .~~____- - ~~~~~

~~ ____ Percentage correct

Model choice predictions All periods Stable periods**

.. ~. ~. ..___~..~___.

Stated first choice (Post’) 52.1 56.4 Stated first choice (Pre’) 50.8 51.8 Last period choice 37. I 45.1 Derived first choice (Pre) 37.6 40.8

(Fishbein Model)

(Fishbein Model) I h i v e d first choice (Post) 37.5 39.9

Predict Coke always chosen 26. I 30.0 Random 12.5 12.5

* Post means that first clioice, either stated or derived, is bwed on questionnaire 2 tlie first time it was mked after the choice, while pre means the questionnaire 2 response obtained most recently before nrtunl choice was utilized.

**Periods 1 , 2, 4, 8, 11, 12.

Fishbein model’s ability to predict actual choice with stated pr(If(wmcw, last pvriod clioicc, most popular brand (Coke) :tl\\.iiy,i predict ed, and random prcdiction. The model fares well in compalison with all of the alternative prcdictiv(1 models cxcqlt for stated first choice and last pcriod choire. I t is not surprising that stated first choicc. is a bctter predictor than dwived firht choice, and sirice statvd first choice provitlcs a corrcbct pr cdiction only about 30 percent of the timtl, a 40 pcwent suc(~’qs rat(, for thv E’ishbcin modrl suggc that thci irifo~ ma- tiori loss is riot c.xccwsivc1 in th(, trarisitiori from thc prc$mmce rn(Iaqur(1 of attitude to tho Fidibcin mvasur(’ of attitude. In addition to disvriminal dispersion, a roawn for thv lon success rate for thc statcd prefvrcmcch model is that there was a certain “desiro for variety”. In fact, most rcspondcnts indi- cated that they frequently consumcd a brand othcr than thv on(’ th(1y pwferrcd.

Behavioral Science, Volume 17, 1972

Page 7: Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

538 F. M. BASS, E. A. PESSE~IIER, AND D. R. LEH~IANK

TABLE: 5 c:OIZR,EL.\TIONS I(ET\\'EEN SIMII,. \ l< B R A N U S ON

ATTILIHUTE l?ATINCS ~~

~~~

Brand Pair

Pepsi Sprite Diet I'egsi Fresca Attribute Coke 7-up Tab Like

~ ~~~~~~~ . -~ ~~

CarlJoiution . GG .53 .72 . 3 3 Calories .6ti . 4 l .52 .31 Saeetncss .ti3 .49 . 7 0 .32 Thirst ilriencliing ,75 .55 . ti3 .50 Pupolarity wit11 .31 .2(i .48 .25

Vlnvor' 1.00 1.00 1.00 1.00 l't~ckiiging . 4 2 .23 .10 . 3 6 Aftcr taste . 54 .37 . ti0 .20

otliers

~ ~~ ~~~~

* Theso are one because of the way flavor u a s defined on the cluestionnaire.

Whilc tlir rcsults suggest thc difficultjr associated \+Tit11 predictions of arid compl(tc uiiderstaiiding of choice behavior, xvo coil- dude that, thc model and mttasuremmts based on specific attributes of the chojcc: object arc meaningfully related t,o this behavior.

VARIETY SEEKING

A simplistic dot~orminist~ic view of t'lie rcdationship betnecn atti tude, stated prof- ormcc, and choico would suggest that attitude scores based 011 spccific attributw of the choice objoct would predict, stat'ctl preference order and t'hat a c:oiisumc:r would always choose the brand most prc- ferred. Such rz simplistic t,hcory implics t,liat choice jvould changc: vnly if at'titudos

change. No allon-nncc is made for variety seeking behavior. We suggest that a more rcdistic t,heory is : T h e probability of choositiy the most preferred brand i s greatest, bud there is a stochastic componext ~ l f choice wJIich arises because of variety seekin.y. Under this version of thc theory, uncertainty about choice arises for a given choice occasion because of t,lie random character of timing of t8hc desire for variety. Forced brand switching-for example, switching when thc favorite brand is denicd-should be prc- dominant'ly Do brands which have similar attributes to the preferrcd brarid if thc product specific attribute theory of attitudc (Fishbein modcl) is a valid component of th(L choice mechanism. Nevertheless, we should also cxpect to find acertain amount of variety sceking swit>ching under unconstraincd choice conditions. We examine switching under constraincd and unconstraincd con- ditions in subsequent sections.

The cight brands used in the experimcmt iverc originally chosen in a11 attempt t,o hitve two similar brands in each of four quadrants of a spacc whcre lemon-lime versus cola and dict versus nondich were tho two major axes. ,4n application of the TORSCA multi- dimensional scaling program (Young S! Torgerson, 1967, p. 498), a program which maps paircd comparison similarity j udg- merit,x int,o dimensions roughly analogous to at,tribut'es wc have attcmpt,ed to m(mure by diroc t questioning, t'o t,he similarit,y data

TAB1,lC 6 SIVITCHIYG P \TTERNb GOMI' \RlSC)Xh W H E X ]3R,lNI) CHOSEN L 1ST IS 1)ENIEl)

~~ ~ _ ~ _ _ ~ ~

__ _ _ _____ ______ Coke Switching

Coke 7 Up Tab Like Peps1 bprite gi:l Fresca Total n

A\erxge percentttge switching1 48 Ifi 2 b 15 6 2 5 100 239 Period 4-5 percentage switchingz 22 2 3 5.3 13 3 5 100 64 5th Day. No Coke

Period 4-5

7-Up and Sprite Switching Coke 7-111) 'Tab Like Pepsi Sprite p"$ Fresca Total Period 8-9 n

Rme Average Percentage 10 35 2 10 17 10 4 11 100 l i 9

Period 8-9 Switching4 30 3 10 28 - 8 23 100 40 9th 1)ay: No 7-Up or Sprite

Switcliina3 -

~- ~. ~~ . ~. -~ 1 Rend: 230 subjects switched from coke during the three periods 1-2,2-3,3-4; 48 percent switched froin coke to coke. 2 Read: fi4 subjects who previonsly chose coke switclied as noted when coke was not available. 3 Read: As 1 ahove except 179 7 - U p and Sprite choosers, periods 1-2,2- 3 , 3 -4,6-7 and 7-8. 4 Kcad: .is 2 above except 40 7 - U p and Sprite choosers.

Behavioral Science, Volume 17, 1972

Page 8: Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

ATTITUDES, HRANI) PKEFERENCE, AND CHOICE .i 3 0

collcctcd initially support.d this design although Like was not perceived to be a diet drink. The fairly high correlations on the eight attributes for the foul brand pairs shown in Table 5 tend to support the simi- larit,y configuration mentioned above.

To tjhc cxt’ent that switching, whcn it occurs undcr forced circumstances, is t.0 brands which have similar properties t’o the prd‘crrcd brand, the proposition that. choice is governed by perceptions of and va1uc.s for spwific attributes is supported. In order to obtain a incasure of such switching an out- of-stock condition was utilized in the vxper- inient. On the fifth day no Coke was avail- able, while on the ninth day, 7-Up and Sprite were riot offered. From Table 6 it is apparent that the major beneficiary of an out-of-stock condition is the brand most similar to the out-of-stock brand. For ex- ample, in normal periods 15 percent’ of thc people who sclected Coke last time se- Iccted Pepsi the next time. When Coke w a s denied, the switching figure to l’opsi jumped to 53 percent.

When 7-Up and Sprite were denied, sub- jects had to switch to diet lemon-lime drinks to obtain a similar flavored drink. The gain in patronage for Frrsca was marked but tlic! remaining subjects predominantly chose to switch t,o nondiet) cola drinks. This parti- tioning of the marlwt into relatively self- eont,ained diet, riondiet segments can alro be observed in the switching behavior of Coke buyers in the upper half of Table (i. Dick Pepsi and Tab do not gain subjclcts forced to switch from Coke.

Switching behavior undcr conditioris of forced change thus tends to support, th(\

that choice is iriflucnced by at’- t’itudt: as derived from brlicfv about arid values for product specific attributes. WP turn ~ i o w t,o tlw prediction of unconstrained v11( lice: bdiavior.

A sccond view of t8ho flavor-diet partitions of thv market wts uncovered when attc.mpt.s uwe made to predict brand purchase with scalcd dollar metric data. Thc scale valuc for ~ a c h brand expressed the subj ect,’s expcctcd relative frequency (or probability) of choo+ ing (purchasing) the brand. A previously developed model was used to compute the

rxxpcctcd frequencies (probabilitie8) of choice by using special purpose, paired prefcrcncv judgments obtained during the initial inter- virw (Prsscmicr, et al., 1971). Here, subject5 srlcctcd the preferred brand in each pair arid indicated how much its pricr could in- crease before they would select (purchaw) the othcr brand. Table 7 displays the cochf- ficients of detcrmination obtaincld M hrri predicting individuals’ rclativc f r rqu tmy of choice (purchase) for the various brands used in this study.

I’rcfrrence measurcs provide fairly good predictions of an individual’s relative. f r v quency of purchase for the more popular brands. A person who normally buy\ Coke, hon ever, will occasionally buy somcthirig other than his favorite brand for the sake of variety. This switching will occur at random intervals and iq therefore very difficult to predict. We can cxaminc, hot\ cvcr, thr extent and character of varicty wcdiing behavior.

In TabIe S we shou the distribution of th(. predictc.d and actual number of brands choscn over nine choice occasions w h ( ~ i all brands were availablc. The predict!cd num- ber of brands chosen is determined by count- ing the number of brands for cach subject for \\ hich the dollar mrtric predicted prob- ability of choicr exceeded . l . Using thiy rulc only 13 pcrcrnt of the subjccts n c r c prc- dictcd to confincl thrir c8hoicc.s to oricb brarid

TABLIS 7 PREDICTED EEL ITIVE FHEQUENCY o r B I ~ \hi)

CHOICE FROM I<ES(’\LE I)OLI.ZJl % f E l R I ( * PREFERENCE 1) iT4*

_ _ - Coefficient of Determination

Brand Market - - ~ ~ ~ - ~

cola Li~2: Yondiet met share2

Coke . 408 7-rp .164 l k l , .012 Like .004 I’epsi .296 Sprite ,064 Diet Pcpni ,016 Frescu ,032 -

Mean

,501

.291

,548

,190

.38 .37

.460

,270

,510

,232

,501 .4FO

,548 .510

.50

,294 ,270

.190

.232

.25 __

1 See tl’e~sernier, e t ul., 1971). Parameters used here: p = R.O,

2 I.:xr,thnental share of purcliase when all brnnds were avail- y = 2.0.

able.

Behavioral Science, Volume 17. 1972

Page 9: Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

540 11’. A[. BASS, E. A. PESSEAIIEH, AND D. R. LEHMANN

* Predicted b y canntirig number of brands for which t h e clolliw metric predicted probability of purchase exceeded .I. Note that siime positive probability of piirchasc is assigned t o every Iirand, lending to a bias favoring the number chosen exceeding tlic riiimber preclieted, especially in the low rnngc of the numher Iiredicted.

\\ hilc almost 50 percmt it crc expccted to clioosc ttvo brands. The. actual nurnbw- of brandb choscn in thcl cxpcrimmt tmdcd to bo larger for n 5ubjcxct than thc p r d i c t d numbcr choscn by a \ubjcct, possibly n rcflcction of c~xpcrimmtal bias in addition to possiblr bias in thtl cbtimatc of probability of choice. In any evcwt, i t i h clear from Tabk S that only a imall fraction of sub- jwts c+onfincd choiw to a 5inglc brand. Attempts to pr‘ldict rhoicc from statcd prc.fcrcnce or attitude w o r f ~ ~ , th(wforc., 51 ill not fare n (4 \\hen thc. prcldietion i u that a Gigle brarid I\ ill be cho 11 and \\lirn thc. normal pat tmi of bcha or, c q ~ i in thc :tbscIicc of attitude change, is for choicc not confinc.d to a single brand. I t seems rcawri- ablr, then, to eonvcrt attitudes, in somv nay, into probability statc>ment\ about

behavior, the uncertainty a function, in part a t least, of the stochastic variety seeking cwmponent of behavior.

We have seen in Table 6 that under forced switching subjects are more likely to switch to something similar to the brand d(micd than to something different. Pure variety seeking behavior, however, may bc characterized by a switch to something different. In Table 9 the distribution of the most frequently chosen and second most frequently chosen brands is shown. Ten pcrccnt of the subjects chose Coke most often and chose 7-Up second most often arid seven percent of the subjects chose 7-Up most frequently and Coke second most frequently. The first arid second most frequently chosen combinations which in- volve flavor changes are in italics in Tablcb 9. For about 30 percent of the subjects thv second mo5t frequently clioscn brand n as a brarid with a different flavor from the most frcquently chosen brand. There n as no dclibcrate stimulus for attitude changc. during the experiment and attitude changes tvecre not marked, as indicated by the mcans in Tablc 2. We conclude, therefore, that switching behavior was not induced in any substantial way by changer in attitude but is rather a more or less normal behavior pattern. While the experimental conditions may have encouraged more switching than I\ ould ordinarily occur, it docs riot appc’ar rcasonable to assume that switching is cntirdy attributable to expr~rimcntal condi- tions.

Wc note that, with the oxcqition of Tables 7 arid 8, none of the predictions of

Behavioral Science, Volume 17, 1972

Page 10: Behavioral Science Volume 17 Issue 6 1972 [Doi 10.1002%2Fbs.3830170604] Frank M. Bass; Edgar a. Pessemier; Donald R. Lehmann -- An Experimental Study of Relationships Between Attitudes,

ATTITUDES, BRAND I'REFERENC-E, AND CHOICE 54 I

prd":rcwx and rhoicc ir based on statist,ical chmatcw of parameters. Therefore, in this sctisc, t'he ctntirc: samplo is a holdout samplc sincc we havn not usrd the same data to cstim:tt,c: parameters and to prcdict.

CONCLUSION

In this ctxpcrimcnt with soft drink brands \vc huvch studied the relationships bct\\-ccm pc:rcc:ption of arid value for specific attri- hut,w of brarids and prefcrcnces arid choice :tmong the brands. We have confirmed the cwnclusions of our earlicr studies to the effect that the Fishbeiri model of attitude is meaningfully related to the preference ordcr for brands and that measures of stated prcfcrerico do provide significant predictions of calioice bchavior. Choice behavior, h o w cvcr, is not necessarily constant even though stated preference and attitudes are un- (:hanging. If askcd to state a preference for wiiic, for examplc, a consumer may prefer \&tc \vine in the abstract, but the consumw also has a riecd for variety and may choosc rod wine on certain occasions, particularly \ihcn his choice of food is more compatible with red wine than \\it11 white wine.

In this study of the choice of soft drink brarids we have found that brand s\vit,ching t,cnds t'o be characterized more by swit.ching to similar brands than to dissimilar brands. Ncvcrthelcss, therc is also a subst,aritinl :mount of snitching to dissimilar bl-:mds. Thc: predominant switching to similar brands tends to confirm the theory tha,t, choice is iIifluericed by attitudes which &:rive from beliefs and d u o s for product, specific at,tributcs. Siricc choico bchavior is not constant even when attitudes arc uri- changing, ho~vevcr, attitude baaed prcdic- t'ions of choice must, be probabilistic. Only about onr.-half of the time did subjects :tctually choosc the brand they said thcy preferred most,. H o \ ~ c v c ~ , wlicn thcy did not rhoosc the brand they most prefcrrcd, thcy

\vcrc mor? likely to choosc a similar brand t,han a dissimilar brand.

It does appear that for purposes of undor- standing as well as for mariagem(mt, the focus on attributes of importance and for mct'hods and models for mc~asurclment and cvnluat,ion of them is just>ificd.

REFERENCES Ihss, I?. M., Pesseniier, 13. A , , Teach, 1:. I)., &

Talarzyk, W. W. Prefereiice measurements ir i market,iiig research. In American Market - ing Association, Fall Conference, I ' r o c ~ ~ t l - ~ M / s , 1969, 88-95.

Bass, 17. M . , & Talarzyk, W. W. A n attitude motlr?l f4)r the study of brand preference. .I. itlarkel- i n q Ites., 1972, 9, 93-96,

Frstinger, I,. A iheory of c .ogt i i / i re dissonance. I.:vaitstori, Ill.: Itow Petersoti, 1957.

Fishbein, M. Attitude and the predict ion of hehavior. In M. Fishbein (Kd.), ISearhtiqs in al l i lude theory. New York: Wiley, lO(i7,

( Ireeri, P., & Carmorie, F. V i t / l i d i n / e n s i o n r / s m l - i n g atid 'related techniques i t 1 murkeliriq analysis. Boston : Allgri & Bacori, 1970.

Laricaster, K. J . A iiew approach to roiisunier llieory. J . pol . Econ., 19(iG, 74, 132-157.

Lchni:trin, I ) . 1:. Television show preferelice: Application of a choice model. .I. ?~tarkeling

3loiiipour, R . , Engel, J . ] I . , & Talarzyk, W. W. ~liiltidixnerisiorial scaliiig aiid iinfoltliiig analysis: A predictive model of hraritl share. Paper clelivered at Assoc. of Consumer Ilesearch, Amherst, Pvlass., August 1970.

Nietlell, I,. A , , & Teach, ll. L). Preference arid perceptual niappiiig of a coiiveriierice good. I n Americaii Statisticnl Associat ion, 13us. & I ~ k o n . Stat.. Sect., I'roc~eedin~gs, lW9,

Pcssernier, E. A , , Burger, P. C. , Teach, 1:. I ) . , K . Tigert, I). J . Using laboratory h r m d pref- erence scales t o predict cotisiimer britnd piirchases, Mgmt Sci., 1971, ti, 371-385.

Yoring. F. W., & Torgerseii, W. S. TOl{SC:A, :I

JWKTIlAN IV program for the Shephard- Ilrriskal niiiltidirnensioiial scaling aria1 13ehau. Sci. 1907, 12, 198-.4'3!).

Zimhardo, P., & Ehhesen, 15. H. Ztt..ifEltenc~iti~g

Park, Calif. : Atidisoii-Wesley, 1969.

477-492.

I t e ~ . , 1971, 8, 47-55.

188-193.

crllil7ctZes and changing OPhaf~ior. hlenlo

(MaIiiiscript received April 21, 1971)

Behavioral Science, Volume 17, 1972