Group Title: Cognitive differentiation
Title: Cognitive differentiation:
CITATION PDF VIEWER THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00097525/00001
 Material Information
Title: Cognitive differentiation: a structural variable underlying the Fishbein attitude model
Physical Description: xiii, 175 leaves ; 28cm.
Language: English
Creator: Durand, Richard Miller, 1947-
Fishbein, Martin
Publication Date: 1975
Copyright Date: 1975
 Subjects
Subject: Attitude (Psychology)   ( lcsh )
Marketing thesis Ph. D   ( lcsh )
Dissertations, Academic -- Marketing -- UF   ( lcsh )
Genre: bibliography   ( marcgt )
non-fiction   ( marcgt )
 Notes
Thesis: Thesis--University of Florida.
Bibliography: Bibliography: leaves 163-174.
Additional Physical Form: Also available on World Wide Web
General Note: Typescript.
General Note: Vita.
Statement of Responsibility: by Richard M. Durand.
 Record Information
Bibliographic ID: UF00097525
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: alephbibnum - 000167546
oclc - 02858261
notis - AAT3936

Downloads

This item has the following downloads:

PDF ( 5 MBs ) ( PDF )


Full Text












COGNITIVE DIFFERENTIATION: A STRUCTURAL VARIABLE
UNDERLYING THE FISHBEIN ATTITUDE MODEL















By

RICHARD M. DURAND


A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL OF
THE UNIVERSITY OF FLORIDA IN PARTIAL
FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY



UNIVERSITY OF FLORIDA


1975


























To Ileene and Rick for their
love, patience, and sometimes understanding.















ACKNOWLEDGMENTS


Sincere appreciation is offered to members of the

supervisory committee: Dr. Zarrel V. Lambert, Dr. R.

Eugene Klippel, Dr. Ralph B. Thompson, Dr. Jack Feldman,

and Dr. Robert C. Ziller. The author would especially

like to express his debt of gratitude to Dr. Zarrel V.

Lambert, chairman of the supervisory committee. The two

years spent as his graduate assistant was a very worth-

while learning experience.

In addition, the author would like to express his

appreciation to Dr. Donald B. Butterworth for introducing

him to the study of Marketing, Dr. Franz R. Epting for

introducing him to the field of cognition, and to Dr.

Olli T. Ahtola for his unselfish hours of instruction on

attitude theory.















TABLE OF CONTENTS


Page

ACKNOWLEDGMENTS . . . . . . . . .. iii

LIST OF TABLES. . . . . . . . . .. vii

LIST OF FIGURES . . . . . . . . .. x

ABSTRACT. . . . . . . . . ... ... xi

CHAPTER

I INTRODUCTION . . . . . . . 1

Problem Definition. . . . . . 1
Concepts of Attitude and Cognitive
Complexity. . . . . . . 3
Concept of Attitude. . . . 3
Concept of Cognitive Complexity. 5

II CONCEPTUAL FRAMEWORK . . . . .. 10

Attitude Measurement Through Expectancy-
Value Models. . . . . . ... 10
Rosenberg. . . . . . .. 11
Adequacy-Importance . . .. 13
Issues Related to the Adequacy-
Importance Model . . . .. 14
Fishbein Model . . . . . 20
Cognitive Differentiation: Measurement
and Meaning . . . . . .. 27
Approaches by Scott and Crockett 27
Bieri's Theory and Measurement 28
Generalizability of Cognitive
Differentiation Across Content
Domains. . . . . . .. 35

III RESEARCH HYPOTHESES. . . . . .. 40

Hypothesis 1 . . . . . 41
Hypothesis 2 . . . . . 42
Hypothesis 3 . . . . . 42
Marketing Implications . . .. 43









TABLE OF CONTENTS (continued)


CHAPTER Page

IV METHODOLOGY . . . ..... . .. 46

Research Design . . .. . . .. 46
Product Selection. . . ... .46
Sample Selection . . . . .. .47
Instrument Design. . . . .. .48
Instrument Pretest . . . .. .57
Hypothesis Operationalization and
Analysis. . . . . . . .. .57
Hypothesis 1: Cognitive Differen-
tiation and the Number of Cogni-
tive Elements. . . . . ... 58
Hypothesis 2: Predictive Efficacy
of the Intra-Individual versus
Cross-Sectional Analytic Procedures
in the Attitude Model. . . .. .68
Hypothesis 3: Generalizability of
Cognitive Differentiation. ... .71

V RESULTS AND DISCUSSION . . . . .. 72

Hypothesis 1: Cognitive Differentiation
and the Number of Cognitive Elements. 72
Toothpaste . . . . . .. 72
Automobiles. . . . .... .75
Hypothesis 2: Predictive Efficacy of the
Intra-Individual versus Cross-Sectional
Analytic Procedures in the Attitude
Model . . . . . . . ... 77
Toothpaste . . . . . ... 78
Automobiles. . . . . . .. .80
Hypothesis 3: Generalizability of Cog-
nitive Differentiation. . . . .. .88
Aberrations of the Fishbein Attitude
Model: The Effects of Their Deletion
on Study Results. . . . . .. 90
Hypothesis 1: Cognitive Differenti-
ation and the Number of Cognitive
Elements . . . . . .. 91
Hypothesis 2: Predictive Efficacy
of the Intra-Individual versus
Cross-Sectional Analytic Procedures
in the Attitude Model. . . .. .95









TABLE OF CONTENTS (continued)


CHAPTER

VI


CONCLUSION . . . . . . . .

Hypothesis 1: Cognitive Differentiation
and the Number of Cognitive Elements.
Hypothesis 2: Predictive Efficacy of
the Intra-Individual versus Cross-
Sectional Analytic Procedures in the
Attitude Model . . . . .
Hypothesis 3: Generalizability of
Cognitive Differentiation . . .
Areas for Further Research . . .


APPENDICES

I-A TOOTHPASTE INSTRUMENT. . . . .

I-B AUTOMOBILES INSTRUMENT . . . .

I-C INTERPERSONAL GRID INSTRUMENT .. .

I-D QUESTIONNAIRE (Toothpaste) . . .

I-E QUESTIONNAIRE (Automobiles) . .

II FREQUENCY DISTRIBUTION OF CORRELATION
COEFFICIENTS . . . . . . .

REFERENCES . . . . . . . . .

BIOGRAPHICAL SKETCH . . . . . . .


121

133

148

152

155


158

S 163

175


Page

112


112



114

117
117
















LIST OF TABLES


TABLE Page

1 BIERI'S MODIFICATION OF THE REPTEST FOR
ASSESSING COGNITIVE DIFFERENTIATION ... . 30

2 ATTRIBUTES ELICITED FOR TOOTHPASTE. ... . 49

3 ATTRIBUTES ELICITED FOR AUTOMOBILES ... . 50

4 CORRELATIONS BETWEEN NUMBER OF COGNITIVE
ELEMENTS AND COGNITIVE DIFFERENTIATION SCORE
FOR TOOTHPASTE. . . . . . . .. 73

5 CORRELATIONS BETWEEN NUMBER OF COGNITIVE
ELEMENTS AND COGNITIVE DIFFERENTIATION SCORE
FOR AUTOMOBILES. . . . . . .. 76

6 CORRELATION COEFFICIENTS DERIVED FROM INTRA-
INDIVIDUAL AND CROSS-SECTIONAL ANALYSES FOR
TOOTHPASTE. . . . . . . . .. 79

7 CORRELATION COEFFICIENTS DERIVED FROM INTRA-
INDIVIDUAL AND CROSS-SECTIONAL ANALYSES FOR
AUTOMOBILES . . . . . . . . 81

8 CORRELATION COEFFICIENTS BASED ON STANDARDIZED
AND NON-STANDARDIZED DATA FOR DIFFERENT EN-
TRANCE PROCEDURES--CROSS-SECTIONAL ANALYSIS 85

9 CORRELATION MATRIX BETWEEN DIFFERENTIATION
LEVELS FOR THREE COGNITIVE DOMAINS. ... . 89

10 CORRELATIONS BETWEEN NUMBER OF COGNITIVE
ELEMENTS AND COGNITIVE DIFFERENTIATION BY
ENTRANCE METHOD FOR TOOTHPASTE AND AUTOMOBILES
--ELEMENTS ENTERED IN TIE SAME ORDER. ... . 92

11 CORRELATIONS BETWEEN NUMBER OF COGNITIVE ELE-
MENTS AND COGNITIVE DIFFERENTIATION BY
ENTRANCE METHOD FOR TOOTHPASTE AND AUTOMOBILES
--ELEMENT ORDER ALLOWED TO VARY . . .. 93









LIST OF TABLES (continued)


TABLE Page

12 MEAN CORRELATION COEFFICIENTS DERIVED FROM
THE INTRA-INDIVIDUAL ANALYSIS FOR TOOTH-
PASTE--COGNITIVE ELEMENTS ENTERED IN THE
SAME ORDER. . . . . . . . .. 96

13 MEAN CORRELATION COEFFICIENTS DERIVED FROM
THE INTRA-INDIVIDUAL ANALYSIS FOR TOOTHPASTE
--COGNITIVE ELEMENT ENTRANCE ORDER PERMITTED
TO VARY . . . . . . . . . 97

14 MEAN CORRELATION COEFFICIENTS DERIVED FROM
THE INTRA-INDIVIDUAL ANALYSIS FOR AUTOMOBILES
--COGNITIVE ELEMENTS ENTERED IN THE SAME
ORDER . . . . . . . . . . 98

15 MEAN CORRELATION COEFFICIENTS DERIVED FROM
THE INTRA-INDIVIDUAL ANALYSIS FOR AUTOMOBILES
--COGNITIVE ELEMENT ENTRANCE ORDER PERMITTED
TO VARY . . . . . . . . ... 99

16 MEAN COGNITIVE DIFFERENTIATION SCORES FOR
AUTOMOBILES--COGNITIVE ELEMENT ENTRANCE ORDER
PERMITTED TO VARY . . . . . ... .102

17 MEAN COGNITIVE DIFFERENTIATION SCORES FOR
AUTOMOBILES--COGNITIVE ELEMENTS ENTERED IN
SAME ORDER. . . . . . . . .. .103

18 MEAN COGNITIVE DIFFERENTIATION SCORES FOR
TOOTHPASTE--COGNITIVE ELEMENT ENTRANCE ORDER
PERMITTED TO VARY . . . . . . .. .104

19 MEAN COGNITIVE DIFFERENTIATION SCORES FOR
TOOTHPASTE--COGNITIVE ELEMENTS ENTERED IN THE
SAME ORDER. . . . . . . . .. .105

20 MEAN NUMBER OF COGNITIVE ELEMENTS FOR
AUTOMOBILES--COGNITIVE ELEMENT ENTRANCE ORDER
PERMITTED TO VARY . . . . . . .. .107

21 MEAN NUMBER OF COGNITIVE ELEMENTS FOR
AUTOMOBILES--COGNITIVE ELEMENTS ENTERED IN
SAME ORDER. . . . . . . . .. .108


viii









LIST OF TABLES (continued)


TABLE Page

22 MEAN NUMBER OF COGNITIVE ELEMENTS FOR TOOTH-
PASTE--COGNITIVE ELEMENTS ENTRANCE ORDER
PERMITTED TO VARY ............. 109

23 MEAN NUMBER OF COGNITIVE ELEMENTS FOR DIFFER-
ENT ENTRANCE PROCEDURES AND COGNITIVE
DIFFERENTIATION SCORE FOR TOOTHPASTE--
COGNITIVE ELEMENTS ENTERED IN THE SAME ORDER. 110
















LIST OF FIGURES


FIGURE Page

1 STAGED ENTRANCE PROCEDURE . . . ... 60

2 PLOT OF CORRELATION COEFFICIENTS BETWEEN
THE PREDICTED AND ELICITED ATTITUDE AS THE
NUMBER OF COGNITIVE ELEMENTS VARIES . . 63









Abstract of Dissertation Presented to the Graduate Council
of the University of Florida in Partial Fulfillment
of the Requirements for the Degree of Doctor of Philosophy

COGNITIVE DIFFERENTIATION: A STRUCTURAL VARIABLE
UNDERLYING THE FISHBEIN ATTITUDE MODEL

By

Richard M. Durand

August, 1975

Chairman: Zarrel V. Lambert
Major Department: Marketing

Recent emphasis of consumer behavior research has

dealt with the use of various expectancy-value attitude

models in an attempt to gain a better understanding of

consumers' perceptions of products and hence their

purchasing behavior. These attitude models attempt to

predict attitude from the system of beliefs an individual

holds about a product or brand.

Much of the current work on the prediction of atti-

tudes, however, fails to take into consideration individu-

al differences in the structure of cognition. In opera-

tionalizing the attitude models, the implicit assumption

is that cognitive structure, composed of the interrelation-

ship and organization of beliefs, is the same for each

individual. This is reflected in the use of attitude

instruments with a standard set of belief statements to

predict attitude. The concept of cognitive differenti-

ation (a structural variable of cognition), on the other

hand, points out that attitudes concerning a range of








stimuli in a product category can be based on few or many

units of information and that this range varies from

individual to individual.

The purpose of this study was to investigate the

effect of cognitive differentiation on the predictive

efficacy of an expectancy-value model of attitudes, more

specifically the Fishbein model.

Bieri's Reptest was used to measure cognitive

differentiation. The product categories investigated in

this study were automobiles and toothpaste. One hundred

two male students from Southern Illinois University at

Carbondale participated in the study.

Three hypotheses were presented and analyzed in the

study. The first hypothesis was that the optimal number

of cognitive elements, in terms of predictive efficacy,

for the Fishbein model, was positively related to cognitive

differentiation. Six different entrance procedures were

used to enter the cognitive elements into the Fishbein

attitude model. Each of the entrance procedures was

applied in two ways: (1) the cognitive element order

was permitted to vary and (2) all elements were entered

in the same order across all respondents.

No significant correlation coefficients between the

optimal number of cognitive elements and differentiation

were found in the case for automobiles. The largest








correlation for toothpaste was .31. Although significant

it was not large enough to warrant the rejection of the

null hypothesis. It is interesting to note, however, that

all of the relationships for automobiles and toothpaste

were in the predicted direction, except in one instance.

The second hypothesis dealt with comparing the pre-

dictive efficacy of the intra-individual versus cross-

sectional analytic procedures in applying the Fishbein

attitude model. The results indicated that the predictive

efficiency of the attitude model is greater when the analy-

tic procedure is based on an intra-individual analysis

rather than a cross-sectional one.

Three cognitive domains were included in examining

the third hypothesis that cognitive differentiation was

generalizable across different domains. These domains

were automobiles, toothpaste, and interpersonal relations.

The intercorrelations between the differentiation scores

for each domain, although significant, were too low for

the construct of cognitive differentiation to be considered

generalizable.


xiii














CHAPTER I


INTRODUCTION


Problem Definition


Recent emphasis of consumer behavior research has

dealt with the use of various attitude models in an attempt

to gain a better understanding of consumers' perception of

products and to predict and gain insight into the behavior

of consumers. These attitude models attempt to predict

attitude from an individual's system of beliefs or percep-

tions held about a product or brand. With a better under-

standing of the beliefs underlying an attitude, marketing

strategies can be developed in a more efficient manner and

demands by consumers more effectively met.

Much of the current work on the prediction of attitudes,

however, fails to take into consideration individual differ-

ences in the structure of cognition. In operationalizing

the attitude models the implicit assumption is that the

cognitive structure, composed of the interrelationship and

organization of beliefs, is the same for each individual.

This is reflected in the use of attitude instruments with

a standard set of belief statements used to predict

attitude.








The purpose of this study, therefore, is to investi-

gate the effect of a structural variable of cognition (cog-

nitive differentiation) on the predictive efficacy of an

attitude model. If the predictive efficacy is increased

by taking into account individual differences in cognitive

structure, then the usefulness of the attitude model will

be improved with greater insight into the underlying beliefs

of an attitude.

The attitude model which will be examined is the one

postulated by Fishbein (29, 30, 31, 32] and the structural

variable will be cognitive differentiation postulated by

Bieri [12, 13, 15, 17], in Fishbein's model, an attitude is

considered to be a function of the strength of an individual's

beliefs about an attitude object and the evaluation of those

beliefs. Cognitive differentiation, on the other hand, re-

fers fundamentally to the extent to which an individual

distinguishes between a set of objects. In a marketing con-

text, an individual with a differentiated cognitive system

has the capacity to view attitude objects in a multidimen-

sional way because more dimensions (beliefs) are available

for product perception. An individual with a relatively

undifferentiated system can be expected to hold beliefs that

are relatively homogeneous or similar to one another, thereby

limiting product perception to fewer dimensions.








Concepts of Attitude and Cognitive Complexity

The purpose of this study is to investigate the

relationship between an expectancy-value attitude model and

cognitive differentiation (a cognitive complexity variable).

In light of this goal, this section is developed to present

the broad theoretical framework on which this study is

based. More specifically, the concept of attitude will be

defined and the two major schools of thought dealing with

the composition of attitudes will be discussed. Further-

more, the concept of cognitive complexity will be defined

and a theoretical tie between the two concepts will be

established.


Concept of Attitude

There are virtually as many definitions of attitudes

as there are major attitude theorists. One definition of

attitude on which most researchers seem to agree is one

proposed by Gordon Allport. He defined attitude as "a mental

and neural state of readiness, organized through experience,

exerting a directive or dynamic influence upon the individu-

al's response to all objects and situations with which it

is related" [4, p. 45].

This definition leaves a good deal of room for differ-

ent interpretations as to the theoretical conception of the

composition of an attitude. There are two major schools on

the issue of the composition of an attitude. One school,








currently headed by Fishbein [30, 31, 32] and supported by

Shaw and Wright [74] and Thurstone [81], treats attitude

as a unidimensional concept. The second school conceptu-

alizes attitude in a multidimentional manner. Advocates of

this position include, among others, Krech, Crutchfield,

and Ballachey [60], Katz and Stotland [41], and Rosenberg

and Hovland [63].

Fishbein, following Thurstone [81], treats attitude

as a unidimensional concept of affect. This affective com-

ponent encompasses the evaluation, feelings of like or

dislike, of an attitude object by an individual. The cog-

nitive component, which refers primarily to how the attitude

object is perceived (beliefs about the object), is viewed

as being related to attitudes by Fishbein [32, pp. 478-9].

These beliefs held by an individual toward an object are

determinants of attitude and not a part of the attitude

[32, p. 479].

The group of theorists who view attitude as a multi-

dimensional concept conceive attitude as containing conative

(behavioral) and cognitive components in addition to the

affective component. While this theoretical approach to

attitude has a number of supporters, surprisingly few

attempts have been made at operationalizing this multi-

dimensional approach to attitudes [e.g., 49].

Several major reasons have been proposed for treating

attitude as unidimensional concept of affect rather than as





5



a three-component concept. First, in most cases only the

affective component is measured by attitude researchers

even if they are proponents of the tripartite conceptuali-

zation of attitude [32, p. 479]. Second, treating attitude

as a unidimensional concept of affect is on a firmer theoreti-

cal base and describes the true state of affairs more ade-

quately [74, p. 11]. Third, the operationalization of a

multidimensional view of attitude is far more difficult

than using a unidimensional affect measure. Finally, the

single affect score has been shown to be highly related to

an individual's beliefs about the object by Zajonc [93],

Fishbein [31], and Rosenberg [60, 61].


Concept of Cognitive Complexity

Whether attitude is treated as a unidimensional measure

of affect or is viewed as a multidimensional concept, the

cognitive component plays an important role. In the former

case, this component is used in the prediction of affect

(attitude) while in the latter it is one of the three parts

of an attitude. Both conceptualizations basically treat the

cognitive component as being comprised of the various beliefs

that an individual holds about a given attitude object

[32, 41, 50]. The concern then, in an attitudinal framework,

is the identification and subsequent measurement of the

magnitude and direction of these beliefs (or elements of

cognition).








The present study is delving beyond the examination of

the cognitive component in terms of content (magnitude and

direction of beliefs) and will be investigating the inter-

relationship and organization of the beliefs. This inter-

relationship and organization of the elements in the cogni-

tive component refers to an individual's cognitive struc-

ture [17, p. 185; 71, p. 405; 92, p. 321; 93, p. 159].

These cognitive structures have been hypothesized to play

a significant role in a number of psychological properties

such as perception, learning, and other psychological

processes [74, p. 173]. Therefore, it is postulated that

cognitive structures also play a role in attitude formation.

In order to describe the organization and relation-

ship between the cognitive elements, a number of morphologi-

cal properties of cognitive structures have been proposed.

These properties have generally been grouped under the broad

heading "cognitive complexity' [79]. Cognitive complexity,

while there is no single body of theory that entirely en-

compasses the concept, is generally considered to be

associated with an underlying response rather than with

what an individual perceives. In other words, complexity is

concerned with the number of cognitive elements used and

the organization of them in cognitive space. The greater

the cognitive complexity of an individual the more versatile

a system he will have for perceiving objects, events, or

people due to the relationship and organization of the

cognitive elements.









The two most accepted approaches to the study of

complexity, although a number have been proposed [e.g.,

69, 93], are cognitive differentiation and cognitive

integration. Differentiation refers to the relative number

of dimensions used by an individual and, more specifically,

reflects the extent to which an individual's system of con-

structs or cognitive dimensions can distinguish between a

set of objects in a cognitive domain. In other words,

a differentiated individual will utilize a greater number

of dimensions in interpreting and perceiving an object than

will a less differentiated individual. In an attitudinal

framework, the differentiated individual may hold more

beliefs about an attitude object than will one who is less

differentiated. Cognitive integration, on the other hand,

is defined as "the extent to which dimensional units of

information can be interrelated in different ways in order

to generate new and discrepant perspectives about stimuli"

[65, p. 25]. An individual who is cognitively integrated

can combine various independent dimensions concerning an

object thereby enabling alternative interpretations or

perceptions of the object.

Both of these approaches, differentiation and integra-

tion, appear to be reasonably close in that differentiation

seems to be a precondition for integration [78, p. 154].

Several authors have attempted to explain the relationship

between these two concepts on a theoretical level. Witkin








et al. [89] theorized that in order to have greater diff-

erentiation a more complex reintegration of the system was

necessary. Harvey, Hunt, and Schroder, who are primarily

integration theorists, state that "differentiation does not

necessitate integration," but rather "integration must be

preceded by differentiation" [36, p. 22]. Schroder, Driver,

and Streufert, while arguing that differentiation is not a

key aspect of integration, state that with greater differ-

entiation there is a higher probability of integrative

complexity [65, p. 166].

In an attempt to ascertain whether integration and

differentiation are disparate processes, a number of empiri-

cal studies have been carried out using various tech.iiques

for measuring these variables. Wyer [91] found that his

model of integration was unrelated to either his conceptu-

alization of differentiation or that of Scott [71]. Vannoy

[85] found that integration, as measured by Schroder and

Streuferts' Sentence Complextion Test [66], was not signifi-

cantly correlated with Bieri's measure of differentiation

[17]. Findings by Streufert and Fromkin [78] have also

been reported as showing little or no correlation between

various measures of differentiation and integration.

Since there is no apparent empirical support for treating

integration and differentiation as related processes, a

decision must be made on which one to utilize in examining

how a structural variable of cognition relates to an

attitude model.









This study will focus on cognitive differentiation

rather than integration for several reasons. First, the

concept of differentiationhas been espoused by several authors

as being useful in a number of different domains of inter-

est [e.g., 7, 27, 73]. In other words, the theory and

instruments designed to measure differentiation can be

readily modified to be applicable for a number of differ-

ent cognitive domains. Differentiation instruments can be

developed for product and/or brand categories, thereby

making it possible to implement this concept in the study

of consumer behavior. Second, differentiation has a strong

theoretical base from which predictions can be made and

implications drawn. Third, differentiation has received a

greater amount of attention in the psychological literature

and is conceptually clearer than integration. Finally,

being perhaps one of the simpler properties of a cognitive

structure [27, p. 6], it is an appropriate place to begin

the conceptualization of the tie between attitudes and

cognitive structure. The exclusion of integration from

this study does by no means imply that integration is not

applicable to attitude research. Rather, it is felt that

its introduction into attitude research might be more

appropriate after exploratory research is conducted into

the relevance of differentiation.















CHAPTER II


CONCEPTUAL FRAMEWORK


Attitude Measurement Through
Expectancy-Value Models


The measurement of attitudes through expectancy-value

attitude models has gained wide support in marketing as

evidenced by the increasing number of published articles in

the area. The reasons for this support are numerous.

Through the use of this approach to attitude measurement,

for example, it is possible to obtain a greater amount of

information pertaining to the cognitive elements underlying

an attitude toward an object than if more traditional

measures are used (e.g., Likert and Thurstone scales) [1,

p. 13]. The increased information is useful in providing

a greater understanding of consumers' perceptions of the

strengths and weaknesses of products on relevant attributes.

This facilitates the development and improvement of market-

ing strategies.

In light of the advantages that the expectancy-value

orientation to attitudes has over other more traditional

approaches, there are three major goals to this section.

First, the Rosenberg [60, 61, 62], adequacy-importance,








and Fishbein [29, 30, 31, 32, 33] models will be presented.

While other expectancy-value attitude models exist [e.g.,

1, 58, 93], these are the three basic underlying models.

Second, the empirical results of a number of studies will

be provided to illustrate the applicability of these atti-

tude formulations to various product categories in marketing.

Finally, a number of studies reporting various modifications

of these models will be examined. The primary purpose of

this is to present a number of issues dealing with improving

the predictive validity of attitude models. These issues

will be presented primarily in the adequacy-importance

model section.


Rosenberg

In order to examine the structural relationships

between attitudes and beliefs about the attitude object,

Rosenberg [611 utilizes an instrumentality-value or means-

end approach. This type of analysis assumes that "an

attitude toward any object or situation is related to the

ends which the object serves; i.e., to its consequences"

[57, p. 153]. To investigate the relationship between

attitude and an individual's cognitive structure (beliefs

about the attitude object), Rosenberg hypothesized that

[61, p. 367]:

The degree and sign of affect aroused in an indi-
vidual by an object (as reflected by the position
he chooses on an attitude scale) vary as a function
of the algebraic sum of the products obtained by
multiplying the rated importance of each value









associated with that object by the rated
potency of the object for achieving or blocking
the realization of that value.

Algebraically, this hypothesis has been expressed

in model form as [29, p. 394]:

n
A = I.V.
o i=l
i=l


where

A = the attitude toward the object o,

I. = the belief or probability that the object will
lead to or block the attainment of a given valued
state "i",

V = the "value importance" or amount of affect
expected from the valued state "i", and

n = the number of beliefs.

This hypothesis was confirmed by Rosenberg through the use

of a chi-square procedure.

While the Rosenberg model has not been employed exten-

sively in predicting attitude in a marketing framework,

various studies have utilized this model. Bither and

Miller [18] used the Rosenberg model to investigate its

applicability as an index to differentiate among univariate

affect ratings of various types of automobiles. The index
n
was derived from the predicted attitude portion ( Z IiV.)
i=l
of the model. They found that the subject's position on

a univariate rating of brand appeal (affect) was highly

associated with the value importance and attitude object

instrumentality index. This finding was supported whether









a cross-sectional (inter-subject) analysis or intra-

individual analysis procedure was employed.

Hansen [35] and Klippel and Bither [48] concentrated

their analysis among lower-priced items. The applicability

of the Rosenberg model in the prediction of choices among

menu items, hairdryers, restaurants, and books was shown

by Hansen. His findings also indicated that both components

of the Rosenberg model (I and V) added significantly to

the predictive value of the overall attractiveness score

(affect). Klippel and Bither found the model to be pre-

dictive of consumers' choices between brands of mouthwash

although the correlations were low.


Adequacy-Importance

The adequacy-importance model is the most widely used

attitude model in research on consumer behavior. It is a

variation of the Rosenberg model [59, 60, 62] and the Fish-

bein model [29, 30, 31] which is discussed in the next

section. Two primary arguments for deviating from the

"pure" Fishbein-Rosenberg formulations have been presented.

One suggests that the different methods of analysis are

due to "differences in purpose between attitude theories

developed in social psychology and brand preference" [8,


1
The name "adequacy-importance" was used by Cohen,
Fishbein, and Ahtola [23] in their rebuttal to the Sheth-
Talarzyk [76] and Bass-Talarzyk [10] and by Ahtola [1].









p. 461]. The second argument is based on alternative

interpretations to a given theory [75]. The basic

adequacy-importance model can be expressed quantitatively

as [1, p. 35]:

n
A = Z P.D.;
o i=l 1 1


where


A = an individual's attitude toward o,

P. = importance of attribute (dimension) i for the
1
person,

D. = his evaluation of o with respect to the attribute
Dimension i, and

n = number of attribute dimensions.

This model has been used by Bass and Talarzyk [10], Sheth

and Talarzyk [76], Wilkie and McCann [86], Hansen [35],

Bass, Pessemier, and Lehmann [9], Wilkie and Weinreich [88],

Moinpour and MacLachlan [54], and Scott and Bennett [67],

among others.


Issues Related to the Adequacy-Importance Model

The purpose of this section is to focus primarily

on several issues centering around increasing the predictive

validity of the adequacy-importance attitude model.2 The

issues of major concern in this analysis include (1) the


2
An extensive review of the research on expectancy-
value models in marketing and the issues associated with the
predictability of the models is discussed in detail by
Wilkie and Pessemier [87].








use of an intra-individual (within-individual) or cross-

sectional (inter-individual) analytic procedure, (2) the

normalization of the individual components of the model,

and (3) the number of cognitive elements (PiD.) used and

the order in which the elements are entered into the pre-

dictive portion of the equation (ZP.D.).

The results of two studies based on a common data bank

with essentially the same attitude model formulation

illustrate the issue of using an intra-individual versus

cross-sectional analytic procedure. Sheth and Talarzyk

used a simple regression procedure in examining the relation-

ship between a preference scale (attitude) for a given brand

and the predicted attitude derived from the adequacy-importance

model shown above. Analysis was on a cross-sectional basis

with the resultant average r s ranging from .013 to .057

when both components of the model (P. and D.) were used.

When only the belief component was included, the average
2
r s for each product category examined ranged from .091 to

.179. This is well below many other studies using different

statistical procedures.

In the study performed by Bass and Talarzyk, the

analysis was conducted at the individual, rather than cross-

sectional level using confusion matrices. Their results,

although not directly comparable to the Sheth and Talarzyk

findings, provided strong evidence that the model was, in

fact, useful to explain brand preference. Examples of








other studies where the analysis was on an intra-individual

basis includes studies by Nakanishi and Bettman [56],

Wilkie and McCann [86], Hansen [35], Bass, Pessemier, and

Lehmann [9], and Wilkie and Weinreich [88]. A basic reason

for using an intra-individual approach rather than a cross-

sectional one is that the latter assumes respondent homo-

geneity in scale measurement as well as the functional

relationship of brand preference to attitudes [87]. In

other words, different individuals have different anchor

points and response sets which would tend to confound the

results if a cross-sectional analysis is employed.

In order to offset this criticism of cross-sectional

analysis, Bass and Wilkie [11] have proposed that the com-

ponents of the model (P. and D.) can be normalized on an

intra-individual level. Normalizing these components

adjusts the within-subject variance in responses which

eliminates the problem of different anchor points and

response sets.

Bass and Wilkie, in testing the effect of normalizing

the attitude model, utilized the same data base employed

by Bass and Talarzyk and Sheth and Talarzyk. The hypothesis,

concerning the effects of normalization on the predictive

efficacy of the attitude model, was tested by comparing the

amount of variance explained between the predicted attitude

and brand preference for the non-normalized and normalized

responses. The average variation explained for the









normalized attitude model was .39 compared to .15 for the

non-normalized one. This difference is significant at the

.001 level.

The final issue to be examined deals with the number

of attributes or belief components included in an attitude

model to estimate brand preference or attitude. Most

authors appear to assume that five to seven attributes are

salient (relevant) to the individual and therefore use that

many in the attitude model [e.g., 7, 11, 76, 86]. Two

studies, those by Churchill [21] and Wilkie and Weinreich

[88], have examined the question of how many attributes

should be included in the model to predict attitude.3

Churchill used a cross-sectional procedure whereas-Wilkie

and Weinreich analyzed their data on an intra-individual

basis.

Churchill used a range of attributes rather than using

a fixed number of attributes, to examine whether one could

increase the predictive validity of the adequacy-importance

model4 by varying the number of attributes included in the

model. Each individual was asked to rate two writing pens

on 19 product attributes, the importance of each attribute,



Nakanishi and Bettman 156] take into consideration the
fact that different respondents use different numbers of
cognitive elements in their analysis but the manner in
which their results are presented prevents any generaliza-
tion to be made.

Churchill, while claiming to be using the Fishbein
model, was in fact using a modification of the adequacy-
importance measure.








and evaluate each pen. These data yielded 19 preference

predictions for each individual based on the order of

importance of each attribute for each subject. The first

prediction, for example, was based on the attribute con-

-sidered most important by the individual, the second pre-

diction included the two most important attributes for the

individual, and so on. This was done for each subject with

each resultant prediction correlated with the actual evalu-

ation of the pens. In other words, 19 correlation coeffi-

cients were obtained for the sample, in a cross-sectional

manner, with the predicted preference for each subject being

composed of different attributes based on their ratings of

importance. Churchill found that the predictive accuracy

increased, leveled off, and then declined as the number of

attributes increased. The correlations between actual and

predicted preference ranged from .118 for one attribute, to

a peak of .558 for 17 attributes and .552 for 19 attributes.

Wilkie and Weinreich [88] asked a sample of 29 house-

wives to rate seven supermarkets on seven store attributes.

Four studies were performed so that the various analytic

procedures could be compared. The first study followed

the more traditional approach to attitude studies in



5While two attitude models were presented,one with
raw scores and the other with normalized scores, only the
former will be presented. On an intra-individual design,
there is no reason for normalizing the responses.








marketing. All seven attributes were used in predicting

preference of stores with the resultant correlation co-

efficient derived from correlating the predicted and stated

preference. The second study allowed the number of

attributes, included in the model, to vary from one to

seven. A mean determinism score was used to decide the order

or attribute entry for the sample. This determinism score

incorporates both the importance score for each attribute

and the variance in perceived satisfaction on the attri-

butes over all stores. The greater the determinism score

the higher the entry of the variable into the model. The

third study derived attitude scores based on attributes

introduced in the model on an individual basis. In other

words, the determinism score for each individual specified

the order of attribute entry. The same number of attributes

were used for each subject. Finally, in the fourth analysis,

each individual's determinism score was used to select the

attribute to be entered with the attributes being entered

until the maximum correlation with preference was ascer-

tained.

The results of this study by Wilkie and Weinreich

appear to provide support for the hypothesis that the

predictive accuracy of an attitude model can be improved



The determinism procedure is presented in detail in
the methodology section.








when the number of attributes and order of attribute entry

are varied for each individual. They found in study one that

when all seven attributes were used, the Spearman Rho was

.6. When the number of attributes were varied across all

subjects, a Rho of .63 was highest when five attributes

were used. The results of study three showed that the

maximum Rho was .62 with three attributes. Finally, the

highest mean correlation resulted when the attribute entry

and number of attributes entered varied on an intra-

or within-individual basis, .76.


Fishbein Model

Attitude, although treated as a unidimensional concept

of affect, is defined by Fishbein as "a compound in which

the elements are beliefs and the effective value of the

compound is some function of the affective value of the

constituent beliefs" [33, p. 488]. In other words, "an

attitude toward any object is a function of (1) the strength

of his beliefs about the object and (2) the evaluative

aspect of those beliefs" [31, p. 117]. Mathematically,

Fishbein's model is expressed as [29, p. 395]:

n
A = Z B.a.;
i=l

where

A = the attitude toward the object o,

B. = the strength of belief i about o, that is, the
S"probability" or "improbability" that o is
associated with some other concept x.,









a. = the evaluative aspect of B., that is, the
evaluation of x., and
1
n = the number of beliefs about o, that is, the
number of responses in the individual's habit-
family-hierarchy.

While other theorists might operationalize the concept of

attitude somewhat differently than Fishbein, many such as

Shaw and Wright 174], Rosenberg [61, 62], and Peak [57]

would also support the rationale of this basic functional

relationship.

Fishbein defines "belief about an object" operationally

as "the 'probability' or 'improbability' that a particular

relationship exists between the object of belief (e.g., an

attitude object) and any other object, concept, value, or

goal" [29, p. 389]. In other words, a belief refers to the

probability that a relationship exists between an attitude

object (e.g., Colgate toothpaste) and some other concept,

object, value or goal (e.g., whitens). The different types

of beliefs have been classified into six basic categories

by Fishbein. The classification includes [31, pp. 110-111]:

1. Beliefs about the component parts of the object.

2. Beliefs about the characteristics, qualities or
attributes of the object.

3. Beliefs about the object's relation with other
objects or concepts.

4. Beliefs about whether the object will lead to or
block attainment of various goals or "valued
states".

5. Beliefs about what should be done with respect
to the object.








6. Beliefs about what the object should, or should
not, be allowed to do.7

All of the beliefs held by an individual concerning a

given attitude object are viewed as a belief system or

habit-family-hierarchy of responses by Fishbein. Following

Hull [37], the higher the location of beliefs in the habit

hierarchy, the greater the probability that an individual

will perceive an association between the attitude object

and another object, concept, value or goal. Furthermore,

the higher the belief is in the hierarchy, the stronger and

more salient it is in the formation and prediction of an

attitude.

Fishbein uses a free association test in order to

determine which beliefs are salient to an individual out

of the infinite number of possible beliefs held toward

an attitude object and the hierarchical position of each

belief [29, p. 396]. Individuals are presented with an

attitude object and asked to list the characteristics which

they feel best describe the object. The characteristics

obtained are assumed to be those in an individual's habit-

family-hierarchy and the order in which the beliefs are

elicited are assumed to represent the approximate belief



The first four types of beliefs are considered by
Fishbein to fall into what most researchers call an indi-
vidual's cognitive component or structure. Types five and
six are action or behaviorally oriented [31, p. 111]. In
any event, the first four types are those most frequently
used by marketers.









strength ranking. Their relationship between hierarchical

position and belief strength was investigated by Kaplan

and Fishbein [40]. They found that for the first six and

nine beliefs in the response hierarchy, the correlations

between hierarchical position and belief strength were .90

and .72, respectively.

The concept of salient beliefs and their location in

the habit-family-hierarchy are major components of Fish-

bein's attitude theory and his model shown in the formula

above. According to Fishbein,

. the only beliefs that serve as determinants
of an individual's attitude are those that are
present in his habit-family-hierarchy of responses.
That is, although all of an individual's beliefs
about an object serve as indicants of his attitude
toward the object, it is only the individual's
salient beliefs, i.e., those in his hierarchy, that
serve as determinants of attitude. [29, p. 395]

Theoretically, then, the predictive efficacy of Fishbein's

model is at least partially dependent on the saliency of

beliefs used to predict attitude. By including only the

salient beliefs of an individual the best estimate of

attitude will be obtained [29, p. 395].

Support for the proposition that the saliency of the

beliefs affects the predictive efficacy of the attitude

model has been provided by Rosenberg [61] and Kaplan

and Fishbein [40]. Rosenberg found that when an individu-

al's salient beliefs were used to estimate affect there

was a stronger relationship between affect and the summative
n
portion of his model ( E IiV.) than when all 35 beliefs
i=l









provided by the experimenter were utilized. Kaplan and

Fishbein also found that when only the salient beliefs of

each subject were included in the analysis the estimate of

attitude was better than when the total set of beliefs

elicited by an individual were used.

From a practical standpoint, Fishbein argues that a

standard list of beliefs must be provided to the sample

rather than having each subject elicit his own beliefs to

use in the attitude model [29]. This requires that some

decision must be made on the number of beliefs that can be

considered to be salient for an individual and therefore

how many beliefs should be included in the attitude instru-

ment. The major concern is to reduce the loss of predictive

validity resulting from the inclusion of non-salient

beliefs which serve to enter some degree of noise into the

model. In other words, the predictive efficacy of the

Fishbein model is expected to be greater when the propor-

tion of salient to total beliefs is large.

The number of beliefs generally considered to be

salient for an individual at any one time is from six to

eleven [29, p. 395]. Empirical support for this statement

is found in studies by Woodworth and Schlosberg [90] on

span of attention, information processing by Miller [52],

and Kaplan and Fishbein [40]. Based on this evidence,

usually a list of five to ten beliefs is used for attitude

measurement [1, p. 23].








The Fishbein model, while being used quite exten-

sively in the area of social psychology, has not been

employed in marketing without substantial modification.

This is particularly interesting in that the model provides

detailed information regarding the belief component of an

attitude and has been shown to correlate around .7, gen-

erally with independent measures of affect [2].

One marketing study which attempted to use the Fish-

bein model was reported by Sampson and Harris [64], They

found correlations between the Fishbein model and elicited

attitude ranging from .06 to .26. While these correlation

coefficients are very low, they do not accurately reflect

the predictive validity of the model for marketing use.

Sampson and Harris, due to their use of somewhat different

scales, inclusion of non-salient belief statements, and

measurement of belief statements in an incorrect manner,

nullify any conclusions that can be drawn on the Fishbein

model based on their results [1, pp. 46-8]. Chapman [201,

while not presenting specific research findings, reports

on a corporation that has been applying the Fishbein model

to commercial surveys where correlations of .6 and above

have been found.

Even though the Fishbein model has not been used as

extensively in the area of marketing as has the adequacy-

importance model, Fishbein's formulation will be used in

this study. One major argument for using the Fishbein








attitude model over the adequacy-importance model is the

strong theoretical background underlying the Fishbein model.

The adequacy-importance model is a modification of the

Fishbein-Rosenberg formulations and does not have a theoreti-

cal base independent of them. Furthermore, ample studies

have used the Fishbein model, albeit in the psychological

arena, and have attested to its predictive validity.

Although the Rosenberg model also has a strong theoretical

foundation, it is not being selected because the theory is

couched in terms of the attitude object's attainment of

positively valued states or blocking of negatively valued

states.

This conceptualization of cognitive structure is more

difficult to operationalize in a marketing context than is

Fishbein's theory. Fishbein accepts any kind of salient

belief statement which provides greater latitude in the

development of the attitude model and increases the

information available about the belief system underlying a

given attitude.

A final reason for selecting the Fishbein model over

the other two rests with the psychological concept of cogni-

tive differentiation, discussed in the next section. Fish-

bein's conceptualization of salience is theoretically tied

to differentiation, in that the level of differentiation is

hypothesized to affect the number of salient beliefs an

individual holds. Furthermore, with the acceptance of any









type of belief statement, the operationalization of the

Fishbein model is closer than either of the other models

discussed to the conceptualization and operationalization

of the concept of differentiation used in this study.

This similarity in the operationalization of cognitive

differentiation and the Fishbein attitude model will be

developed more fully in Chapter IV where the measurement

methodology is presented in detail.


Cognitive Differentiation:
Measurement and Meaning


The concept of cognitive differentiation is a struc-

tural property of cognition. It refers to the degree to

which an individual distinguishes among the elements in a

given cognitive domain. More specifically, an individual

who has a more differentiated cognitive system or structure

has the capacity to view or perceive objects, persons,

and events in a more multidimensional way than one with a

less differentiated system [14, 16].


Approaches by Scott and Crockett

A number of measurement techniques for assessing

cognitive differentiation have been proposed. Scott

[68, 70, 71], for example, devised a method of measuring

differentiation based on information theory [6]. Scott

either provides or has subjects generate a list of objects,

such as nations, and then asks them to sort the objects








into various groups that belong together based on common

attributes held by each. From this sorting procedure, a

dispersion measure is derived which is interpreted to repre-

sent the number of groups-worth of information yielded by

the classification system of an individual. The major

assumption underlying this differentiation measure is that

attributes are dichotomous. In other words, an object is

sorted into one category or another based on whether or not

it contains a particular attribute. Crockett [24], rather

than using a sorting procedure to measure differentiation,

developed a technique whereby short essays, written about

the stimulus object (people), were analyzed. Cognitive

differentiation was defined as "the number of interpersonal

constructs in these descriptions . [24, p. 51].


Bieris Theory and Measurement

Bieri, in defining differentiation "as the capacity

to construe social behavior in a multidimensional way"

[17, p. 185], assumes that differentiation can be measured

in terms of the dimensional characteristics of a cognitive

structure [15]. This dimensional analysis of cognitive

structure has been drawn and modified from George A. Kelly's

Personal Construct Theory [44]. Because Bieri's underlying

theory and measurement technique is based on Kelly's work,

8
Bieri uses the term cognitive complexity in essenti-
ally the same way that most complexity theorists call cogni-
tive differentiation. Therefore, so as to limit any
confusion due to differences in terminology, the term
differentiation will be used.









part of Kelly's theory and a description of his instrument

will be provided.

Construct theory, as espoused by Kelly, centers on

the fundamental idea that "a person's processes are psycho-

logically channelized by the ways in which man anticipates

events" [42, p. 46]. These ways by which events are

anticipated and the cognitive world construed are called

constructs. In other words, constructs are dimensions or

elements used by individuals to characterize how some events,

objects, or people are similar but yet different from other

events, objects, or people. An example of a construct

possibly used by an individual in construing toothpaste is

expensive-inexpensive. Some brands are viewed as being

expensive and others as not. Constructs are assumed to

be bipolar in nature.

The Role Construct Reperatory Test (Reptest) was

developed by Kelly to provide a means of examining the

constructs used by an individual to construe and give

structure to his environment. In the "gridform" of the

Reptest, a subject is typically presented with a list of 20

to 30 different role figures (e.g., father, mother, girl-

friend). These figures are written down at the top of a

rectangle, much like that of Table 1 showing Bieri's

Technique. The subject is asked to consider, in groups

of three, how two of the individuals listed are alike

and different from the third in some important aspect.










TABLE 1

BIERI'S MODIFICATION OF THE REPTEST FOR
ASSESSING COGNITIVE DIFFERENTIATIONa



1. Yourself

2. Person your dislike

3. Mother

4. Person you'd like to help

5. Father

6. Friend of same sex

7. Friend of opposite sex (or spouse)

8. Person with whom you feel most
------ uncomfortable
9. Boss

10. Person difficult to understand




+ +


F- I HI Ih i


I i 'I0 -. 0 a aD LI -I Ir
+ 0 (D 0
r1 m B H O

H- a ra o i l tn C <

ource: Bieri et al. [17, p. 191].



H a fT I-.




H- Hl-- a
r+-






(D
o hH (D 0 H
( D o 0 H c 0 3
'0 1 ua n a a
0 a '0 rd a H- M 0. m
Se H- 0 c 0r ci- P.- 0. C
H 0 aD a 0 0 U H- a
: hi P- h hi H5 < ct
Wi H 0 a U, (D (D (
rt Q, (D f0
(D 0.


wiS LI ie


aSource: Bieri et al. [17, p. 191].








This construct dimension is written in bipolar form such as

pretty-ugly along the right side of the rectangle. Then the

subject is asked to put a check mark under each individual's
9
name to which this construct can be applied. If the con-

struct cannot be applied to an individual, the space is left

blank. This process is repeated 15 to 20 times or more but

with different groupings. The result of this involved

procedure is a rectangle with a series of voids (blanks)

and checks with a list of constructs at the right-hand column.

Grid data can then be subjected to numerous types of analysis.

The major analytical technique for examining grid

measures of structure, within the framework of Kelly's theory,

is the original non-parametric form of factor analysis

proposed and devised by Kelly [43]. This technique was

later revised and adapted to computer analysis by J. V.

Kelly [45]. The degree of differentiation was defined as

the amount of variance not explained by the first factor

[39 .

Bieri has modified the Reptest developed by Kelly so

that the instrument is more easily administered. Table 1

represents the modified Reptest of Bieri. The 10 roles are

selected on the basis of being meaningful to an individual

and are provided by the experimenter rather than elicited

from the subject as is done in Kelly's procedure. A subject


9
A more detailed description of the Reptest can be
found in Kelly [44, Vol. 1], Bonarius [19, pp. 2-3], and
Bannister and Mair [7, pp. 38-77].








rates each of the 10 persons on a scale of +3 to -3. No

neutral point is provided.

An index of differentiation is derived from the modi-

fied Reptest through a matching procedure devised by Bieri

[14, p. 7; 17, p. 190]. Once the subject has rated each

role person on every bipolar construct, each row of ratings

is compared to the rows below it. In the comparison of any

two rows, when there is an exact agreement in the rating of

a role person, a score of 1 is given. For example, in the

role person "yourself" shown on Table 1, if you have

ratings of +3 for both "outgoing" and "adjusted," then a

score of 1 is given for that agreement. After making all

possible comparisons between rows one and two, rows one and

three are then compared with the resulting number of agree-

ments marked down. This matching procedure is continued for

all possible row comparisons.

After all comparisons are made, then the scores for

each match are added so that you arrive at one score. In a

10 x 10 matrix, since there are 45 possible row compari-

sons within a column (role person)and 10 rows, there is a

total possible differentiation score of 450. An individual

with a score of 450 is assumed to have an undifferentiated

cognitive system in that the construct dimensions were used in

a functionally similar and homogeneous manner to construe

the people on the grid. Those subjects with a score of 100

for example would be relatively more differentiated because









constructs were used differently in discriminating among the

role persons. In other words, there is an inverse relation-

ship between degree of differentiation and score of the

modified Reptest.

As discussed above, Bieri has modified Kelly's Reptest

inthree main ways. In the first place, Bieri does not have

the subjects elicit the constructs as does Kelly. He pro-

vides the constructs which are then rated. The second modi-

fication centers around the rating method for the Reptest.

Bieri has each subject rank each role person on a +3 to -3

scale along each construct while Kelly has the subject use

a checkmark if the construct applies to other role persons.

Finally, Bieri's matching procedure is different from Kelly's

procedure for scoring differentiation.

Each of the Bieri modifications has been examined

empirically in order to ascertain what effect they have had

on the predictive efficacy of Kelly's formulation. Tripodi

and Bieri [82] and Jaspers [39] examined the differences

between differentiation scores when constructs are provided

or are elicited. Both found that there are highly signifi-

cant correlations between Bieri's procedure (provided

constructs) and the one by Kelly (elicited constructs).

In terms of Bieri's modification of scoring the Reptest

grid, Bieri [17] reports a +.90 correlation between J. V.

Kelly's [45] non-parametric factor analysis technique and

his matching procedure where the scales were reduced to









two, rather than six categories. These modifications thus

do not appear to significantly alter the results from Kelly's

Reptest on scoring and analyzing the structural aspects of

the constructs.

Bieri's conceptualization of differentiation and sub-

sequent measurement of it has been utilized in a number of

psychological areas. Examples of the various areas include

stimulus affect and perception [38, 53], social concept

attainment [58], certainty of judgments made by individuals

[83], and information transmission using both clinical and

social stimuli [17].

The test-retest reliability of Bieri's technique is

quite satisfactory. Tripodi and Bieri [82] had test-

retest reliability coefficients of .86 when differentiation

was based on an individual using constructs provided by the

experimenter and .71 when based on constructs elicited from

the subject. In other studies, the test-retest coefficients

have been .70 [83] and .71 [84]. Epting [28], while

analyzing three Bieri grids with social issues rather than

role persons, found the test-retest reliability coefficients

to range from .51 to .65.

The theory and operationalization of cognitive differ-

entiation that will be used in this study is that proposed

by Bieri [13, 14, 15, 16, 17]. Bieri's conceptualization

of differentiation has been selected over that of Scott,

Crockett, and others for several major reasons. First,








while other major complexity theorists have also discussed

differentiation, their emphasis has primarily been on inte-

gration [e.g., 36, 65]. Second, Bieri's concept of differ-

entiation, being founded extensively on the work of Kelly

[44], has a strong theoretical foundation. Third, Bieri's

measurement technique has been shown to be applicable to a

number of cognitive domains when appropriate modifications

are made. Finally, there is no assumption that the

attributes used by an individual to construe his world are

dichotomous. Rather, a six-point scale is used to reflect

that the construing of an object, along a dimension, is

not necessarily black or white but rather some shade of

gray. The Bieri approach to measuring differentiation,

therefore, is conceptually closer to measuring attitude,

via the Fishbein model, than is the sorting procedure by

Scott [71], the essay analysis of Crockett [24], or other

various techniques [e.g., 36, 91, 93].


Generalizability of Cognitive Differentiation
Across Content Domains

A number of differentiation theorists have empiri-

cally examined whether the construct, cognitive differenti-
10
ation, is generalizable across different domains. If an

individual who is relatively differentiated in one cognitive


10
The emphasis is placed on the generalizability across
different content areas when the same measurement instrument
is being used.








domain is also relatively differentiated in other domains,

then the usefulness of the concept to marketing prac-

titioners is greater than if differentiation is domain

specific.

It is Crockett's hypothesis that "individuals with

complex cognitive systems with respect to other people need

not necessarily have complex systems with respect to other

domains" [24, p. 54]. In one unpublished study, Crockett

[241 reported that there is some degree of generality of

differentiation within a given domain (people). No direct

examination of the generalizability across distinct domains

was presented.

Bieri [13, 15] emphasizes that the concept of differ-

entiation is only applicable to an individual's social world.

In fact, Bieri has stated that cognitive differentiation is

concerned only with the dimensional versatility of a person

in his social judgments [12]. There is no theoretical base,

however, to support this contention since the concept of

cognitive structure is theoretically applicable to physical

as well as social stimuli. Moreover, Kelly, in an extensive

discussion on the nature of the grid method, argues that

since the construing process is applicable to all -events,

the grid method is generalizable [44, pp. 301-2].

In two different studies, Epting [27, 28] examined

the generalizability question using the Bieri modification

of the Reptest. The first study [27] examined the correla-

tion between Bieri's traditional measure with role persons









and related constructs and a modified measure designed to

ascertain the level of social issue differentiation. The

social issues ranged from "allowing Red China to become a

member of the United Nations" to "abolishing the death

sentence." The correlation between the two grids, although

significant at the .05 level, was relatively low with a

coefficient of .39. In a more recent study, Epting [28]

developed three different Bieri grids to assess differenti-

ation on three different sets of social issues. The corre-

lation between the differentiation scores obtained from these

grids ranged between .56 to .60.11 Although it can be

argued that the three areas used by Epting do not constitute

completely independent content domains, the results cer-

tainly provide some evidence for the generalizability of

differentiation.

Allard and Carlson [3] provide a more direct analysis

of the generalizability proposition. They used Kelly's

Reptest procedure and designed three grids which encompassed

famous figures, geometric designs, and the person roles

which Kelly used on his original instrument. The inter-

correlations among the three measures of differentiation

ranged from .57 to .67, all significant at the .001 level.


11
Epting reports that he administered the grids twice
to each respondent. The correlations presented here are
those derived from the first administration.









Scott [72] has summarized a number of studies per-

formed both by his students and by himself dealing with this

question of generalizability. One specific section reported

in his paper deals with the domain specificity versus

generality of the various structural constructs developed by

him. In his examination, Scott formed a composite score of

various techniques to measure differentiation and then com-

pared them across domains. The domains included acquaintances,

family, groups, nations, school, and self. In order to

examine the proposition of generalizability, the results of

three different samples were presented. The mean inter-

domain correlation for a Boulder, Colorado, sample was .72

(6 domains), .66 for a sample of students at Kyoto, Japan

(4 domains), and .58 for students at Wellington (2 domains).

Based on these results, Scott concludes that a person's score

on a differentiation variable is not dependent, necessarily,

on the content area assessed. Because the scores reported

were composites, this conclusion is not as strong as it

might have been had the analysis dealt with individual meas-

ures of differentiation across domains.

The findings reported above, although not providing

overwhelming support for the hypothesis that differentiation

is generalizable across content areas, certainly do not

refute it. In fact, the evidence is weighted more heavily

in favor of generalizability than against it. The diversity

of previous findings suggests then that additional studies





39



encompassing dissimilar domains are needed before a definitive

conclusion can be drawn regarding the generalizability

issue.















CHAPTER III


RESEARCH HYPOTHESES


Before discussing the questions to be investigated in

this study, a few of the major theoretical points concerning

Fishbein's attitude theory and the concept of cognitive

differentiation are reiterated.

Fishbein emphasizes that the concept of saliency of

beliefs is a crucial one in the predictive efficacy of the

attitude model. The conventional operationalization of the

attitude model, however, assumes that the cognitive struc-

ture of each individual is composed of the same number of

salient beliefs from which an attitude is formed. This

assumption is inherent in the operationalization because

typically a standard set of belief and evaluative statements

is provided to each subject from which a predicted attitude

score is abstracted. When the proportion of salient to

total beliefs included in the standard set of evaluative

statements is higher, the predictive validity of the

Fishbein model is expected to be greater.

The concept of differentiation points out that atti-

tudes concerning a range of stimuli in a product category









can be based on few or many units of information and that

this range varies from individual-to-individual [65, p 7].

Interpreting these units of information in an attitudinal

framework as cognitive elements suggests that by taking

cognitive differentiation into consideration a more appropri-

ate number of cognitive elements may be used in the measure-

ment of attitude. To the extent that an individual is

highly differentiated in a given cognitive domain, the

inclusion of a greater number of cognitive elements may

improve the predicted measure of attitude. Conversely,

if an individual is relatively undifferentiated, the pre-

dictive validity of the model may be higher when only the

few salient dimensions are used.

A final theoretical area applies to the generaliza-

bility of cognitive differentiation across cognitive areas.

If differentiation is generalizable across different product

domains, the concept's applicability and use in marketing

will be greatly facilitated compared to what would be the

case if the concept is domain or product specific.

The research questions examined in this study are

expressed in the alternative (Ha) and null (Ho) forms of

the hypotheses presented below.


Hypothesis 1

Ha: There is a positive relationship between
cognitive differentiation and the optimal
number of cognitive elements used in the Fish-
bein attitude model to predict a given attitude.








H : There is no relationship between cognitive
differentiation and the optimal number of
cognitive elements used in the Fishbein
attitude model to predict a given attitude.


Hypothesis 2

Ha: The predictive efficacy of the Fishbein
attitude model is greater when an intra-
individual analytic approach is used, where
the number and entrance order of cognitive
elements included in the model is allowed
to vary for each individual, than when a
cross-sectional procedure is utilized,
where the respondents are aggregated and
the number and entrance order remain the
same.

Ho: The predictive efficacy of the Fishbein
attitude model is not affected when an intra-
individual analytic approach is used, where
the number and entrance order of cognitive
elements included in the model is allowed to
vary, than when a cross-sectional procedure
is utilized, where the respondents are
aggregated and the number and entrance order
remain the same.


Hypothesis 3

Ha: Cognitive differentiation is generalizable
across two distinct product categories, one
high involvement product category and one low
involvement category, and interpersonal
relations.

H : Cognitive differentiation is not generalizable
across the two distinct product categories
studied and interpersonal relations.

Each of these hypotheses is discussed in greater detail in

the subsequent section describing the operationalization

and analysis of the hypotheses.








Marketing Implications

If the null hypotheses are rejected in favor of the

alternate ones, this study holds implications for the

academician studying consumer behavior as well as the

marketing practitioner.

Analysis of hypothesis 1 will provide insight into

whether or not cognitive differentiation, a structural

variable of cognition, influences the number of dimensions

or cognitive elements that provide the best prediction of

attitudes. This information will provide greater insight

into how attitudes are developed, how they can be changed,

and how to more effectively measure and predict them.

Furthermore, assuming a positive relationship between

cognitive and the number of optimal elements entered into

the attitude model, a decision rule can be developed in

terms of how many cognitive elements to include in the

attitude model.

The first step in using this research, if hypothesis 1

is supported, is to determine if cognitive differentiation

is linked to easily identifiable variables. Linking cogni-

tive differentiation to these variables provides the market-

ing practitioner with the basis for identifying segments of

the market who are differentiated and those who are not.

Different marketing programs and strategies can then be

directed at the relevant target market.









An individual who is cognitively differentiated, for

example, may require more information along more product

attributes for a change in attitude to occur than will the

less differentiated individual. If differentiated individu-

als are the target market for a particular product, the

marketing practitioner would realize that greater emphasis

must be placed on the dissemination of product information

to those individuals than may normally be required. In

other words, greater emphasis would have to be placed on

the promotional program with a resulting increase in pro-

motional expenditures. If, on the other hand, the cost of

providing additional information becomes prohibitive, a

company might do well in aiming only at a less differenti-

ated segment of consumers when an attitude change strategy

is deemed necessary. In either instance, knowing the

differentiation level for the target market should aid the

practitioner in improving the impact of a promotional

program by providing insight into the amount of information

required.

With respect to hypothesis 2, if an intra-individual

analytic procedure is found to be superior in predicting

attitude, its use should provide greater confidence in

using attitude models for developing and improving marketing

strategies. This is important both for the academic

researcher and the marketing practitioner.




45



If the concept of cognitive differentiation is

generalizable across different cognitive domains (hypothesis

3), the use of cognitive differentiation would be greatly

facilitated. An academic researcher or marketing practi-

tioner would not have to be concerned with whether or not the

influence of cognitive differentiation deviates between

different product categories. Rather, they would realize,

for example, that regardless of the product category more

information is needed for a differentiated consumer than

for an undifferentiated one.















CHAPTER IV


METHODOLOGY


Research Design


Product Selection

Two products were selected as focal points of this

study. In light of the generalizability hypothesis (hy-

pothesis 3), the two products were selected primarily on

the basis of representing two distinct product categories.

The criteria used for the selection of one product category

were as follows: low unit price, purchased relatively

frequently, low risk decision, and where peer group pressure

would be minimal. In purchasing a brand in this product

category, the level of involvement in the decision-making

process would be expected to be relatively low. The criteria

for the second product category were as follows: high unit

price, purchased on an infrequent basis, high risk decision,

and where peer group pressure would be expected to play a

role in brand selection. A relatively high degree of in-

volvement in the decision-making process would be expected

for this product category. Use of such distinct product

categories permitted an examination into whether differenti-

ation is generalizable. In both cases, the product

46









categories had to be ones in which consumers hold attitudes

which they can reasonably express.

Two product categories which fit the above criteria

are automobiles and toothpaste. An automobile is a highly

visible and high risk decision whereas the type of tooth-

paste purchased has little social visibility and financial

risk. Most people, including those respondents who par-

ticipated in this study, hold attitudes and beliefs about

the various brands within each product category and should

be able to express them.


Sample Selection

The sample consisted of 102 undergraduate males

enrolled in the College of Business and Administration at

Southern Illinois University at Carbondale. Males were

selected so that differences between attitudes and variations

in differentiation due to sex would not be introduced into

the data.

A convenience sample rather than a random procedure

was employed due to length of the questionnaire. To avoid

respondent non-interest and boredom which might result in

haphazard responses, each respondent who completed the

questionnaire was paid $3. Since the purpose of the study

was to examine the theoretical relationship between differ-

entiation and attitudes and not an attempt to generalize

to a larger population, the use of students and a convenience

sample was not viewed as a limiting factor.









Instrument Design

Eleven instruments were utilized in collecting the

data necessary to test the three hypotheses set forth

previously. They included a measure of (1) beliefs about

the toothpaste brands (Fishbein's B. measure), (2) the
1
evaluative aspect of the beliefs about toothpaste brands

(Fishbein's ai measure), (3) attitudes towards the tooth-

paste brands, (4) differentiation measure for toothpaste,

(5) determinism measure for toothpaste, (6) beliefs about

the automobile brands (Fishbein's B. measure), (7) the
1
evaluative aspect of the beliefs about automobile brands

(Fishbein's B. measure), (8) attitudes toward automobile

brands, (9) differentiation measure for automobiles,

(10) determinism measure for automobiles, and (11) differ-

entiation measure for interpersonal relations.

To develop these measures, two preliminary protests

were administered for each product category. Following

Kaplan and Fishbein [40], an unstructured, free-recall

questionnaire was presented to 73 male students in order to

elicit the salient attributes and the relevant brands to

be included later in the research questionnaire. The

respondents had three minutes to answer each question. The

free-recall questions for toothpaste and automobiles

were:









1. Please list what you believe to be the charac-
teristics and attributes of toothpastes.

2. Please list the brands of toothpaste that you
can recall.

3. Please list what you believe to be the charac-
teristics and attributes of automobiles (with
standard equipment) within the $2300 to $3500
price range.

4. Please list the brands of automobiles (with standard
equipment) that you believe fall within a $2300 to
$3500 price range.

The resulting list of salient attributes for tooth-

paste, which was included in the final research question-

naire, is presented in Table 2 along with the percentage of

times each attribute was elicited. The frequency with which

each attribute was elicited ranged from a minimum of 26.4%

for price to a high of 68.1% for taste/flavor. It is

interesting to note that a majority of these attributes

have also been included in other marketing related studies

[e.g., 10, 23].


TABLE 2

ATTRIBUTES ELICITED FOR TOOTHPASTE

Attributes Percentage of Times Elicited
taste/flavor 68.1%
freshen breath 62.5
cavity prevention 54.2
clean teeth 45.8
whiten/brightena 37.5
texture 31.9
color 30.6
price 26.4
Over 50% of the respondents who said whiten/brighten also
included clean teeth, so the two attributes were deemed to
be different on an a priori basis.









Eight brands of toothpaste were selected for inclu-

sion on the basis of question 2 above. These brands were

Close-up, Colgate, Crest, Gleem II, McClean's, Pearl

Drops, Pepsodent, and Ultra Brite. Each brand was mentioned

a minimum of 20 times (28%) and is a well-advertised

national brand with differing advertising appeals.

Table 3 shows the salient attributes for automobiles

and the percentage of time each was elicited in the pre-

liminary pretest. Each attribute was elicited by a minimum

of approximately 18% of the respondents, in the case of

handling, and reached a high of 50.7% for gas mileage.

Again, many of these attributes have been used in other

attitude studies [e.g., 51].


TABLE 3

ATTRIBUTES ELICITED FOR AUTOMOBILES

Attributes Percentage of Times Elicited
gas mileage 50.7%
economy 35.6
size 35.6
dependability 35.6
quality 30.1
style 24.7
performance 23.3
comfort 21.9
luxury 19.2
handling 17.8


In an attempt to ascertain what influence the "price

restriction" in question 3 had on the attribute elicitation

procedure, a follow-up study was employed. The same question









without any price restriction was administered to essen-

tially the same respondents two weeks after the first one.

A Spearman rank correlation coefficient of .76 (p < .01)

was found for the two series of attributes ranked according

to the number of times each was mentioned. The major

difference between the two was price. Price was ranked

eleventh with seven respondents mentioning it in the "price

restriction" questionnaire compared to seventh with 23

respondents when no price restriction was given.

In the final research questionnaire the price re-

striction was used so that the attributes associated with

the brands would be more homogenous. The concern was that

the attributes might not be stable or might differ between

high priced prestigious automobiles and those automobiles

which college students are more likely to purchase. There-

fore, by including only those brands in the relevant price

range this potential problem was circumvented.

Ten makes of automobiles were selected for inclusion

in the questionnaire. They were VW Beetle, Cutlass, Dart,

Javelin, Maverick, Mustang II, Pinto, Nova, Road Runner,

and Vega. These makes were included due to their familiarity

to the subjects, as indicated by question 4, and because

they represented several major automobile manufacturers with

a wide variety of prices and images within the price range

of interest. In other words, these 10 automobile makes








provided a representative sample of those available in the

relevant price range. The number of makes was limited to

10 so that the size of the questionnaire would not be

unwieldly or create undue respondent fatigue.

After these preliminary protests were completed, the

final questionnaire was developed. Form A in Appendix I-A

shows the Fishbein differentiation, determinism, and

attitude measures for toothpaste. Form B includes the same

for automobiles (Appendix I-B).

The belief statements (Bi scale) for both product

categories was a seven-point scale ranging from -3 to +3

indicating the probability that a given product attribute

(e.g., freshen breath for toothpaste) is associated with a

brand (e.g., Crest). The evaluative aspect of the beliefs

(ai) was also a seven point scale which ranged from -3 to

+3. This component indicates the respondent's evaluation of

the product attribute from bad (-3) to good (+3) regardless

of the specific brand.

The attitudes toward each brand of automobile and

toothpaste was measured in terms of as unidimensional

measure of affect similar to that used by Bither and Miller

[18], Klippel [47], and Mazis and Klippel [51]. This

measure was a seven-point scale for each brand with one

pole labeled "extremely low appeal" and the opposite pole

"extremely high appeal."









Determinism scores for each attribute were ascer-

tained by Alpert's [5, 55] "Dual Questioning Method," The

information for this procedure was collected for use in

determining the entry order of the cognitive elements into

the predictive equation of the attitude model.1 In the

"Dual Questioning Method" two questions concerning each

product attribute were asked. First the subjects were asked

how important each attribute was in selecting a product

and, secondly, whether they felt there were many differences

between brands on that attribute. According to this pro-

cedure an attribute would not be considered a determinant

attribute, even though it might be rated as extremely

important if no differences were perceived between the

respective brands.

Bieri's Reptest was used to measure cognitive differ-

entiation for both product categories and interpersonal

relations. For automobiles the Reptest was comprised of

a 10 x 10 matrix with brand names and product attributes

substituted for role persons and personal constructs. In

the case of toothpaste the matrix was 8 x 8 and again

the brand names and attributes replaced the interpersonal

elements in the instrument.



In addition to the determinism approach of ordering
the entry of the cognitive elements into the equation, other
methods were also employed. Each of the entrance procedures
and the relevance of the entry order is discussed under hy-
pothesis 1 in the Hypothesis Operationalization and Analysis
section.








Essentially the same brands and product attributes

contained in the attitude instruments were used in the

Reptests because they were operationally defined as salient

in the free elicitation procedure for attitudes. These

brands and attributes should provide a more valid measure

of differentiation than if non-salient attributes or brands

were included. The product attributes, however, were not

couched in the same terminology as those in the attitude

model.

Cognitive differentiation is a measure of the re-

spondent's ability to discriminate between brands on the

amount of a given attribute present in each brand. There-

fore, the Reptests presented asked how much of an attribute

does a given brand contain. A value of +3 indicates the

brand contains very much of the attribute and a -3 value

indicates the brand does not contain the attribute. No

neutral point is provided. The Fishbein attitude model,

on the other hand, is interested in the probability or

likelihood that a relationship exists between a product

attribute, regardless of the level of the attribute, and a

given brand and whether possessing that attribute is good or

bad (the evaluative dimension). The use of these two

instruments, while measuring two different things, facili-

tates the comparison between the two psychological concepts

of attitude measurement and cognitive differentiation more









than some other measure of differentiation such as that

developed by Scott [71].

The bipolar extremes used in Reptests were elicited

in a fashion similar to that developed by Ahtola [1]. For

automobiles (Appendix I-B,Form B) and toothpaste (Appendix I-A,

Form A), the respondents were asked to describe the brands

of toothpaste and automobiles with respect to several

characteristics. Furthermore, they were asked to describe

the brands only in terms of characteristics and not in

terms of their preference. These questionnaires were

administered to the same respondents one month after the

original salient attributes were elicited. The bipolar

extremes selected were those elicited most by the

respondents.

The only modification to the original Bieri Reptest

(Appendix I-C)for interpersonal relations used in this

study was the substitution of "instructor" for "boss" as

a role person. Since a substantial proportion of students

had not held a full-time job for a period longer than two

months of summer employment, a change was deemed necessary

because they had little work experience on which to rely in

differentiating "boss" from other roles. Both the original

and modified Reptests were 10 x 10 matrices with 10 role

persons and 10 bipolar constructs.


Scott's method of measuring differentiation was pre-
sented earlier in the discussions under the heading Cognitive
Differentiation: Measurement and Meaning.








In administering the questionnaire, half of the

respondents were asked to complete Form A (toothpaste)

first and half Form B (automobiles). This was done to

minimize any systematic biases which might occur by having

all respondents answer one of the forms first. The Reptest

for interpersonal relations was last in each instance. The

presentation order of the various brands and attributes was

varied in a random fashion and all brands were rated within

attributes, i.e., the brands were rated for each attribute

rather than rating each brand along all attributes. These

precautions were taken in an attempt to reduce any possible

halo effect by reducing the opportunity to compare the

responses with prior ratings [86].

The parts of the questionnaire pertaining to Fishbein's

belief statements (Bi) and the Reptest were separated by

other instruments to minimize any possible perceived simi-

larity between the measures and resulting respondent con-

fusion. The instructions and operations were also quite

different. The order of the various measures in Forms A

and B was as follows: (1) the evaluative aspects of the

beliefs, (2) belief statements, (3) attitudes toward brands,

(4) importance measure for the determinism score, (5)

difference measure for the determinism score, and (6) differ-

entiation measure.








Instrument Pretest

A pretest was conducted in order to ascertain whether

the instructions were clear and therefore elicited the

appropriate responses and if the time necessary to complete

the questionnaire was reasonable. Five College of Business

and Administration male students from Southern Illinois

University at Carbondale volunteered and were compensated

for participating in the pretest. Following the completion

of the questionnaire, each respondent was asked to express,

in his own words, why he responded the way he did to each

set of questions. Furthermore, each respondent was asked

to indicate if any questions and instructions were not clear

and, if so, how they could be improved. Each of these

respondents came from the same population as those who

participated in the final study.

Only minor modifications were necessary because it

appeared that the respondents had no difficulty in compre-

hending the questions and were answering them properly.

The time required to complete the questionnaire ranged

from 22 to 45 minutes with only one individual taking more

than 40 minutes.


Hypothesis Operationalization and Analysis


The purpose of this section is to describe the statis-

tical procedures employed in examining the three research

hypotheses. In each instance the alternative form of the









hypothesis is presented and operationally defined, followed

by discussion of the statistical methodology.


Hypothesis 1: Cognitive Differentiation
and the Number of Cognitive Elements

A respondent's cognitive differentiation score, based

on a modification of the Bieri Reptest (see Chapter IV)

constitutes the predictor (independent variable) variable

in hypothesis 1. The optimal number of cognitive elements

used for predicting a given attitude represents the cri-

terion (dependent variable). Cognitive elements are

defined as the B.a. components in the Fishbein attitude
x1
model. The "optimal" number of such elements is operationally

defined as the number yielding the best correlation between
n
the elicited and predicted ( Z Bia.) measures of attitude.
i=l1
What constitutes the best correlation is based on six

separate criteria which are described in subsequent para-

graphs.

The alternative form of the hypothesis states that

there is a positive relationship between cognitive differ-

entiation and the optimal number of cognitive elements used

in the Fishbein attitude model to predict a given attitude.

A low score derived from the Reptest indicates a high degree

of differentiation whereas a high score indicates a low

degree of differentiation. Therefore, the alternative form

of the research hypothesis suggests a negative correlation








between the cognitive differentiation score and the optimal

number of cognitive elements for the Fishbein model.

The examination of this relationship requires two

stages of analysis. The first involves ascertaining the

optimal number of cognitive elements to include in the

equation for predicting an attitude and the second involves

correlating this number with the differentiation score.

To determine the optimal number of cognitive elements,

each of the elements was introduced into the equation one at

a time, according to a specified entry criterion, until all

of the elements were included. Each time another element

was added, the elicited attitude toward the brands was

correlated to the predicted attitude. This entrance pro-

cedure is shown in Figure 1 in flow chart form for auto-

mobiles.

For the purpose of this study, six entrance criteria

or decision rules were used in defining the optimal number

of cognitive elements because no single criteria was without

possible limitations. The entrance criteria, with one

exception, were applied in two ways. In one, the entrance

order of the cognitive elements was permitted to vary

between individuals. In the second, the entrance order of

the cognitive elements remained stable across all respondents.


This staged entrance procedure is similar to that used
by Willie and Weinreich [88].











FIGURE 1

STAGED ENTRANCE PROCEDURE








Enter ith cognitive element
on the basis of a specified decision criteria



n
Compute the E B.a.
i = 1



n
Correlate the E B.a. value
i = 1
with the unidimensional measure of
attitude (elicited attitude)

I


yes i<10
hen

no







For example, in the first way cognitive element number two

might enter into the predictive equation first for respondent

five but fifth for respondent 20. In the second way, the

cognitive element would be entered into the equation in the

same order for both respondents.








The concern was to utilize several criteria to guard

against accepting or rejecting the hypotheses based on

results that might be influenced by choice of entrance

criteria. At the same time, however, they were similar

enough to permit a comparison of the results across the

various entrance criteria.

The six basic criteria were

1. Peak Salience

2. Peak Determinism

3. Peak Regression

4. Significant Salience

5. Significant Determinism

6. Significant Regression.

The first three measures are labeled "peak" because the

number of optimal elements, subsequently compared with the

differentiation scores, was simply that which provides the

highest absolute correlation coefficient between the elicited

and predicted attitude measures. This is an approach similar

to that used by Wilkie and Weinreich [88], which does not

take into consideration whether the addition of one more

cognitive element increases the predictive efficacy of the

model significantly. Using a peak approach, if the addi-

tion of one element increased the correlation from .58 to .59

between the elicited and predicted attitude measures, then

that number of cognitive elements would be used even though

the .01 increase may in fact be meaningless.








The last three measures labeled "significant,"

however, take into consideration whether the addition of a

cognitive element increases the correlation coefficient

significantly. These latter measures decrease the likeli-

hood that the number of elements included in the analysis

is the result of a spurious and insignificant increase in

the correlation coefficient.

As elements are added to the equation, the correlation

coefficient may either increase or decrease. Hence a number

of "peaks" may exist as illustrated in Figure 2. The "sig-

nificant" designation pertains to the number of cognitive

elements producing the highest correlation coefficient that

is significantly greater, in a statistical sense, than the

next lowest coefficient.

To determine the optimal significant number of ele-

ments to be included in the next stage of analysis, the

correlation coefficients corresponding to various numbers

of cognitive elements in the attitude model were compared.

For example, point b in Figure 2 would first be compared to

point a. If point b was significantly higher, then c would

be compared to point b, and so on. In this illustration, if

c was higher than b, but d was not significantly higher than

c, the number of cognitive elements subsequently analyzed in

relation to differentiation would be 4.









FIGURE 2

PLOT OF CORRELATION COEFFICIENTS BETWEEN
THE PREDICTED AND ELICITED ATTITUDE AS
THE NUMBER OF COGNITIVE ELEMENTS VARIES


Correlation
Coefficients

1.00 -


.8


.7


.6 -


.5


.4


.3


.2


I I I I I I I I I I


1 2

Number


I I I I I I I
3 4 5 6 7 8 9

of Cognitive Elements in the Model


The formula, used for comparing the various r's, was

developed by Hotelling and takes the form of a "t" test

[34, p. 190]. This formula tests whether the r's are








significantly different and takes into consideration the

intercorrelation between the r's when the number (i) of

cognitive elements included in the model is different.

The formula is



tdr = (r12 r13) (n 3) ( + r23
1 2 2 2
2(1 r23 r2 r13 + 2r23 r12 r13)


where

r12 is the correlation between the elicited and
predicted attitude at point i (e.g., when two
elements are included),

r13 is the correlation between the elicited and
predicted attitude at point i + n (e.g., when
three elements are included),

r23 is the correlation between the predicted atti-
tudes at point i and at point i + n, and

n is the number of brands being evaluated in the
equation (10 for automobiles and 8 for tooth-
paste).

The alpha level used to determine the significant level in

the analysis was .35 for a one-tailed test. This alpha

level was selected so as not to be so stringent as to

eliminate legitimate increases in r but powerful enough to

take into consideration only spurious increases. Since

this analysis revolves around being able to reject the null

hypothesis that the two peaks are equal when in fact they

are different, a Type II error is of critical importance.

Hence a large alpha level was selected which reduces beta

(the probability of a Type II error).









The term "salience" applied to two of the entry

criteria (peak and significant) means that the cognitive

elements were entered into the predictive portion of the

equation according to the saliency hierarchy. The hierarchy

was operationalized as the number of times each attribute

was elicited in the preliminary phase of the study. In the

case of toothpaste, for example, since taste/flavor was

elicited by 68.1% of those in the pretest it was entered

into the model first, followed by freshen breath with 62.5%

and so forth. The last cognitive element to be entered

was price with 26.4% of the respondents eliciting the con-

struct. The same approach was followed for automobiles.4

Based on the operationalization of the salience procedure,

the cognitive element entrance order did not vary between

respondents.

The entrance procedure designated as "determinism"

was based on a measure developed by Alpert [5, 55]. This

measures the importance of each attribute and whether the

respondent perceives differences among the brands with

respect to that attribute. Although the attribute is very

important, theoretically if there are no perceived differ-

ences aong the brands along the attribute, it would not be

determinant in the formation of an attitude. Determinism

score for each respondent was computed according to the formula


The ordering of economy, size, and dependability of
automobiles, since each was elicited the same number of time,
were randomly assigned.









D. = P. x I.;
1 1 1

where

D. is the determinism score for attribute i,

Pi is the perceived difference between brands
along attribute i, and

I. is the importance of attribute i.


The attribute with the highest mean determinism score

was entered first, the one with the second highest mean

score was entered second into the attitude model, and so

forth. This procedure is similar to that of Wilkie and

Weinreich [88] although they operationally defined P.
1
differently than did Alpert.

When the entrance of the cognitive elements were

allowed to vary on an intra-individual basis, the deter-

minism score for each attribute for each individual was

used. In the case where the entrance order remained the

same across individuals, a mean determinism score for

each attribute across respondents was computed.

The final entrance procedure is designated "regres-

sion." This approach was operationalized through the use



5Wilkie and Weinreich operationalized Pi by taking
the standard deviation of the attribute ratings across
stores rather than asking the respondents to state the
extent to which they perceived differences among the stores
along an attribute.








of a stepwise regression procedure, BMD02R [25]. In this

procedure, the order of variable insertion into the regres-

sion equation followed the rank order of the partial cor-

relation coefficients, beginning with the largest and

concluding with the smallest. The partial correlation

serves as a measure of the importance of each of the

variables [26, p. 69] which follows conceptually the

salience and determinism entrance procedures. The elicited

measure of attitude was the criterion or dependent variable

and the cognitive elements were the independent or predictor

variables. The order in which the cognitive elements were

inserted in the regression equation was the order followed

in entering them into the predicted portion of the attitude

model. This stepwise entrance procedure is an effective

complement to the theoretically based determinism and

saliency procedures.

Where the entrance order was allowed to vary from

individual-to-individual, the stepwise procedure was applied

to each individual to define the ordering for that individu-

al. When the entrance order remained the same across all

respondents, the stepwise procedure was applied to the

aggregated sample to determine the element entry order.



While the stepwise regression procedure may not pro-
vide an optimal solution, it is recommended over the other
variable selection procedures by Draper and Smith [26,
p. 172].








Once the optimal number of cognitive elements was

ascertained for each entrance procedure the final stage in

testing hypothesis 1 was completed. In this step the

optimal number of cognitive elements, derived earlier, was

correlated with the respondents' respective cognitive

differentiation scores for toothpaste and automobiles.

Ten correlation coefficients for each product category

resulted for each of the two product categories, four

correlation coefficients for the determinism and regression

entrance procedures, and two for the salience procedure.7


Hypothesis 2: Predictive Efficacy of the
Intra-Individual versus Cross-Sectional
Analytic Procedures in the Attitude Model

The difference in the predictive efficacy of the Fish-

bein attitude model when two different analytic techniques

are used is the focal point of this hypothesis. The alter-

native form of the hypothesis states that the predictive

efficacy of the Fishbein attitude model is greater when an

intra-individual analytic technique is used, where the

number and entrance order of the cognitive elements included

in the model are allowed to vary for each individual, than

when a cross-sectional procedure is utilized, where the



Again, the salience procedure did not permit the
entrance order to vary between individuals based on the
elicitation procedure. The entry order is the same regard-
less of whether or not the entry order is stable or is
permitted to vary across individuals.








respondents are aggregated and the number and entrance

order remain the same.

It should be noted at this juncture that there are

two sources of variation in the predictive efficacy of the

attitude model. One is the variation in the cognitive

element entrance order. The second is the variation

between individuals in the number of elements entered into

the model.

In the operationalization of this hypothesis, pre-

dictive efficacy was defined as the correlation between

the elicited and predicted measures of attitude. The intra-

individual analytic approach required that each respondent

be treated as a subsample in correlating the elicited and

predicted measures of attitude. Hence, a correlation

coefficient for each respondent resulted for each cognitive

element entrance procedure (e.g., determinism, salience).

In the case of the cross-sectional procedure, one correlation

coefficient resulted for each of the cognitive element

entrance approaches.

The correlation coefficients from the intra-

individual and cross-sectional analysis were transformed

using Fisher's r to Z procedure before the two sets of

data were compared. This transformation was necessary,

prior to a statistical comparison, because correlation

coefficients have skewed distributions. In the absence of








transformation, statistical tests comparing the mean

correlation coefficients would not satisfy the normality

assumption. The relationship of a correlation coefficient,

r, to Fisher's Z, with Z approximately normally distributed

with a variance of is [34, p. 163]



Z = 1/2 [loge (1 + r) loge (1 r)].



A series of Z tests were used to test the differences

between the correlation coefficients derived from the cross-

sectional analysis and the intra-individual approach. For

each cognitive element entrance procedure, the Z score of

the correlation coefficient derived from cross-sectional

analysis was compared against the mean transformed Z scores

derived from the intra-individual analysis. These tests

were made using the formula [34, p. 190]

Z Z


S- 3 n 3
1 2


for two unmatched samples of Fisher Z's. A separate test

was performed on the correlation coefficients arising from

each cognitive element entrance procedure.








Hypothesis 3: Generalizability of
Cognitive Differentiation

The acceptance or rejection of this hypothesis is

based on the positive intercorrelation of the differenti-

ation scores derived from three different cognitive domains.

The hypothesis in alternative form states that cognitive

differentiation is generalizable across two distinct product

categories, one high involvement product category and one

low involvement category, and interpersonal relations.

In operational terms, this hypothesis states that a

person who exhibits a relatively high degree of differenti-

ation will tend to have similar levels of differentiation

in other domains. Those with low differentiation in a

domain will tend to be low in others. The three cognitive

domains, in this study, were automobiles, toothpaste, and

interpersonal relations. The differentiation scores for

each were derived from a Reptest.














CHAPTER V


RESULTS AND DISCUSSION


Hypothesis 1: Cognitive Differentiation and
the Number of Cognitive Elements


In analyzing hypothesis 1, the results on toothpaste

and automobiles will be presented separately. Once both

sets of findings have been discussed a comparison of the

findings will be made.


Toothpaste

The optimal number of cognitive elements was posi-

tively related to cognitive differentiation in each of the

six entrance procedures, as suggested by the alternative

hypothesis. Whether the entrance order was permitted to

vary between respondents or whether each element was

entered in the same order across respondents did not in-

fluence the acceptance or rejection of the hypothesis. Only

when the significant determinism entrance procedure was

employed, however, was the relationship between cognitive

differentiation and the optimal number of cognitive elements

included in the attitude model significant at the .05 level.

The correlation coefficient, as shown in Table 4, was -.31




















































































O D


-4j
0






0 4- v
v m


0
4--


C)


0












O)
1)
r-
o

4



4J



-)




-4





C)

04
4-P

0
co







a)



0



0

-4




HO
-)-
(d
s-C-













0 (0
0

0-



0
( 1)


0

O




-C>
0
U E








when the cognitive elements were entered in the same order

and -.223 when the element order was allowed to vary

between respondents. A negative sign resulted because a

low numerical differentiation score stems from a relatively

high degree of cognitive differentiation.

The remaining significant correlation coefficient,

significant at the .10 level, existed in the case of the

peak determinism entrance procedure when the entrance order

of the cognitive elements was allowed to vary. None of

the three correlation coefficients, even though they were

significant at the .10 level or greater, have an r amount

of variance explained, greater than .10. These results

provided only partial support for the hypothesis that a

positive relationship exists between the optimal number of

cognitive elements and cognitive differentiation and tooth-

paste. If the r 's were larger or if significant relation-

ships had been shown in other entrance procedures, then the

alternative hypothesis could have been accepted.

It does not appear, furthermore, that allowing the

cognitive elements to vary between respondents enlarged

the correlation between differentiation and the optimal

number of elements in the attitude model. The correlation

coefficients within each entrance order procedure (elements

allowed to vary versus elements in the same order) were not

significantly different (p<.05) from one another. No pattern








seemed to be present, in that for half of the entry methods

the correlation coefficients were higher when the elements

were entered in the same order and for the other half when

the element order varied.

The coefficients presented in Table 4 were based on

99 respondents who were included in the toothpaste portion

of the study. Three were discarded out of the 102 completed

questionnaires because the correlation between the pre-

dicted and elicited measures of attitude was undefined.

This was due to the absence of any variance between the

cognitive elements in evaluating the eight brands of tooth-

paste.


Automobiles

The analysis yielded very little evidence that the

optimal number of cognitive elements was significantly

related to cognitive differentiation for automobiles.

Although the signs of all but one of the correlation coef-

ficients were in the predicted direction, the value of the

coefficients shown in Table 5 (based on 102 respondents)

were not significant at the .05 level. This statistical

evidence dictated acceptance of the null hypothesis of

no relationship between the optimal number of cognitive

elements and cognitive differentiation for this product

category.




























ZH

ao




Ha
U) o4




SE-H
P0


U C
044
0
0






0H

a
0


OH

HM>




OH
04






a
ta1



0(





U


0)

O
0


'4
41



fL



4-)









r-


























0
0)
a,












0a

















4-I
r0
















4
0)



0)
0
4o





o

m











a0)4

U40
0c0

00l4-
040








The results for toothpaste and automobiles, evaluated

together, seemed to hint at some relationship between

cognitive differentiation and the optimal number of cogni-

tive elements. No element entrance approach, however,

provided significant results both for toothpaste and auto-

mobiles. Only in the case of the significant determinism

entrance procedure were significant correlations found for

toothpaste. If other entrance procedures had produced

significant results for toothpaste but not automobiles,

one might argue that the relationship between cognitive

differentiation and the optimal number of elements is product

specific. Since this was not the case a strong relationship

did not appear to exist between cognitive differentiation

and the number of elements included in the model.


Hypothesis 2: Predictive Efficacy of the
Intra-Individual versus Cross-Sectional
Analytic Procedures in the Attitude Model


To test whether the predictive efficacy of the Fish-

bein attitude model is greater when the analysis is on an

intra-individual basis than on a cross-sectional basis,

correlations between the predicted and elicited attitude

were calculated for each individual separately and for the

sample as a whole. These two sets of correlation coeffi-

cients were then compared. The discussion of results will

be separated on the basis of the two product categories

examined.








Toothpaste

The coefficients for the cross-sectional (respondents

aggregated for analysis) and the intra-individual (attitude

model applied to each respondent individually) analyses

(Table 6) were compared through a series of Z tests set out

in Chapter IV. The mean correlation coefficients for the

intra-individual analysis (99 respondents) were derived by

transforming the original r's to Fisher's Z scores. Then

the mean of the Z scores for each entrance procedure was

computed. These Z scores were transformed into the r values

which are shown in Table 6. These mean correlation co-

efficients were based on the distribution of coefficients

derived for each respondent and are shown in Appendices II-A

(elements allowed to vary) and B (elements entered in the

same order).

In testing this hypothesis, the cross-sectional

coefficient for the salience entrance procedure (.463 in

Table 6), for example, was compared to the peak and sig-

nificant salience coefficients (.751 and .735, respectively)

derived from intra-individual analysis. Similar compari-

sons were made for the regression and determinism procedures'

coefficients, resulting in a total of 10 tests for differ-

ences between coefficients corresponding to tne cross-

sectional and intra-individual analyses for each procedure.





79


4-
00 >, Uen--



0 0



O 0O


Q 0 0 O0"








M -,-- L O Q
i -~o e H 0 .
u m 0) O )










K H e 0-4 ) o
n 4- *--

H Ol 0 0















a--
r .4 o o 4i )n
a a4 41 C




O m 00 OM-
SE -i HO -
4- 0 :

-! 4- (0 e
00 H [0 N-
C) 44O E 4J O





> u 1 -, Q )
L: o Qe-1 .0 0 H

xw .1 0 7 4- 4
0 0 'i n 5 ) -400







0 w ) C 0 -W
0 0n (n U) 0 4
H H )l 0) 44
CD 0 00 C t7nH


HO en in Q004 -4i
0E H 0U m 'e n


OW 0 0 w(o



O ) '0 U 0 0

0O 00 0 l -4M
H '0 Q) 41 r45 U(0
Q) H> M 4-0

4J Q 4-4 0 O 4-400
-o 0 0 0 oen 0 0 I



rH -0 4 0 u 4H




0 r eo u 4 0








In each of the 10 comparisons, the mean correlation

for the intra-individual analytic approach was signifi-

cantly greater (p<.05) than the correlation derived from

the cross-sectional procedure. The null hypothesis, there-

fore, is rejected in favor of the alternative hypothesis

which states that the predictive efficacy of the Fishbein

attitude model is greater when an intra-individual analytic

approach is used than when a cross-sectional procedure is

implemented for toothpaste.


Automobiles

The coefficients for the cross-sectional and the

intra-individual analytic procedures for automobiles

(Table 7) were analyzed in the same fashion as for tooth-

paste. In each instance, the correlation of the cross-

sectional analysis entrance procedure was compared to the

appropriate intra-individual analysis correlation coeffi-

cients.

The intra-individual mean correlation coefficients

were significantly greater (p<.05) than those coefficients

derived from the cross-sectional analysis in all but one

instance. The intra-individual analysis significant

determinism correlation coefficient was significantly

greater than the cross-sectional procedure at the .10,

rather than the .05, level when the elements were entered

in the same order for all respondents. These results









42

u m

4-4H o0 .Q > 0
0 C O > O
C r 0 0
C., O
--4 -c
(1 U) 4 4
Q) C (

CO 0
H H to




0 00

H4J -1 -1
S -Ho I C 04




H m 4- 1 o o J -J
Z rO 4-4 N .0Q C 0


0 C) O' 0



CC 0 4 0 4
H -.-- 0 0 4
HZ -H N 0-4

0 r10 0 2H








< r m ol o 0 c,
ZO LO 0-H H




0 0 o -, 00
N N W C0









E 0 4 t- 1-
H HO 00 r- 0 -H --0
4 0 OC 0 0i
o H4C r- d 4 42)t
H NI C ] 2 00

NQ p 4 C t-H H0


HZ m C N H o
SQO 40 0 N 0 0
Cli-l 0N0 4-4 300
) O *1 ** H





0 0 0 41 -.1
U) 41 -H- 4O

.w )T ) 0 0 4-1
CI Ol 0 C
U 0-H 0 m 0 0o O0
c -0 0m0 rt O


K4 P V)0 0 Q)
0 0 4-1 H E- t o


0 0 0 0 0 0I
H 40 > 0 C 40 0-
SC 0 -H CO 1C 0*
C to 0 0 (U 3 OC
H 420) >n Q4 0 01 0
0 00 m 0 to 0 -4





42 0 0H 0 H 0-H
In o r 1 a i c i





C HC H-i 0 r-> H- Q t 2n
H -1 4) a U 0 .0
> c En 0 cc )c 4- D u 0




r- r-a -A 3 U) 4-J
4j aj, w ro o a) m








provided strong support for the rejection of the null

hypothesis that the intra-individual approach does not

improve the predictive efficacy of the attitude model over

the cross-sectional approach for automobiles.

One possible explanation for the significant differ-

ences in the predictive efficacy of the Fishbein attitude

model when the analysis proceeds on an intra-individual

basis versus a cross-sectional one has been provided by

Bass and Wilkie [11]. The argument posed by Bass and Wilkie

is that normalization of the responses adjusts for within-

subject variance in responses. This is necessary because

different individuals may have different anchor points and

response sets in evaluating brands and product attributes

which would not be reflected in a cross-sectional analysis.

Normalization of the responses, accordingto Bass and Wilkie,

will eliminate this problem and increase the predictive

efficacy of the attitude model when a cross-sectional

analysis is used.

The formula proposed by Bass and Wilkie to normalize

the belief (BN) of the kth consumer of the ith attribute

for the jth brand is [11, p. 265]


BNijk = Bij. /j Bi .
i]k ]k ] ij]k

Similarly the formula for normalizing the importance

weight (IN) is [11, p. 265]








INik = Ijk/Ei Ilk'


It should be noted that a modification of Fishbein's atti-

tude model was employed in the Bass and Wilkie study where

the belief and importance measures ranged from 1 to 7

rather than from -3 to +3 as in the case of the original

Fishbein model. The Fishbein attitude model cannot be

normalized in the Bass and Wilkie manner because of the

way in which the components of the predicted attitude

(Bi and a.) are measured. Adding a negative response

in the Fishbein model very likely reduced the denominator

in the Bass and Wilkie formula and therefore artificially

inflates the importance or belief measure.

One alternative to the above approach to normalizing

the predicted measure of attitude for this study is to

standardize the belief and evaluative components on the

basis of Z scores. This standardization approach, unlike

the one proposed by Bass and Wilkie, allows for negative

and zero responses. The formula for standardizing the

belief ( B) of the kth consumer of the ith attribute for the

jth brand is


X.j Xik
ijk ik
^k---^ ---k









where

Xijk is the response of the kth consumer
on the ith attribute for the jth brand,

Xik is the mean response of the kth consumer on
the ith attribute across all brands, and

oik is the standard deviation for the kth consumer
on the ith attribute across all brands.


Similarly the formula for standardizing the evaluative

component (ai) is


Xik. Xk
Zik = ii k
ik ak

where

Xik is the response of the kth consumer on the
ith attribute,

Xk is the mean response of the kth consumer, and

ok is the standard deviation for the kth consumer
across all attributes.


When the predicted and elicited measures of attitude

in this study were standardized in dle manner described

above, the coefficients were lower than those based on the

non-standardized data in all but one instance. The

determinism entrance procedure for toothpaste was the

only case in which the correlation coefficient was higher

on the standardized data than on the non-standardized

data (Table 8). The increase was not significant, however,

at the .05 level.

The elicited attitude was also standardized for each
respondent across brands.




























Z



w 2



4-i H


z 1



H W


C H

m0H
o H
O2UO








H C 1
S 04 0 I


FZ 4 : (n
N FI r
H eO



<0

OZU



HO


0


0 o N ,o


MI

0
C,


0

0
4-1
C')

'HO

,4 m ,D 0%


k0
4-
0l

0
42




0




*-








) O
'a
C











0
41




0














-,4
4) 0
N Q
a
41
0
-

0

NJ
-1C





42 0 0



43 m O v









-H- O

54 0 0) en
O n
'U 3




E-IC)Q CO Q0 0


Uo


(0

U)



.)
0






4o
,Q
O
->
04







0
OU

a)



















.-u
a


0 0
0





,-I
0

















r O

0
.,O






-o
>4i




4-1 o
(0



mIn
0C






u 4











0









Using the Fishbein attitude model employed in this

study, it is apparent that standardizing the data does not

improve the predictive ability of the cross-sectional

approach over the intra-individual technique. A theoreti-

cal explanation for the apparent failure of the standardiza-

tion process rests with the operationalization of the

Fishbein model. The standardization process redefines

the respondent's cognitive element in a fashion that is

contrary to the Fishbein attitude theory. For example,

assume that a respondent has an average belief rating for

a Ford Pinto of -2 across five attributes. By standardizing

the belief ratings, any belief rating of -1 would then become

a positive Z score. Multiplying this positive Z score times

a +2 rating on the evaluative scale (ai) provides a cogni-

tive element with a positive Z score rather than a negative

cognitive element required by the attitude model.

In comparing the results from toothpaste and auto-

mobiles for both the non-standardized and standardized

data, an additional comment needs to be made. For both

the salience and determinism entrance procedures in the

non-standardized cross-sectional analysis (Tables 6 and 7),

the correlation coefficients for automobiles are signifi-

cantly greater (p<.05) than for toothpaste. This differ-

ence is similar to findings of Mazis and Klippel [62].who

observed substantial differences in correlations between

automobiles, toothpaste, and mouthwash.




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs