v. -‘< R

A" ~.
.' -A..'-s l

IE3;-
«4

r "revamp
$24“ w"
. ,; »

A“.
l ” fi;,r\fi.
, “wait? ‘-‘“
‘xug‘m. v.9 "is

 

’ a
v . 3—:
#~ P"

 

 

 

 

 

1" -

:‘»;J.,. - ‘6 . -: “
u . . 1 . ‘a "f ’
y».

\
:3.

‘ a.
*fig.
1- 9" ‘v'é

{cam
Nix

, ‘ .,

1,11%" '

w’xfi"
TI ‘
1“

.g n" 3%“ '
‘ _ tee/{act
, .

 

I
~' at”: 1-1.3

~ - “3"
w”
. v."’ gg‘Ef’a-‘éh a
£4 _ ;::\_— 3". " k A."
«WWW fiwg' ' ‘ ‘3' 1'
Jflnr $34" I .‘

2 i.

.,.
1.3+:

Me

I RARIE

ltllllllllllllllllllllllllllljlllll

31293 0104617 6

 

This is to certify that the

dissertation entitled

Parallel Processing of Satisfaction Formation:
An Empirical Analysis of Competitive Alternatives
presented by

Robert D. Mackoy

has been accepted towards fulfillment
of the requirements for

Ph.D. degree in Business Administration

 

we

Major pr ssor

Date AUQUSt 4, 1994

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

 

LIBRARY
Mlchlgan State
University

 

 

 

rue: ll RETURN BOXtomnovothbMouflunywrm
TOAVOID FlNESrottmonorbdmdutodm.

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU lommmmd Oppomlnlylmtltwon
W1

 

PARALLEL PROCESSING OF SATISFACTION FORMATION:
AN EMPIRICAL ANALYSIS OF COMPETITIVE ALTERNATIVES

By
Robert D. Mackoy

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Marketing and Logistics

1994

ABSTRACT

PARALLEL PROCESSING OF SATISFACTION FORMATION:
AN EMPIRICAL ANALYSIS OF COMPETITIVE ALTERNATIVES

By
Robert D. Mackoy

The notion of competition is an integral element of the choice process.
Yet postchoice processes, such as satisfaction formation, are modeled in a
manner inconsistent with the competitive reality in which they, too, occur.
Current models of satisfaction formation focus exclusively on the single,
chosen product while assuming that evaluations of competing products are no
longer salient. This dissertation reviews the satisfaction literature in
marketing and presents a model of satisfaction formation which explicitly
considers competitive alternatives. Parallel yet interrelated paths of
satisfaction formation are hypothesized to lead to satisfaction with the
competing targets (both the choice and nonchoice alternatives) and then to
overall satisfaction. Also addressed are issues related to postchoice processing.
In this dissertation, dissatisfied consumers are hypothesized to process more
extensively than do satisfied consumers. The relative importance of predictive
expectations, desires, attitudes and disconfirmation of expectations as
antecedents of satisfaction are investigated within this framework.

The research hypotheses are tested using data from the 1992 US.
presidential election within a longitudinal research design. Using primarily
LISREL two-group analyses, support is found for each of the following
research hypotheses:

0 Research hypothesis 1: Multiple targets of satisfaction may remain salient

post choice during satisfaction formation.

0 Research hypothesis 2: The processing paths associated with each of the
multiple targets of satisfaction are interrelated, i.e. there are cross-over
effects.

0 Research hypothesis 3: Dissatisfied subjects process more extensively or
completely than do satisfied subjects.

0 Research hypothesis 4: Desires have a greater impact on satisfaction

formation than do expectations.

The dissertation has significant implications for the satisfaction
formation literature. Specifically, all current models of satisfaction formation
focus on the single product or service of interest; yet this dissertation shows
that, in a competitive environment, alternative products or services mt
chosen may also impact satisfaction formation. One conceptual implication is
that the fundamental structure of future satisfaction models should be altered
to include competitive alternatives. In addition, managerial implications
include the necessity of revising both satisfaction monitoring methodologies

and the strategies for managing key marketing mix variables.

Copyrighted by
Robert D. Mackoy
1994

DEDICATION

This work is loving dedicated to my wife, Tina, whose support, patience
and effort made it possible both to contemplate and to complete this
endeavor.

ACKNOWLEDGMENTS

I gratefully acknowledge the contributions of my dissertation
committee, both collectively and individually. Professor Roger Calantone and
Professor Bixby Cooper each provided helpful comments and guidance early in
the proposal stage, and each provided useful comments on successive drafts.
Professor Richard Spreng was extremely generous with his expertise in the
area of consumer satisfaction and with his time: he spent countless hours with
me and with each version of this dissertation. He also provided valuable
insights about the early years of an academic career. To my chair, Professor
Cornelia Droge, I am especially grateful. Her masterful guidance at each stage
of this work - encouraging, discussing, criticizing, challenging, analyzing --
kept me moving in the right direction. She has taught me how to engage the
research process with curiosity, with thoughtfulness, with thoroughness, with
tenacity, with wonder. For this, for expertly managing the committee, and for
sharing her thoughts about thriving in an academic career, I am truly

indebted.

vi

TABLE OF CONTENTS

List of Tables ...................................................... xi
List of Figures .................................................... xiii
I. Introduction and Overview ................................... 1
H. Logic of the thesis: Relation to consumer behavior ....... 2
I-2. Logic of the thesis: Theoretical support .................. 6
[—3. Logic of the thesis: Example ............................ 7
[-4. Four research themes and expected theoretical
contributions ......................................... 10
I-S. Context of the empirical test ........................... 12
[-6. Managerial implications ............................... 14
II. Satisfaction Literature in Marketing
11-]. Introduction ......................................... 19
11-1.]. Scope of the literature review ................. 19
II-l.2. "Traditional" satisfaction model framework ..... 24
11-2. Satisfaction and its measurement ....................... Z7
II-3. Antecedent constructs and their measurement ........... 42
11-3 . 1. Expectation .................................. 42
lI-3.2. Desires ..................................... 56
II-3.3. Attitudes .................................... 60
Il—3.4. Perceived performance ....................... 61
II-3.S. Disconfirmation ............................. 67
11-4. Principle relationships among constructs ............... 78

“-4.1. Antecedents of perceived performance:
Expectations, Desires, and Attitudes ............. 78

vii

II-4.2. Antecedents of disconfirmation:
Expectations, Desires, Attitudes, and Perceived
Performance ................................ 81

II-4.3. Antecedents of satisfaction:
Expectations, Desires, Attitudes, Perceived

Performance and Disconfirmation ............. 86

II-S Other relevant topics .................................. 103

II—S. 1. Model complexity - Structural .................. 103

II-5.1. Model complexity - Processes ................... 106

II-6 Summary of literature review ........................... 108

Key Theoretical Issues Addressed ............................... 114

III-1 . Introduction .......................................... 114
111-2. Thesis themes, research hypotheses, and testable

hypotheses ........................................... 115

III-2.1. Theme No. 1: Multiple targets ................... 115

III-2.2. Theme No. 2: Parallel paths with cross-over
effects ....................................... 124

III-2.3. Theme No. 3: Differential processing due to
ultimate level of satisfaction .................. 129

III-2.4. Theme No. 4: Importance of desires versus

expectations in satisfaction formation ......... 138

III-3 . Models to test hypotheses .............................. 141
Ill-3. 1. Election outcome satisfaction model ............ 141
III—3.2. Campaign process satisfaction model .......... 143
Method ..................................................... 14S
IV-l. Data collection and timing ............................. 145
IV-Z. Operationalization of key constructs .................... 147

IV-2.1. Election Outcome Satisfaction Model -
Operationalization of key constructs ........... 149

IV-2.2. Campaign Process Satisfaction Model -
Operationalization of key constructs ............ 156

IV-3. Other relevant variables ............................... 159

V. Results ...................................................... 162

V—l.
V-Z.

General description of subsamples ...................... 162
Results of the Election Outcome Satisfaction Model ........ 166
V-2.1. Differences between the two groups (EOSM) ..... 166
V-2.2. Results of the EOSM two-group LISREL analysis . . 170
V-2.3. EOSM two-group results related to research

hypothesis #1 ................................ 174
V-2.4. EOSM two-group results related to research

hypothesis #2 ................................ 177
V-2.5. EOSM two-group results related to research

hypothesis #3 ................................ 183
V-2.6. EOSM two-group results related to research

hypothesis #4 ................................ 184
V-2.7. Summary of EOSM results ...................... 186
Results of the Campaign Process Satisfaction Model ........ 186
V-3.1. Differences between the two groups (CPSM) ..... 187
V—3.2. Results of the CPSM two-group LISREL analysis . . 187
V-3.3. CPSM two-group results related to research

hypothesis #1 ................................ 191
V-3.4. Additional observations from the CPSM two-

group analysis ............................... 201
Results of the EOSM LISREL analysis of satisfied Clinton
voters and dissatisfied Bush voters ....................... 202
V—4.1. Analysis of EOSM with satisfied and dissatisfied

V-4.2.

V-4.3.

voters -- results related to research
hypothesis #3 ................................ 203

Analysis of EOSM with satisfied and dissatisfied
voters -- results related to research
hypothesis #1 ................................ 208

Analysis of EOSM with satisfied and dissatisfied

voters -- results related to research
hypothesis #2 ................................ 209

ix

V-S. Results of the CPSM correlation analysis of satisfied
versus dissatisfied voters ............................... 210

V-6. Summary of results relating to the research hypotheses . . . 220

VI. Discussion and Implications ................................... 224

VI-l. The competitive model of satisfaction formation .......... 225

VI-Z. Differential processing ................................ 232

VI-3. Total effects of desires versus expectations ............... 237

Vl-4. Other findings in relation to past research ............... 241
Vl-4.1. The relationship of expectations to

disconfirmation and to satisfaction ............. 241

VI-4.2. Disconfirmation .............................. 244

v1-4.3. Attitudes ..................................... 24s

Vl-4.4. Context of the research ........................ 247

VI-S. Limitations ............................................ 248

VI-6. Concluding remarks ................................... 250

List of References .................................................. 252

\DQVO‘M-fi-WNt—b

r—tr-fii—‘h—it—i
Ari-‘O

15

16

17

18

19

21

LIST OF TABLES

Measures of Satisfaction in Empirical Studies ................... 32
Specific Measures of Satisfaction ............................... 34
Measures of Expectations ...................................... 45
Specific Measures of Expectations .............................. 47
Specific Measures of Performance .............................. 63
Measures of Disconfirmation in Empirical Studies ................ 68
Specific Measures of Disconfirmation ........................... 69
Classification of Studies Which Include Disconfirmation ........... 71
Antecedents of Performance--Relative Importance ............... 79
Antecedents of Disconfirmation—Standards Disconfirmed .......... 82
Antecedents of Disconfirmation-~Relative Importance ............ 83
Characteristics of Studies Testing Antecedents of Satisfaction ...... 87
Antecedents of Satisfaction-Relative Importance ................ 9o
Characteristics of Disconfirmation and Satisfaction Measures

Used in Satisfaction Models ..................................... 95
Operationalization of Constructs - Timing of Measurements

for Each Model ............................................... 148
Operationalization of Constructs - Election Outcome Satisfaction

Model ....................................................... 155
Operationalization of Constructs -- Campaign Process Satisfaction
Model ....................................................... 158
Operationalization of Constructs -- Campaign Process Satisfaction
Model ....................................................... 161
Organization of Chapter 5 ..................................... 163

Description of the Sample --Differences Between Clinton and
Bush Subsamples on Selected Variables ......................... 165

Election Outcome Satisfaction Model -Means and Standard
Deviations of Each Group ...................................... 167

xi

LIST OF TABLES (continued)

22

23
24

26

'3

31

32

33

35

39

Election Outcome Satisfaction Model «Components of Attitude
Towards the Candidate ........................................ 169

Election Outcome Satisfaction Model «Correlation Matrices ........ 171

Election Outcome Satisfaction Model «Within Group Indicators

of Fit ........................................................ 172
Election Outcome Satisfaction Model «Unstandardized Estimates

and Standard Errors of Structural Paths ......................... 176
Election Outcome Satisfaction Model «Common Metric

Standardized Solution ......................................... 178
Election Outcome Satisfaction Model «Psi Matrix Results .......... 182
Election Outcome Satisfaction Model «Total Effects of

Expectations versus Desires .................................... 185
Campaign Process Satisfaction Model «Means and Standard

Deviations of Each Group ...................................... 188
Campaign Process Satisfaction Model «Correlation Matrices ...... 189

Campaign Process Satisfaction Model «Within Group Indicators
of Fit ............................................. . ........... 1 92

Campaign Process Satisfaction Model «Unstandardized
Estimates and Standard Errors of Structural Paths ................ 195

Campaign Process Satisfaction Model «Psi Matrix Results ......... 196

Campaign Process Satisfaction Model «Common Metric
Standardized Solution ......................................... 198

EOSM « Unstandardized Estimates and Standard Errors of
Structural Paths -- Satisfied Clinton Voters versus All
Clinton Voters ................................................ 204

EDSM « Unstandardized Estimates and Standard Errors of
Structural Paths « Dissatisfied Bush Voters versus All
Bush Voters .................................................. 205

Election Outcome Satisfaction Model -- Summary of
Differences in Model Results Using All/Satisfied Clinton
Voters and All/ Dissatisfied Bush Voters ......................... 206

Campaign Process Satisfaction Model «Subgroup Correlation
Matrices ..................................................... 213

Campaign Process Satisfaction Model «Summary of Subgroup
Correlation Analyses ......................................... 217
xii

10
11

12
13

14

15
l6
17

LIST OF FIGURES

Pre-Decision versus Post-Decision Models ........................ 4
Pre-Decision versus Post-Decision « Proposed Model .............. 8

Classification of Satisfaction Formation Constructs

from the Traditional Model ..................................... 26
Targets of Satisfaction: Variations in the Traditional Model ........ 116
Hypothesized Relationships: Satisfaction with Choice, Satisfaction

with Nonchoice, and Overall Satisfaction ........................ 118
Election Outcome Satisfaction Model: Theme No. 1 Research

Hypothesis ................................................... 120
Hypothesized Effects of Satisfaction Antecedents Regarding

Separate Targets on Overall Satisfaction ......................... 122
Hypothesized Paths in the Campaign Process Satisfaction Model

for Testing the Multiple Targets Research Hypothesis ............. 123
Traditional Model: Different Targets Require Parallel

and Unrelated Paths ........................................... 125
Hypothesized Model: Parallel Paths and Cross-over Effects ......... 127
Hypothesized Relationships Among Key Antecedent

Variables and Satisfaction ...................................... 131
Hypothesized Paths for Satisfied Versus Dissatisfied Voters ......... 133
Hypothesized Paths in the Campaign Process Satisfaction

Model for Those Dissatisfied with the Process ..................... 135
Hypothesized Paths in the Campaign Process Satisfaction

Those Satisfied with the Process . . ............................... 137
Complete Election Outcome Satisfaction Model ..................... 142
Complete Campaign Process Satisfaction Model .................... 144

Election Outcome Satisfaction Model -- Distribution of
Standardized Residuals .......................................... 173

xiii

LIST OF FIGURES (continued)

18

19

20

21

Results: Election Outcome Satisfaction Model --
Clinton Voters ............................................

Results: Election Outcome Satisfaction Model «
Bush Voters ..............................................

Campaign Process Satisfaction Model « Distribution
of Standardized Residuals ...................................

Results: Campaign Process Satisfaction Model «
Clinton Voters .............................................

Results: Campaign Process Satisfaction Model «
Bush Voters ...............................................

Overall Satisfaction with Process (CPSM) .....................

xiv

179

180

193

199

211

Chapter I
Introduction, Problem Statement, and
Expected Contributions

1. Introduction and Overview

Since marketing scholars first began to investigate consumer
satisfaction, the literature has been dominated by two conceptualizations
regarding the satisfaction formation process. These widely accepted but
rarely discussed conceptualizations are that:

1) satisfaction formation occurs post-purchase relative to a single
relevant product or service target (the "choice") and is unrelated to
alternatives considered but not chosen (the "nonchoice");

2) there is a single underlying satisfaction formation process that
operates across individuals and usage contexts.

This dissertation questions these two conceptualizations by addressing four
general research themes: (1) multiple targets (both "choices" and
"nonchoices"); (2) interrelated processes of satisfaction formation relative to
the multiple targets; (3) differential processing for satisfied versus dissatisfied
consumers; (4) expectations versus desires as antecedents of satisfaction
formation. The first three of these themes address directly the widely accepted
conceptualizations enumerated above.

In this introductory chapter, the logic behind the four research themes
of the dissertation is introduced in relation to the field of consumer behavior
as a whole, and then is illustrated using examples based on common consumer
experiences. Next, the four themes are briefly presented along with the

theoretical contributions each is expected to make. The context within which

the four research themes will be analyzed is then presented. Managerial
implications of the dissertation are presented next. Finally, the structure of

this dissertation is summarized.

I- l . Logic of the thesis: Relation to consumer behavior

Marketing scholars have long recognized the centrality of the concept
of satisfaction to their discipline. Early in the twentieth century, it was
recognized that the drive to satisfy desires coupled with the reality of limited
resources resulted in the need to choose from among competing alternatives.
For example, one early marketing text states: "No one can satisfy all of his
wants. People are never entirely satisfied Because the human being has
many wants and only a part of them can be satisfied, choice must be made
The consumer undoubtedly finds the greatest satisfaction when his choices
accord most fully with all of the factors enumerated" (Nystrom, 1929, pages 51,
72). In other words, consumer satisfaction and consumer decision-making
were recognized as being interdependent.

As the discipline evolved, marketing scholars used taxonomical
distinctions to help analyze the structure and function of marketing processes.
Alderson (1957) bisected consumer behavior into buying processes and
consuming processes en route to developing his functionalist theoretical
perspective. The distinction between buying and consuming as discussed by
Alderson continues to be useful when looking at the field of consumer
behavior as a whole. "Buying" refers to the information-processing aspects of
consumer decisions and consumer choice, while "consuming" refers to often-
neglected but "....equally important experiential aspects of consumption"

(Holbrook and Hirshman 1982, page 139). Westbrook and Oliver (1991) go so

far as to consider this distinction to be an "....emerging dialectic between
decision processes and experiential perspectives of the study of consumer
behavior" (p. 84).

Study of consumer satisfaction is one topic which resides primarily in
the post-decision, consuming realm of consumer behavior. Traditional models
of satisfaction formation typically regard consumer choice as a given and
specify satisfaction to be a function of antecedents such as expectation, desire,
attitude, perceived performance, and disconfirmation relative to the choice
already made. Thus, most satisfaction models have been structured and tested
in a manner consistent with Alderson's (195 7) distinctions, that is, with
consumption as totally separate from the "buying" process.

However, should the fact that it is possible to distinguish between the
purchase decision and consumption imply that it is desirable to study each as if
they are unrelated? Consider the implications of the "emerging dialectic" as
illustrated in Figure 1. The consumer considers purchasing one of three
alternative products or services. To explain how the choice is made, numerous
models have been proposed, tested and supported. Often such choice processes
include intensive information manipulation on an attribute-by-attribute basis
comparing expected characteristics or performance levels. Such comparisons
are made with the consumer's internal standards and/ or with the
characteristics and performance levels of the other alternatives. The end
result of this "buying" process is a selected alternative, in this case,
Alternative B.

Consumer satisfaction, as part of the "consuming process," is universally
modeled as a response or a judgment of the consumption experience related

only to Alternative B. Expectations, desires, attitudes, perceived performance,

 

 

 

 

 

 

 

Pre-Decision Post-Decision
Afi
Sat' fact' 0v all
3% wit: ch02: * Satisfaction
Cfi
Traditional
Choice Models: Satisfaction Models:

Alternatives A, B, and C are Only the antecedent constructs in relation
considered and comparatively to the choice (here, alternative 8) are
evaluated. Expectations and considered relevant.

desires are formed relative to

all three alternatives.

Figure l
Pre-Decision Versus Post—Decision Models

and disconfirmation related to Alternative B are considered sufficient to
understand satisfaction with the consumption experience. In other words,
satisfaction is modeled as a single process related to a single target. When one
considers the inten_sive choice processing which involved Alternatives A and
C in addition to B, it seems unreasonable to assume that those alternatives not
selected become totally irrelevant to post-choice sgt_i§_f_2_1ction formation
processes. Expectations, desires and attitudes regarding Alternative B evolved
in relation to expectations, desires and attitudes regarding the two other
alternatives. Likewise, in some usage contexts, the perceived performance of
Alternatives A and C may influence the consumption experience of the
consumer who selected Alternative B, especially if the consumer is dissatisfied
with B.

Some researchers (e.g., Cadotte, Woodruff and Jenkins 1987; Spreng and
Olshavsky 1993) have used nonchoice alternatives as standards of comparison
in satisfaction formation processes; however, the "comparison" is thought to
occur at a distinct, single point (e.g., disconfirmation) in a total process which
s_tfll focuses only on the choice alternative.

In summary, it is unrealistic to assume that intensively "processed" but
unselected alternatives ("nonchoices") have no impact on satisfaction
formation regarding the selected alternative (the "choice"). Given that many
consumer decisions of interest to marketers occur in a competitive context «
one in which alternatives exist, information is abundant, processing is
encouraged, and risk is involved « it becomes clear that the satisfaction
literature has not developed a satisfaction model which explicitly considers

competition.

[-2. Logic of the thesis: Theoretical support

While marketing scholars have neglected the relationship between
consumer decision making and satisfaction formation, theoretical support for
such a relationship is apparent from two other sources. In A Theory of
Cognitive Dissonance, Festinger (1957) explicitly recognizes that dissonance
can be created by the simple act of choosing one alternative while rejecting
others. He states: "....There will be some cognitive elements corresponding to
the positive aspects of the unchosen alternative and some elements
corresponding to the negative aspects of the chosen alternative which will be
dissonant with the cognition of having chosen one particular alternative..."
(p. 36). Thus, Festinger provides a theoretical link for hypothesizing that
nonchoice alternatives may continue to have an impact on consumers' post-
choice experiences. Although Festinger's analysis is limited to the period
following the choice and prior to actual consumption (p. 43), there is no
logical reason for concluding that dissonance can not extend into post-
consumption processes.

Theoretical support for the major premise of this dissertation may also
be found in regret theory from the field of economics. Loomes and Sugden
(1982) offered regret theory as a parsimonious alternative explanation of some
types of observed "non-rational" consumer choice behavior. Numerous
scholars (e.g., Kahneman and Tversky 1979) have demonstrated that
consumers systematically violate commonly accepted axioms of choice under
uncertainty. Rather than construct elaborate explanations of such behavior,
Loomes and Sugden (1982) question the axioms themselves. Their theory is
based on the notion that consumer choice is not "....independent of the nature

and combination of actions simultaneously rejected..." (p. 82). In their theory,

the anticipation of regret (and/ or "rejoicing") is combined with factors
derived from conventional utility theory. The entire constellation of
alternatives in the choice set is evaluated in terms of expectations following
the decision. For example, using the situation depicted in Figure 1, traditional
choice theory would state "choosing alternative B is preferred to choosing
alternatives A or C." Regret theory would state that "choosing alternative B
and simultaneously rejecting alternatives A and C is preferred to 1) choosing
alternative A and simultaneously rejecting alternatives B and C or 2) choosing
alternative C and simultaneously rejecting alternatives A and B." The
rationale for the latter framework is that consumers anticipate (expect) regret
from choices not made and factor that anticipation into their decision. For our
purposes, the important point is the implication that consumers develop
expectations regarding their future responses to nonchoice alternatives.
Thus, it is logical to hypothesize that nonchoice alternatives may remm

salient in both buying and consuming processes.

I-3 . Logic of the thesis: Example

In this dissertation, the relevance of Alternatives A and C, the
unselected alternatives, will be explored. Specifically, it will be proposed that
multiple "targets" and parallel satisfaction processing paths for these targets
are relevant to satisfaction formation (see Figure 2). The logic of multiple
targets and parallel processing of satisfaction formation can be illustrated
with the following example.

Consider a husband and wife who buy new cars at approximately the
same time. She buys a Geo Prizm and he buys a Toyota Tercel. The wife was

confronted with an attempted "bait and switch" maneuver and dealt with a

Pre-Decision Post-Decision

 

 

Satisfaction
Afi + w/ nonchoice
B—->fl->

ll

 

 

 

Satisfaction Overall
w/ choice Satisfaction

 

 

 

 

   

 

 

 

 

é Satisfaction
C + w/ nonchoic
Proposed
Choice Models Satisfaction Model

Constructs relevant to both choice ("B") and
nonchoice ("A" and "C") alternatives affect
choice satisfaction processes and overall
satisfaction.

Figure 2
Pre-Decision Versus Post-Decision -- Proposed
Model

salesman who tried to add on several new costs at the closing. Still, the wife
remained firm in her position and ultimately obtained the car she wanted at
the expected price. Her husband, on the other hand, experienced no problems
or surprises during the entire negotiation and closing process, and in fact was
offered several minor services (free of charge) which had not been expected.

During the first year of ownership, the wife's car needed minor repairs
on two occasions to correct a manufacturing defect. Both events were covered
by warranty. The husband's car did not need any repairs. At the end of the
second year, the wife's car began to exhibit minor rust spots; the husband's car
did not. Though the wife is, in general, satisfied with her car, that satisfaction
is diminished because of her "experiences" with the husband's car. The
husband's satisfaction with his car is intensified because of his experience
with his wife's car. In other words, satisfaction with the choice is partly the
result of expectations, desires, performance, and disconfirmation associated
with the nonchoice. Note that this experience is n_o_t_ simply the result of
having another substitute comparison standard as some have suggested.
Rather it is the result of parallel processing along the entire satisfaction
formation process. The husband's overall satisfaction with his transportation
experience is influenced not only by his satisfaction with the Tercel, but also
by his satisfaction with the Prizm. His satisfaction with the Prizm is
determined in part by his expectations, desires, performance perceptions, and
disconfirmation related to the Prism as well as by the these same constructs
related to the Tercel. (This illustrates the first and second themes of the
dissertation«see Section I-4.)

Because the husband is satisfied with his car, he does not think much

about his original expectations, desires, or attitudes, nor does he think about

10

whether these pre-purchase expectations, desires, and attitudes have been
supported in his experience with the car. Because he is satisfied, he can move
about his daily routine without giving the car much thought. On the other
hand, his wife is mildly dissatisfied with her car. This dissatisfaction "gnaws"
at her. It bothers her that although she spent $800 more for her car, she is
less satisfied than her husband is with his car. She recalls her original
expectations, desires and attitudes: "What did I expect? Were my expectations
unreasonably high or is the performance really worse than it should be? All I
really wanted was a trouble-free car; was that unreasonable? I really believed
that American car companies could once again compete with Japanese
companies, but maybe my husband was right ....". She replays her decision
process in her mind, partly because she doesn't want to repeat her "mistake"
and partly because her dissatisfaction iust bothers her. (This illustrates the
third theme of the dissertation.)

Upon reflection, the wife realizes that she had, to a certain extent,
talked herself into believing that American cars were once again of high
quality. She wanted it to be true, but in the back of her mind she expected to
have a few more problems with her car than her husband had with his. When
she actually experienced the problems, however, she was dissatisfied because
she had desired a trouble-free car, though she had expected less. (This

illustrates the fourth theme of the dissertation.)

[-4. Four research themes and expected contribution
Because this thesis will investigate issues pertaining to the core
structure of satisfaction formation processes, it is expected to make theoretical

contributions to the satisfaction literature. Each of the four themes of this

ll

dissertation will contribute directly to the conceptual understanding of
satisfaction formation processing. In some cases, significant new structural
enhancements are proposed. In other cases, discussion will address areas in
which conflicting results have been obtained. Expected theoretical

contributions are outlined next.

(1) The potential relevance of multiple targets of satisfaction formation
(i.e., choice and nonchoice) will be investigated. Current models of
satisfaction formation processes universally focus on a single target, the
central object or event about which satisfaction is being measured. It will be
demonstrated that, at least in some situations, measuring satisfaction with
multiple targets, individually and collectively, is more relevant than
measuring satisfaction with a single target. Situations in which multiple
targets are likely to be relevant are those in which consumers select a product
or service from among a set of alternative possible choices, and have some
exposure (direct or vicarious) to the performance of nonchoice alternatives.
Since a large proportion of consumer decisions occur in such situations,

conceptual implications of this thesis may be far reaching.

(2) It will be demonstrated that the processes resulting in satisfaction/
dissatisfaction with multiple targets are parallel yet interrelated. The major
implication of this finding is that the complexity of satisfaction formation
processing is not adequately captured in current models of satisfaction
formation. Not only should satisfaction with multiple targets be considered,
but the satisfaction formation process of each target also needs to be included

in satisfaction models, since the processes themselves are interrelated.

12

(3) It will be tested whether dissatisfied consumers process differently from
satisfied consumers. Current satisfaction formation models do not differentiate
between the actual processing of satisfied versus dissatisfied consumers. The
demonstration of differential processing implies the need for using models of

greater complexity, and/ or for developing a contingency processing model.

(4) It will be tested whether the t_ot_al effects of desires are greater than the
total effects of expectations in determining satisfaction with an important,
mghly involvingevent. One could argue that desires is a future-oriented
affect-based construct, just as expectation is a future-oriented cognition-based
construct. If supported, the implication of this theme is that the recent trend
in satisfaction literature towards an increased focus on affect should be

continued or expanded.

I -5 . Context of the empirical test

It seems logical to expect there to be a continuum of usage contexts based
on the degree to which non-selected alternatives (nonchoices) are relevant to
satisfaction formation. At one extreme would be the context implied by the
traditional satisfaction model (i.e., single target and single process). At the
other extreme would be contexts in which all alternatives considered have
significant impacts on satisfaction formation. To explore the proposition that
nonchoices can have a significant impact on satisfaction formation, it makes
sense to select a context in which the probability of uncovering such
relationships (if they exist) is high. If it can be demonstrated that nonchoices

can influence consumption experience satisfaction formation in one context,

13

then the range of contexts within which such relationships occur can be
explored. The first step, however, is to demonstrate that it ga_n occur.

The context for analyzing the four research themes will be the 1992 US.
presidential election. An election entails many of the same characteristics as
many more traditional consumer choice settings: an abundance of
information, choice among alternatives, relevance of traditional antecedent
constructs (e.g., expectations, desires, attitudes, etc.), persuasion attempts that
can create expectations, the probability that individuals will ultimately be
satisfied or dissatisfied. In addition, an election provides the potential to
obtain additional benefits rarely evident in more traditional consumer
behavior settings: wide variance in measures of satisfaction, wide variance in
measures of desires, and clear, objective, widely perceived measures of
performance (i.e., the outcome of the election).

Use of the 1992 presidential election as an arena for testing and
developing satisfaction theory is appropriate for at least three additional
reasons:

1) Marketing of political candidates is a huge industry in its own right.
Tens of millions of dollars are spent on campaigns during presidential election
years, and nearly as much is spent on campaigns, research, and political
strategizing during other years. The high cost of running for office is the
major force behind proposed campaign finance reform legislation. Marketing
research firms using established market research techniques play a
prominent role in identifying target markets and effective communications
program development. All the tools of advertising and public relations are
utilized. The "product" (i.e., candidate) is created and modified to most closely

match the wants and desires of the "consumers" (i.e., voters). Given the clear

l4

role of marketing in elections, it is surprising that political campaigning is a
topic seldom discussed by the marketing discipline.

2) Several topics clearly within the domain of marketing possess
characteristics which parallel those of political marketing. Any situation in
which few competitors compete in a zero-sum situation has similarities to
election marketing. Professional sports marketing, for example, is one
marketing application which may have similar satisfaction formation
processes. Two teams compete head-to-head and one team wins. Fans of each
team have expectations and desires before the contest, and varying degrees of
disconfirmation and satisfaction following the contest. The outcome itself is
generally unambiguous, and people can distinguish between the process and
the outcome. Other possible parallels include governmental and organi-
zational buying in which a limited number of bidders compete in a zero - sum
situation.

3) Findings from this investigation may be relevant to the broad spectrum
of product and service contexts because what is being investigated is the
structural relationship among the antecedent constructs of satisfaction. As
such, the investigation may have implications for satisfaction theory in
general. In other words, because the topic of interest is the relationship
among constructs, the findings may be relevant to the relationships among

satisfaction constructs in general.

I-6. Managerial implications
Top managers understand the importance of satisfaction formation
processes in determining corporate success. For example, Jack Welch,

successful CEO of General Electric, states: "Too often we measure everything

15

and understand nothing. The three most important things you need to
measure in a business are customer satisfaction, employee satisfaction and
cash flow. If you're growing customer satisfaction, global market share is
sure to grow too" (Tichy and Sherman 1993, p. 88).

Conversely, managers recognize that dissatisfaction has serious
repercussions. Numerous consequences of consumer dissatisfaction have been
documented including loss of repeat business, low employee moral, negative
"halo" effect on other aspects of the business, and negative word-of-mouth
communications (Albrecht and Zemke 1985). The cost of adding a new
customer is significantly higher than retaining an existing customer and
even small decreases in customer defection rates can have a major impact on
company profits (Reichheld and Sasser 1990). Given the increasing openness
of global markets, foreign firms which have specialized in satisfying
customers will enjoy competitive advantages over U.S. firms which have not.

Because this dissertation questions the core structure of satisfaction
formation processes, results are expected to have numerous implications for
managerial application.

1) If the parallel satisfaction processing framework is supported,
managers may have to alter their strategies for increasing/maintaining
consumer satisfaction. They will have to consider that consumer satisfaction
with their product or service is, in part, determined by consumer satisfaction
with competitors' products or services. Also, they will consider the
implications of the fact that consumer satisfaction with competitors' goods and
services is at least partially determined by customer satisfaction with their

goods and services. This represents the simple, but critical recognition that

satisfaction formation processes. too, occur within a compatitive environment.

16

Producing high quality products at a fair price is not the only determinant of
satisfaction even among one's own customers. Of course, much more research
will be needed to establish that such parallel processing occurs in the context
of consumer and durable goods. The exact nature of the "cross-over" effects in
various contexts will also need to be specified.

2) Support for the parallel processing framework would imply managers
should immediately begin to alter the way they monitor customer satisfaction.
Most firms which monitor satisfaction currently collect measurements only
with regard to their own products; those firms which do collect information
related to competitors' products collect only summary information instead of
information relevant to the satisfaction process. Satisfaction information with
regard to competitors will help in determining the degree to which parallel
processing occurs in a particular industry, and if it does occur, will provide
guidelines for managing satisfaction formation.

3) If it is determined that satisfied versus dissatisfied consumers arrive at
their ultimate level of satisfaction/dissatisfaction differently, efforts to
promote or discourage specific kjn_ds of processing may be effective in
managing satisfaction formation.

4) By clarifying the relative importance of expectations versus desires in
the satisfaction formation process, firms will have a clearer idea of how to
promote and advertise their products and services. Again, specific findings
from studiesconducted in specific industries will be necessary to determine
ideal applications. However, this dissertation is a necessary first step.

5) If this research effort is successful, at least part of its success will be
due to the wide variance of the satisfaction measures. Thus, the dissertation

will demonstrate the practical implications of developing satisfaction

17

measures which exhibit variance. Currently, many satisfaction monitoring
efforts in industry are ineffective because satisfaction levels are consistently
high. This causes two problems. First, it is impossible to measure any
improvements which do occur because satisfaction levels cannot increase.
Thus, the usefulness of such measures as a factor in determining manager
bonuses, for example, becomes insignificant. Some firms have attempted to
address this by measuring proxies for dissatisfaction such as customer
complaints. However, such a practice is dangerous because then the manager
has an incentive to minimize or suppress complaints even though complaints
are a valuable source of information. Second, relationships between potential
antecedents of satisfaction and dissatisfaction become extremely difficult to
identify or monitor. Thus, managers are unable to establish priorities among
the factors responsible for satisfaction. Obviously, this reduces the probability
that resources will be optimally distributed. Therefore, simply increasing
variance in the satisfaction measure can lead to improvements in the
usefulness of satisfaction measurement.

6) Clarification of the relationship between attitudes and desires, and their
relative importance in the satisfaction formation process, will help managers
make decisions regarding allocation of resources among efforts designed to
alter desires (or perceptions of desires congruence) versus those designed to

improve attitudes.

[-7. Structure of the dissertation
Following this introductory chapter, the dissertation is organized as
follows. Chapter 11 contains a review of the traditional satisfaction literature

with a focus on the constructs used and the relationships among constructs.

18

Chapter 111 provides the formal development of the research and testable
hypotheses associated with the four general research themes of this
dissertation. Chapter IV contains a description of the design, data collection
and analytical methods used to test the research hypotheses. Chapter V
presents results of the empirical investigation, and Chapter VI contains the

discussion of results, limitations, and implications for future research.

Chapter II
Satisfaction Literature in Marketing

II. Introduction

The purpose of this chapter is to review the relevant satisfaction
literature (within the limits described below) in order to develop a thorough
understanding of the current state of satisfaction formation processing models
in marketing. The review will be broken into four broad topic areas. In the
first, the conceptualization and Operationalization of commonly used
constructs will be analyzed. The second will focus on empirical findings
regarding key relationships among the antecedent constructs. The third topic
area will focus on the relationship between antecedent constructs and
satisfaction. The final topic area will examine other relevant findings not
explicitly involving common constructs.

Throughout this chapter, distinctions will be drawn between how a
construct is conceptualized and how it is measured, as the two often appear to
be different. These differences may be responsible for at least some of the

confusion and inconsistencies found in the satisfaction literature.

II-1.1 Scope of the literature review

This review is necessarily limited both with regard to concepts
addressed and to sources reviewed. The purpose of this section is to make the
limits explicit. Thus, the following are discussed in this section: (1) sources
reviewed; (2) the political science literature; and (3) goods versus services

marketing.

19

20

Sources reviewed
With regard to sources utilized, this review of theoretical and empirical

contributions to the satisfaction literature is delimited as follows. First the

 

review will focp§ on contributions from ma_rl_<eting publications only.
Specifically omitted from the formal review are works focusing on job
satisfaction, life satisfaction, and patient/healthcare satisfaction. Occasional
reference to these disciplines will be made when appropriate, but the
extensive nature of each of these literatures precludes comprehensive
coverage in this review.

Second, m review concentrates on con§umer rather thg
organizational satisfaction a_s its central construct. The concept of
organizational satisfaction is fundamentally different from consumer
satisfaction in that the former somehow encompasses a collective judgment or
evaluation of a product or service. No empirical study has demonstrated that
the two types of satisfaction are similar enough to be considered a single
construct.

Third, the review focuses on theoretical and empirical works appearing
in a set of academic publications. Peterson and Wilson (1992) state that over
12,000 books and articles on satisfaction have appeared in the past 25 years. A
complete review of the "satisfaction literature" is thus impractical if not
impossible. Therefore, the review focuses primarily on works appearing in

the following publications: loumal of Marketing, loumal of Marketing
Research loumal of Consumer Research, and loumal of the Academy of

 

Marketing Science. These journals represent the major outlets for quality
theoretical and empirical analyses of consumer satisfaction. In addition,

works presented at the American Marketing Association Summer and Winter

21

Educators' Conferences, Association of Consumer Research conference and the
conference of Consumer Satisfaction/ Dissatisfaction and Complaining
Behavior will be reviewed as these conferences have also yielded noteworthy
contributions to the field. Finally, significant contributions from other
sources which have been referenced in the works listed above will also be

included.

Political science literature

Although the setting used here for investigating satisfaction formation
processes is the 1992 presidential election process and outcome, a review of the
political science literature will not be extensively discussed here. There are
two reasons for this:
1) This dissertation addresses satisfaction theory from a marketing
perspective. The constructs of interest, the measurement of those constructs,
and the hypothesized relationships among constructs all come from marketing
literature. The particular setting of a presidential election was selected to
overcome some clearly documented problems with the conduct of satisfaction
research in traditional marketing contexts. The analysis presented here is
neither designed nor intended to address theoretical constructs or
relationships of interest to political scientists. Therefore, a comprehensive
review of that literature is unwarranted.
2) A preliminary review of the political science literature revealed that
much effort is devoted to predicting the outcome of elections and to
monitoring public opinion and satisfaction with the performance of elected
officials. Very little effort is directed to measuring, or explaining, voter

satisfaction with the outcome of an election immediately after its conclusion.

22

There appears to be an implicit assumption that if a voter's candidate won, the
voter will be satisfied, and vice versa. The formation (e.g., antecedents) of
satisfaction are not of immediate concern, and there is little emphasis on
process as defined in the marketing literature. Therefore reference will only

be made, when appropriate, to specific works from political science.

Goods versus services marketing

The marketing discipline has long distinguished between the marketing
of goods versus services. This distinction remains intact in the traditional
satisfaction literature. For example, Parasuraman, Zeithaml, and Berry (1988)
draw a clear distinction between the two at the level of construct definitions,
while Yi (1990) in his comprehensive review claims to focus exclusively on
consumer satisfaction with products. (However, it should be pointed out that
Yi did include references from the service satisfaction literature on a selective
basis, e.g., his references to Cadotte, Woodruff, and Jenkins 1987).

Justification for separate treatments appears to derive from the classical
distinction between goods and services marketing, such as the simultaneous
production and consumption for services versus sequential production and
consumption of products, and the intangible nature of services versus the
tangible nature of products.

While the distinction between goods and services may be pseful for
some pupposes, the distinction appears to be artificial in the satisfaction
Megature.and thus is not presumed for this review. There are four reasons
for this:

1) There appears to be little difference in the antecedents and

consequences of services versus product satisfaction. Though this argument

23

can only be convincingly made following the review, a brief analysis of the
antecedents and consequences of satisfaction in product versus service
situations reveals more similarities than differences. Antecedents common to
both conceptualizations include expectations, desires, attitudes, perceived
performance, and disconfirmation. Consequences common to both include
word of mouth, complaining behavior, repeat purchase intentions, repeat
purchase behavior, and attitudes. Thus, at least on the surface, satisfaction
seems to be related to similar constructs in both product and service contexts.
2) The satisfaction concept itself seems to be similar if not identical in both
the product and service contexts. In general English usage, the term
"satisfaction" is commonly used in both contexts. Likewise, dictionary
definitions of "satisfaction" do not distinguish between satisfaction with
products and satisfaction with services. Also, as will become clear in the
review, satisfaction is measured similarly in both service and product contexts.
Both employ overall and trait-specific measures, and both employ a range of
scales including semantic differentials anchored by "satisfied - dissatisfied" or
variations employing these terms. Thus, there is no a priori reason based on
the satisfaction construct itself for distinguishing between product and
service contexts.

3) Even those scholars who distinguish between products and services are
quick to acknowledge that such distinctions are at least somewhat artificial.
For example, businesses may be placed along a continuum based on the degree
to which each provides goods versus services. Kotler (1980) states that all
goods producers provide at least some services; many service providers also
include some products or product-like attributes in their service offerings.

Given that those who draw distinctions between goods and services admit

24

ambiguity in those distinctions, it seems more prudent to utilize the
distinctions only when there is clear benefit to doing so rather than to assume
such distinctions are always relevant. Again, there is no clear a priori reason
for assuming the distinction is relevant in analyzing satisfaction formation
processes.

4) The trend in both academic and managerial sectors has been to expand
the "target" of satisfaction. For example, in a product context, the relevant
target of satisfaction seems to be expanding from satisfaction with a specific
product's performance to satisfaction with a broader "consumption
experience" (Yi 1990). Given that there is some mix of service and product in
most marketing exchanges, the practical distinction between the two,
especially in the mind of the consumer, is even more likely to blur as the

target of satisfaction measurement is expanding.

II-l.2 "Traditional" satisfaction model framework

The following discussion of constructs and relationships is structured
around what is referred to throughout this dissertation as the "traditional"
model within the "traditional" satisfaction literature. As used here, the term
traditional satisfaction literature refers to the family of satisfaction formation
models which utilize a single target of satisfaction. It is understood that this
family of models includes a wide variety of hypothesized relationships.

In addition to focusing on a single target, traditional satisfaction models
are built around a core set of constructs and theories (Yi 1990). In the
marketing literature, satisfaction processes have been studied from a variety

of theoretical perspectives. Although the disconfirmation of expectations

25

perspective clearly dominates the literature (Yi 1990), the following
perspectives have also been proposed, and, to a greater or lesser extent, have
been shown to be useful:
0 Value-percept disparity--The difference between perceived performance
and a person's values-related desired level of performance is conceptualized as
the major antecedent of satisfaction (Westbrook and Reilly 1983).
0 Equity theory--The perceived relationship between input/output
(cost/ benefit) ratios of the consumer versus the salesperson is the primary
antecedent of satisfaction (Oliver and DeSarbo 1989).
0 Extraordinary experience-Satisfaction with extraordinary experiences is
determined by a complex set of cultural and personal factors which unfold
over time and cannot be adequately captured in simplistic measures of
expectations and disconfirmation (Arnould and Price 1993).
0 Desires congruency--An extension of the value-percept disparity
perspective, desires congruency is the difference between a desired and
perceived level of performance, and is the primary antecedent of satisfaction
(Spreng 1992; Spreng and Olshavsky 1993).
Although each of these perspectives contributes to our understanding of
satisfaction formation processes, the disconfirmation of expectations model
will provide the primary orientation for this dissertation given its dominance
of the marketing literature. However, these other perspectives will be utilized
frequently throughout this dissertation when relevant (e.g., to demonstrate
the variety of ways in which satisfaction is operationalized).

Within the disconfirmation of expectations perspective, the core
antecedent constructs in the satisfaction formation process are illustrated in

Figure 3. In the traditional model, antecedent constructs fall into two

26

 

 

 

 

 

 

 

Major Major
Pre Performance . Post Performance
Constructs Constructs
l I I I
Expectation
Perceived . f' .
performance Discon Irmation
Attitude
I J
Antecedent Constructs

In the traditional model, only constructs related to the choice alternative are
considered. Thus, constructs illustrated here are all "post choice" constructs.
While the temporal order of pre-performance constructs is unclear, the
generally accepted order of post-performance constructs is as illustrated.

Figure 3
Classification of Satisfaction Formation Constructs
from the Traditional Model

27

categories, those which precede performance and those which follow. Note
that expectations, desires, and attitudes, are the core pre—performance, or pre-
experience, constructs. Perceived performance and disconfirmation are the
core antecedent constructs which m product/ service performance. While
there is little agreement on the temporal order of pre-performance constructs,
post-performance constructs are generally thought to occur in the following
order: perceived performance -> disconfirmation -> satisfaction.

The review which follows begins with a conceptual and empirical
analysis of "satisfaction" as a construct (Section II-2). Next, the pre-
performance and post-performance constructs are examined (Section II-3);
then, relationships among the constructs are reviewed (Section II-4).
Additional constructs not generally considered part of the traditional
satisfaction model are reviewed (Section II-S). Finally, the major points of the

chapter are summarized (Section II-6).

II-2. Satisfaction and its measurement

Satisfaction of consumer needs and wants is arguably the central theme
of marketing as a discipline. Therefore, it is surprising that serious attempts
to define satisfaction both theoretically and empirically did not begin until the
late 19608. Two related forces may be credited with the rise of interest in
satisfaction as a topic worthy of investigation. First, the rise of consumer
behavior as an identifiable subdiscipline within marketing brought the
perspectives and methods of psychology and social psychology to marketing.
The focus on understanding and predicting the perceptions, attitudes and
behaviors of individual consumers resulted in the identification of satisfaction

as a key construct (Hunt 1977). Second, one of the periodic revivals of

28

consumerism as a social and political force occurred during this period.
Consumer satisfaction became a salient public policy issue due to widespread
perceptions that companies were not responsive to consumer demands and
that the federal government was overly supportive of business interests at the
expense of consumers. Perhaps not surprising is the emphasis on consumer
dissatisfaction and complaining behavior which resulted from this

reincarnation Of consumerism.

Cpgnitive versus affective content

Early conceptualizations of satisfaction in the marketing literature
described satisfaction in cognitive terms (Engel, Kollet, and Blackwell 1973;
Howard and Sheth 1969). For example, Howard and Sheth (1969) define
satisfaction as "....the buyer's cognitive state of being adéquately or
inadequately rewarded for the sacrifices he has undergone" (p. 145).

The traditional satisfaction formation model treats the process as being
primarily cognitive, while the satisfaction construct itself is generally
perceived to have definite affective overtones. Only recently have affect-
based processes been proposed to help explain satisfaction. This may be due in
part to the recent theoretical advances made in understanding affect (e.g.,
Izard 1984). Zajonc (1980) indicates that though the importance of affect has
long been recognized, empirical and theoretical emphasis on cognitive
psychology has dominated the discipline of psychology. Given the close
relationship of psychology and marketing, especially in the consumer
behavior school (Sheth, Gardner, and Garrett 1988), it is not surprising that

early satisfaction researchers sought cognitive antecedents of satisfaction.

29

Within marketing, the role of affect in satisfaction is first discussed by
Hunt (1977) who states that affective responses to a product experience follow
the formation of satisfaction via cognitive processes. Westbrook (1980) raises
the question of whether affect enters the satisfaction process prior to or
concurrent with satisfaction. He states "....affective influences do not, of
course, deny the role of cognitive processes such as expectancy confirmation
or disconfirmation, but rather combine with them in the determination of
consumer satisfaction" (p. 50). Customer satisfaction, then, is viewed as a
combination of cognitive and affective elements in an overall evaluation of a
product.

Others also believe that satisfaction is a general, overall feeling. For
example, Westbrook and Newman ( 1978) state "Satisfaction typically is
conceived as the extent to which consumers feel subjectively pleased with
their ownership and usage of products" (p. 456). Likewise, Cadotte, Woodruff
and Jenkins ( 1987) define and measure satisfaction as a summary, affective

C 0118th t.

Dimensionality

The dimensionality of satisfaction has been questioned (Swan and Combs
1976; Maddox 1981) and the issue remains unresolved (Yi 1990). The primary
issue is whether satisfaction is a unidimensional construct (e.g., with anchors
of "completely dissatisfied" and "completely satisfied") or two separate
constructs, one being "level of satisfaction" and the other being "level of
dissatisfaction." The two-factor conceptualization is based on Herzberg's Two-
Factor Theory (Herzberg, Mausner, and Snyderman 1959) and derives support

from two basic observations. First, consumers appear capable of being both

30

satisfied and dissatisfied at the same time and therefore the two constructs
must be at least somewhat independent. Second, consumers who are asked to
recall both satisfying and dissatisfying experiences recall different "levels" of
phenomena for describing why they are satisfied versus dissatisfied (Maddox
1981).

Only equivocal empirical evidence has been found to support the two-
construct conceptualization. Swan and Combs (1976) found weak preliminary
support. Their study was replicated by Maddox (1981) using larger samples of
customers drawn from multiple industries. Maddox found little support for the
two-factor conceptualization, and concludes that such a conceptualization also
yields implications which run contrary to common sense.

Until recently, support for the two factor conceptualization has been
derived almost exclusively from studies using a single methodology: the
"critical incidents" method described above. An alternative explanation of the
findings yielded by this method is possible. Briefly, it is possible that the

satisfaction construct is unidimensional, but, given that satisfied and

dissatisfied consumers may process differently, theLremember different
kinds of factors as being responsible for their ultimate level of sagsfamp.
Because the critical incidents method requires consumers to "retrace" their
process, it is not surprising that their responses are qualitatively different.
Recent evidence from Babin et a1 (1993) provide stronger empirical support
for a two-dimensional conceptualization using confirmatory factor analysis.
The evidence is not conclusive however, as (1) measurement scales selected for
use may have "pre-disposed" the two-dimensional findings, and (2) theoretical

justification for the two dimensions is weak.

31

Measurement

Table l and Table 2 list the empirical studies reviewed which measure
satisfaction as either an antecedent or consequent construct. Certain
characteristics of the satisfaction measurement are also indicated. The variety
of satisfaction measurements is discussed next, and the relationship between
how satisfaction has been measured and specific research findings will be
analyzed later (see section II-4.4).

Table 1 indicates whether the study cited uses some form of
"satisfaction/dissatisfaction" anchors for measuring satisfaction, and whether
the study uses an overall and/ or attribute-specific measurement of
satisfaction. Table 2 presents details of the satisfaction measurement
including number of items and scale points, actual question wording, and scale
anchors.

Table 1 and Table 2 reveal the following:

1) The single most striking observation regarding the actual
measurements of satisfaction is that no two studies measured satisfaction in an
identical manner. To be sure, there are clear similarities among specific
groups of studies, but there is no evidence that a standard satisfaction scale is
emerging in the literature. Researchers may choose not to use previously
developed scales because they believe they are examining satisfaction in a
unique context or because they are testing new satisfaction measures. No
matter what the reason, the consequence of using nonstandard measurement
scales is that comparisons across studies and the subsequent generalizations
must be approached with caution.

The fundamental reason for lack of consistency in the measurement of

satisfaction is the lack of agreement on its definition. This observation is not

Measures of Satisfaction

Anderson and Sullivan 1993
Babin, Dardin and Griffin 1993
Barbeau 1984

Bearden and Teel 1983

Bitner 1990

Bitner and Hubbert 1994
Bolfing and Woodruff 1988
Cadotte, Woodruff, and Jenkins 1 987
Cardozo 1965

Churchill and Suprenant 1982
Cronin and Taylor 1992

Drbge and Halstead 1991

Fisk and Young 1985

Garland and Westbrook 1989
Halstud 1993

LaBarberra and Mazursky 1983
Mano and Oliver 1993

Oliver 1993

Oliver 1980

32
Table 1

"Sat/Dissat"
On Scale

NO
YES
NO
NO
NO
YES
NO
NO
NO
YES
YES

YES

NO
YES
YES

NO

NO

NO

By
Attribute

in Empirical Studies

Overall

<<<<<<

<<<

<<<<<<

Tabka

Oliver and DeSarbo 1988
Patterson 1993

Prakash and Lounsbury 1984
Richins and Bloch 1991

Singh 1991

Spreng and Olshavsky 1993
Swan and Martin 1981

Swan and Oliver 1991

Tse and Wilton 1988

33

1, confinued

”Sat/Dissat "
On Scale

YES
YES
YES
YES
YES
YES

NO

YES

By
Attribute

Overall

<<<<<<<<<

34

.8335 - 83......

5632955 - 8389:...
.0333: . poucoucou.
Lon... - 35.8. 35.8.. 38 - so... :53.

.838... 29.83. . 85a 2935...
.8535 29.25 - 02am 29.25..

.295» - 829.8.

3.2:: - 8238..

.3585 - 622..

SEES - 829.8.

.98.. 28.25 - =3, 38.25.
32

he... - .3.

0.00m

858..» 8.. 2.2a;

20:3 .35 c. 3.53 no .8050 3.98 3:
.633...an $5

3 88 2,2 so: 2:2... as? .26 8 8 a. as. . a.

negative
8.28 958% use 5.; 8:58 8; _.
.2388 3:8! as

:0 35596 8.28 So» Sons .00.. so» In 301..

.359 .993 05 Sena .03 :0» 2:2» 26..
58a 05 c. 33:38 383258.“. ofi .995.

.3233 05 8.8 so...
5:28.... .8. use; _ :26 8 3 g 8.. . a...

.25 on...» a no; ”.0330 338 05 on: 8 3.05 .50..

Sec 3.8.86

E3830 :83
3300: 50» ~09: 8.53 05 no =25 301..

8335:. a co...

Abomp 3085mm; Ea camp 32.0 cu a_wfimb (\z

nectar. uo: 05203

.8383

oopxn

O—

E3. .8
.3; 2.3

m mom— r3500; Us 9.50m

a ramp tongs: use .055

— camp 5055

e mom, .8... ac. coanm

o_. vamp aaofiam
m mam. acute as £38 635

p mmm— Exam was .5803

E3. ‘0
509.52

Cozumwmfiamm $0 mMEDmNOZ Dismam

N 03m...

35

(\z

8.00.0000... 0000.00.00. - .8008. 0000.00.00.

........08 0.0.80 .8. .. .5002... 005.00..
8...... 30.. .. 00000 80.9.00 .00.. _.

800.080... 00>. - .8880 00?.

800.008... 00?. - .8008... 00>.

800.0000... 030.0500. - .8008... 0000.00.00.

009000.98 020.09 00.3 000! 00 00:3
.3505. .. .03....
sabgmu b95058”: I 0028 zggmu

.03§00....= - 8000.00...

.0808 - 808.... 83500085 - 8.0.0...
.8053... - 80.8000. .880 - 800....
000000.00: - 000000.... .0300 . 800.0...

.80 - 800. .028. 0.00 . 30.0 053.

0.00m

.000. .026 00.0.6.0...2 :2.

 

m
0.000 0. 000800.000. 00.0.03 n
.0050000. 003 000....0 .0 00.000 5.2. 00.000.00.009.
80.0000 00.. 00.0.02. ...
0050000 000 8.0.03 0
.500 00800000
00 0000 000 003.00 0. E0300 02:00.. >2. n
"0500000 .3026
"050000.. .3026
"800:0 "0. 000000000 :3. .0”.
”£0.30 .0. 0003.000 23— .0”. N

020030.000 a 00.00290 300000.000 0000 00030.0( 00.000.

80.80. 00.. 00.0.0.5 0

020003 0.00. 000
.00.. 0.000

02.0.0000 .N 0.00»

mp

¢~\-

0—

ER: 00
0000.52

nom—

mma— 002.09.00.31

000. 0.0.000... 80 08.0003

000. 0.00.802. 0.... 80000

mwmp 950» us 0.0....

amp 300.0... Us on;

Nam. .0300 000 0.00.0

Nam. 00000083 000 3.0055

mwmp 3025

8:84 0... 0.080.... 00.080

36

.000 00000.00... 00800.00. - .00. 00000.0 000000.00. ..00E00.00 50.. 00.3 0.03

:0» 0000.000 30.. 0000.00? 3.00 00.00.08
.00000000000: 2.0? - .b000£0_000 .00?
60000088

.0000300. .. .03th. 000 0:000 .08 0.00.... 0 000 30.. 000.00 .002
.089000 300 .00.. 00.0: 000 00:50 00

.888... 20>. - 88.00000 8? 800.09... .0030 2.0 .0... 0.. 0.00; 88.000 :0...

88.000 230.058. - ...0 00 008.000 002.
L...0.000 000 00 0000.00.. 000. :0» 0.00.3 0.003 .000
.88.. 888. 230.008. .00.. 000.... 5.... 88.00000 0... 00.. 00.....0 2.0 .0
. .38. 88.000 ..0 00 002. 0.... 00.; 88.000 0.0 00.. 00.....0 2.0 =0 9.000.008.

.8800... 200.205.. - 88.0000... 200.200...

.0000 00......
.00.. 23.0000. - .00... 9.80.0000 8; 003 08... 0 00 .38..
.0 000.0 .00.. 000000000. :0» 0.002... S
.8808 230.058. - 88.00020 .09. 00.00.00.000 .0090 ..

0.0.000 0.00.0006 .8 0 6.9000 0000 .8 ~— 0
£00. 0.00.0000...
$2 0.... .20 0.... 000. 82.0 00 5.0.... $2

.....0.0 _ 00.... 0.0 . 00.0 2.8.. 00.. .00 ..

.....0.....0 0...... 000 0.0. 0......0 ..

.......0.0.000 2.. 000.... 8.. .8. ..

.000 00.2. 0 003....00600 .02..

20.5.3.0 .8. 0.00; . 0.000 .90 00 00 0. 8.. . 0..

00.000 0.9.... .......0.0.0o0 2.. 00.... 88.08:...

0.000 00.0026

O—

E.

0.

0.00. .00 00000. 00

.000 0.08

00050000 .N 0.00...

cw-~ _.

w

0000002

89 002.0 000 003m

52. 0.0.02 000 003m

mam. b.9205 0:0 205m

.00. 000.0

amp 0003 0:0 00.00.».

.00. 230.50.. 0.... 88......

mag. 000000000

mam. 002.0

0mm. 002.0

37

new (Hunt 1977). Recently, Yi (1990) summarized the problem as follows: "For
the field of C8 to develop further, a clear definition of CS is needed Before
more studies are done, an effort should be made to clarify the concept of CS" (p.
74). Without a clear conceptualization, standardized operationalizations will
not evolve.

2) A second striking observation from Table l and Table 2 is that numerous
studies remrt incompletely, or not at all, exactly how satisfaction was
measured. Some omit critical information regarding number of scale points
(e.g., Oliver 1980; Bearden and Teel 1983). Some do not report the number of
items (e.g., Cardozo 1965). Most, while giving specific information about the
scales used, do not provide the wording questions used to elicit the scalar
responses (e.g., Cadotte, Woodruff, and Jenkins 1987; Westbrook and Reilly
1983). Others provide only examples of specific questions asked (e.g., Bearden
and Teel 1983; Swan and Oliver 1991). Finally, some provide virtually no
information : for example, Fisk and Young (1985) indicate simply that
"....satisfaction with choice of airline was measured." Because standards for
satisfaction measurement do not exist, such omissions further cloud
understanding of specific study results and complicate the process of making
generalizations.

3) Given that no standard measurement exists, the task remains to
document and understand the variation in measurement which does exist. One

observation is that most studies utilize some overall measure of satisfaction

 

whether alone or in combination with attribute-specific measurements.
Attribute-specific measures are based on the notion that a consumer can
evaluate each product/ service trait individually, is satisfied or dissatisfied with

each and forms an overall satisfaction judgment in some additive fashion.

38

Reported reliabilities of such attribute-specific scales are high (e.g., Churchill
and Surprenant 1982) indicating at least some support for this method.
However, given that most studies use only overall measures of satisfaction, it
appears the consensus is that satisfaction is a summary evaluation, and
includes reactions to the gestalt of the product/ service experience. On the
other hand, the prominence of overall measures may simply be due to the
recognition of a practical problem: inability to specify all salient attributes
would tend to reduce the validity of attribute-specific measurements.

4) Nearly half the studies reviewed used single item measures of
satisfaction. Numerous studies reviewed here have utilized single item
operationalizations for satisfaction as a key variable: Anderson and Sullivan
(1993), Bitner (1990), Cronin and Taylor (1992), Garland and Westbrook (11989),
LaBarberra and Mazursky (1983), Prakash and Lounsbury (1984), Swan and
Martin (1981), Tse and Wilton (1988), and Westbrook (1980). Note that this list
includes multiple studies from the lournal of Marketing and the lournal of
Marketing Research. Note also that half of them have been published since
1987.

Yi (1990) criticizes single-item measures on a variety of conceptual and
methodological grounds, but his statement that recent studies have tended to
use multi-item measures ("....in contrast to earlier studies...." ) is not supported.
Studies conducted before 1988 were roughly 50% multi-item and 50% single
item measures; in contrast, nearly 70% of the articles published since 1988
included only single item measures of satisfaction. Although the superiority
of multiple item scales is well established, it is clear that use of single item

scales remains common.

39

At the opposite extreme, four studies utilized measures constructed of
ten or more items (Barbeau 1984; Churchill and Surprenant 1982; Oliver and
Bearden 1983; Cadotte, Woodruff, and Jenkins 1987). The first three of these
used so many items because their measures were, at least in part, summations
of attribute-specific items. All reported at least moderately high reliabilities.
5) Scale length, or number of points per item. also exhibits wide variation.
The minimum number of points appears to be four while the maximum is 11.
Approximately half the studies used seven point scales.

6) The wording of satisfaction scales and guestions likewise exhibit wide
variation. In fact, it is difficult to find two studies which measure satisfaction
in exactly the same way. Some studies, including many which have been
extremely influential in the literature, do not ever mention the word
"satisfaction" in either the question wording or for the scale itself. For
example, Cadotte, Woodruff and Jenkins (1987) use "feelings" as the dependent
variable in their oft-cited study comparing expectations and norms as
standards for disconfirmation. The 10 item "feeling" scale they employed did
not use "satisfaction" in any of the items, yet they refer to satisfaction as the
dependent variable in the text and title of the article. A sizable minority of
studies use scales which did not include the terms "satisfied" and "dissatisfied"
or any variation of those words. Four studies used some variation of the
"delighted - terrible" scale. Four studies also used Likert scales.

Question wording in some studies indicates that constructs other than
satisfaction per se were measured. For example, Cardozo (1965) implies that
satisfaction is simply the subjective evaluation of performance. Bearden and
Teel (1983) and Oliver (1980) use at least one item which appears to be similar

to behavioral intention: "If I had it to do over, I would feel differently about

40

using the business."

Question and scale wording also reflect diversity regarding the degree
to which satisfaction is thought to be primarily cognitive versus affective.
Some unabashedly focus on the affective nature of satisfaction. Oliver (1980)
states that "....all items were emotional in content...." (p. 463). Cadotte,
Woodruff, and Jenkins (1987) clearly favor an affect-based Operationalization
by utilizing scale anchors such as "warm glow/cold feeling," "uplifted/down,"
"elated/tense," and "pleasant/ unpleasant."

Others use both cognitive-based and affective-based items. For example,
Barbeau (1984) measures response to each attribute on an "extremely well" to
"extremely poorly" scale, and measures overall satisfaction on the affect-based
"delighted - terrible" scale. Churchill and Surprenant (1982) explicitly
acknowledge both aspects of satisfaction by measuring half their attribute-
specific items as beliefs and half as affects.

7) Marticles vagwith regard to the actual "target" of satisfaction
measurement. The three most common targets appear to be the product or
service itself (e.g., Barbeau 1984; Churchill and Surprenant 1982; Tse and
Wilton 1988), the consumption experience (e.g., Cadotte, Woodruff, and
Jenkins 1987), and the purchase decision (e.g., Bearden and Teel 1983; Oliver
1980). Yi (1990) enumerates a more detailed list of targets.

The distinctions between targets is significant. For example, one could
be satisfied with one's decision to take a vacation in Florida (e.g., given the
options available, access to information, etc.) but still be dissatisfied with the
vacation in Florida (e.g., colder than usual), but feel satisfied with the totality
of the experience (e.g., good to get away). Only one study explicitly compared

satisfaction with different levels of target (Bitner and Hubbert 1993).

 

41

8) Notice that notwithstanding the ongoing discussion about the
dimensionality of satisfaction, gtisfaction is consistently operaLtionalized as a
unidimensional construct. In only a few cases were satisfaction and
dissatisfaction measured separately (e.g., Westbrook and Reilly 1983), and in
these cases, the scales were simply combined into a single overall measure of
satisfaction. Renewed interest in the dimensionality of satisfaction is
apparent, but no consensus on the issue has been reached (Babin et al 1993).
Singh (1991) implies a two-dimensional conceptualization given his
measure of satisfaction which ranges from "not satisfied at all" to "completely
satisfied." Likewise, Richins and Bloch (1991) and Westbrook (1980) imply two
dimensional conceptualizations with their numerical scales which range from
0% to 100%, where 096 is labeled "not at all satisfied" and 100% is labeled
"completely satisfied." These scales indicate that the opposite of satisfaction is
absence of satisfaction rather than dissatisfaction. However, in two of these
studies (Singh 1991; Westbrook 1980), this is the only item used to measure
satisfaction. The degree to which such measures, as opposed to the more
traditional bipolar scales, alter the revealed relationships between satisfaction
and its antecedents is unclear. The important observation at this point is that

most researchers actually measure satisfaction as if it were unidimensional.

Inconsistent operationalizations of the satisfaction construct mirror the
inconsistent conceptualizations of satisfaction. Numerous calls have been
made for development of agreed upon operationalizations of satisfaction, but
little progress has been made on this problem. Until standard measures of
satisfaction are widely implemented and shown to have reliability and validity

across a range of contexts, the question "what exactly is being measured

42

here?" will remain. Hunt (1977) states that the study of customer satisfaction
is "...much more advanced in measurement and methodology than...in
conceptual development" (p. 16). While much progress has been made since
1977, the statement is still true.

In summary, satisfaction

0 is generally measured as a unidimensional construct,

0 often measured using a single item,

0 has a variety of "targets,"

0 most often uses seven point scales,

0 exhibits wide variation in scale wording, but usually includes the

word "satisfaction" or a variation of that word.

In this dissertation, the few conventions regarding satisfaction
measurement which do exist are followed. Specifically, satisfaction is
conceptualized as unidimensional, measured using a single 7-point item, and is
anchored by "extremely satisfied" and "extremely dissatisfied." In addition,
two levels of targets of satisfaction are used: satisfaction with a specific target

and overall satisfaction.

II-3. l . Expectation

Expectation is the subjective belief that an event or some attribute of an
event will occur, or that a product possesses some attribute, or that some
product will perform in a certain manner. Expectation can be measured at the
attribute and/ or overall level. It can be measured either with regard to
relevant attribute characteristics (e.g., "hot/cold," "dull/bright," etc.) or as an

evaluation of the attribute characteristics (e.g., "excellent/poor"). Olson and

43

Dover (1979), in an effort to clarify the exact nature of expectations,
demonstrated that expectations may be conceptualized as product-specific
beliefs.

Notions of expectation can be complex. Miller (1977) identifies four
types of expectations: the ideal, the expected, the tolerable, and the deserved.
Conceptually, a good case could be made for using any of the four, and the
appropriate choice may vary by situation. Although most research examined
here uses "predictive expectations," the alternatives have not been shown to
be inappropriate.

Czepiel and Rosenberg (1977) propose an algebraic version of the above.
They specify three types of expectation: 1) what the subject would like to
occur, 2) what the subject thinks Mid. occur, and 3) what the subject thinks
w_i11 occur. They suggest that the numerical difference between one of these
expectations and what actually occurs should be summed on an attribute by
attribute basis to calculate customer satisfaction. Several of the studies
examined used variants of this approach.

Swan (1981) contends that these variations of expectations are actually
points along an "expectation continuum." Consumers have a "modal
expectation level" that incorporates the probability of occurrence and the
evaluation of the probable occurrence. Thus, for example, consumers take
into account all the different types proposed by Miller (1977) and arrive at a
realistic modal expectation for a given situation. Simply asking the consumer
what he or she expects to occur or what quality is expected yields an overall
expectation rating. Though other researchers do not refer to the "modal

expectation level," many use this formulation.

44

Expectations have also been linked to attitudes. For example, Oliver and
Linda (1981) state: "Perhaps the most general workable definition is that of
Katona ( 1964) who referred to expectations as a 'subclass of attitudes that point
to the future (p. 34).' As such, they may be seen as prediction of affect" (p. 88).

Expectations is one of the earliest and perhaps most consistently studied
constructs in the satisfaction formation literature. It was used as an
independent variable in early studies of perceived performance. (e.g., Cardozo
1965; Olshavsky and Miller 1972) and in nearly every study of antecedents of
satisfaction. Further, it is the construct most closely connected to the
theoretical roots of the satisfaction literature.

In the service satisfaction literature, measurements labeled
"expectations" usually refer to "desires" (e.g., desired level of service). In

order to avoid confusion, this distinction will be clarified as necessary.

Table 3 lists the studies which utilize expectations in a satisfaction
formation model. It indicates whether expectations are measured and/ or
manipulated experimentally, and whether measurement occurred before or
after "performance." Table 4 indicates whether expectations are measured by
overall versus attribute-specific items, the number of items and scale points
used, question wording and scale anchors. The following are observations
regarding these tables.

(1) Although most of the studies reviewed included expectations as a
potentiaLantecedent varia_bl_e, over one-third did not (see Table 3). Two
additional studies, Cardozo (1965) and Olshavsky and Miller (1972), manipulated

expectations in a laboratory setting, but did not measure them. Tse and Wilton

45

Table 3
Measures of Expectations

Not Manipu- Before After
Measured Measured lated Performance Performance

Anderson and Sullivan 1993 J J
Babin, Darden and Griffin 1993 J

Barbeau 1984 J \/

Bearden and Teel 1983 J J

Bitner 1990 J

Bolfing and Wooduff 1988 J J

Cadotte, Wooduff. and Jenkins 1987 J J

Cardozo 1965 \/ \/ \/

Churchill and Surprenant 1982 J J J

Cronin and Taylor 1992 J J
Drbge and Halstead 1991 J J
Halstead 1993 J J
Mano and Oliver 1993 J

Oliver 1993 \/

Oliver and Bearden 1983 J J

Oliver and Desarbo 1988 J J

Oliver and Linda 1981 J J

Olshavsky and Miller 1972 J

Patterson 1993 J J

Parasuraman, Zeithaml and Berry 1988 J

Prakash and Lounsbury 1984 J J

Richins and Blodi 1991 J

Sindr 1991 \/ \/

Spreng and Olshavsky 1993 J J

Swan and Martin 1981

Swan and Oliver 1991

Swan and Trawick 1981

Tse and Wilton 1988
Westbrook 1980

Westbrook 1981

Westbrook and Newman 1978

Westbrook and Reilly 1983

46

Table 3, continued

Not Manipu- Before After
Measured Measured Iated Performance Performance

v J
J
v \/
v v J
J
v
v
v v

47

.3: use; 23:3? - .33; 232:8.

8:522:
58» 5:32.... doc-Lacuna "835: 3

".326
”83.2. 2.

..Eo=ouxo.. . .uoou be) 3:.
.5535“. 03:28»

and» n 8.3. .9. 28.88 :8 .3.
3.8 .. 8.... .9. 528:882

.0328

859... 3 £3... .figxfia. .8828?
t8€§£9959 98:83... 859....
.838 .o 38.. .93.. 32.. "8.3.5..
9.5.3.... 2. .85... 3 23:10 $98.2
8.8» 583 5.3 85:8.6 322.... so... it. 8 3.2
Log 5.! 2538.3. 835 .06 2.8
0828.5 a 050%

.35» __a 3 Se. - .323 =3 2.?

seasons-E .0 88:23.. .828 .o no.9.
.E... .o c3832. .38.: .o 9.3...

.2855. - .295. hoe-8.. .383 883288...

.238 2.8.55. - .__o3 22.2.? 3335: o to”.

:53: I :30? (\2

Room .3280

e m >
k as v >
>

m m > >
n k >
k m >
s o >
o P P >

£3. .2. .53. so =Eo>o 3.5.3:
5.. 38m .35.: an

mcozmuoooxm Co $5302 050on

v «33

32 Been: a... «2.5

mom. «5:053 use 55.550

mum— 809.8

8.: 8:2... 2. £683 .588

8a, .3683 E. 8.28

mmm— 30h use canteen

Vow _. :38

mmmp Caz-3m us cocoa-.3

48

A90 .38 .8? Enema... iota .o. 33> uooa
.823 uooo "0.5.9.. 35953 coon .o. .66.
A6: 51.9 ace... .32 .o 3.5.. do: .828
woo. a. s... :2. 2.; 8:8 68. .o 33
"83.9.. «8353 5:838»... poo. .o. .6...
.6523.
«.9228... 0.2.2.8 8o .3 5.2.2. 8 2s... a: .82.. a 39.. 3.2 k 2: > $2 .3833 2.. .39....

r328. 3. 3.! .38 3 8 .88..

2.. 88.x. _ 628.28 «8.5 ..<. .335.
.858... 29.2%. - .85. 282.». .326 a... 8.3.3. o. .o. 3.2 m 2 > > 82 :88...

£88.? 83....- one}: ”329.8 9.5.. :3
58...... 83.2.. 23.8. 588 9.5.. so: > > 2.: 3...: 2a 2.225

mwmp coin use .025

$3.38»? .. . .ulfiuu.

€9.38... - .288... ...&1.5. - $9.2. ".335
.335». 53 8
2825.32.39... 3.953202%...

$2.38»... to? . .33...» 22,. .6... 3.3.8 so... 88...... 8 E8882 no.3. . 2 > > 8.: an... E. .36

.955 38.. 8m 28.88 53.. 2...: > 8.... 8.88 .8. .56

.3... - .uoou. ”8.09:0 03.89. gone .0: 2808.8 c.5515

g
.528. . .855 8. c982 8393 3 9.3.88. 288.58 .38 A v. > > 82 528.. B. .55
0.3m scan-ad Sea. .2. IE3. no .335 32...»:
.8. 38m .35... .6

33528 JV 0.2....

t'I-Flcfithlni “V Chafillhl

49

.835. . .05.. 28.0.88. .....on 8 a. 388.8 so» 8... on ....s .8 so» too.

05 up... no» 8a 0.8 26.... 63:3. .26. do...

$3.583»... to? . ....omao... 28.288. ......3 :o» 3:02. .3 o. u. 6098
so» o.... «a... o. woo. 2: .... 9.0.8838 “8.3.022

25.8. .96. ...... do... .. 5.8388 38.8....

. 228%.... ......s 8.8.85.5. ......26
......3... 5.8.... 35.82.. ......E ..3.
5885552.. ......322... .3... 8 ......
....ot8 ...... .. .8...» 5.8... .o 88.5.98...
3308303833.. L032. 2.9.0..

£55.05“. 035.com «6:8... .0 Eco... .0982... "83953
......a... €31.33... .....Eo .
33.3.. 52.35.... . angina .02. .0350... ..o.oo.. ..flofitafi. ....Eiu. 833.5% m

......83 ...-.2...
82026 .....a 3.5.. I: c. 3.23 89.. 3....
655.. 05 .... so» 8
8.23 .3qu 02a ...... 8039:. .30» 028....
£038..
So» .0 23 8.5 8 :38 83.5059 8.8....
.82.... 8 .3 8628......
......32. 2.... o... .2. .. a. as...
.29....3 be... - .29... be... _ .28.. o... o. ......38. 2.. 8:8... .5.A 8.82.

0.08 .3330

vouczcou .V 0.9....

me— xgiflh Ba :95

8a.. £95! I3 anm

8.: b.9230 u... 22%..

8.... .2...

50

..0.5.. .60). - .00.. .005 a.

.voom b0? . .30.. b0...

.uoou .00... - Loon b0).

boom .00? - ..ooo b0?

0.08

..Eonfluoaxo 0005 .o .25...
...... ...... garage... 368.... 5. 883:... I .

..coalE.....80.u.
00.3000... 5.. ..338935. a. .833 .052

....u. on too. .0 ..000 0.5..
no» 9.5.62.0 05.00.83 .000 8 509.0 332..
o». «05 09.858395. "oglEotoa “.300an
503.8... 05 act-5.5.. ... >05... .56 So» .o
on» 0882.. :2. .. 0.88.... .3280... 3
use... .2. oflgota...u 6389a 038......
.205. 28..
0550...... 0 ... 000 8 0.... 2.63 no» 0835?
.o 802.9..8 .... .58.. 8... =5 .23..

a .o 8:052... 081.85.... 68...... .8... m R

cot-0:0 E00. .0: 083. .o
.8: 0.00m .0952

coactcoo ... 030...

\(

..805

\1.

8.3.32
an

8.: .....s. .8. .8380...

cam— x8530;

mom— can—ts .80 3h

51

(1988) and Churchill and Surprenant (1982) manipulated and measured
expectations.

(2) Though most studies measured expectations prior to product/ service
exposure, five did not. "Predictive expectations" measured following the
experience is really a measure of remembered predictive expectations, subject
to modification due to the passage of time, the known performance, the
subjects' reaction to the performance, and other potentially confounding
factors. Thus, it is generally considered to be an inferior measure of
predictive expectations.

All five studies which used post-performance measures of "predictive
expectations" were field studies designed to avoid the artificial environment of
a laboratory setting. Some of the problems involved in using ex m measures
of expectations are discussed in Drége and Halstead (1991), Halstead (1993), and
Yi (1990).

(3) Table 4 provides the details of how expectations were measured for each
of the studies reviewed. Note that all but three studies measured attribute-
specific expectation or attribute-specific and overall expectations. Two of the
three studies which did not measure attribute-specific expectations, Cardozo
(1965) and Oliver and DeSarbo (1988) both were laboratory studies and both
manipulated expectations regarding performance. However, even in these
cases, subjects were provided with a few attribute-specific cues, so it could be
argued that all studies but one either manipulated or measured at least some
attribute-specific expectations.

(4) Table 4 also indicates wide variation in the number of items used to
ogrationalize expectations. At one extreme is the study by Tse and Wilton

(1988) which used 27 items; at the other extreme are studies by Swan and

52

Trawick (1981) - which appears to use a single item scale created by
multiplying an expectation times the probability of its occurrence - and
Anderson and Sullivan (1993).

(S) The number of scale points per item likewise exhibits some variation
ranging from four points to ten points, with seven points also representing
the mode.

(6) Note that Westbrook and Reilly (1983) allowed subjects to choose those
characteristics about which they recalled expectations in an open-ended
format. Thus, the number of items on which expectations were measured
varied from subject to subject. (This measurement technique may have
contributed to the failure to develop a significant overall model.)

(7) Variation is most evident in the specifics of question and scale wording.
Note first of all that actual questions are rarely provided, though attributes
measured are usually specified. Likewise, scale anchors are often omitted
though readers are told that "semantic differential" (or "bi-polar") scales were
used.

(8) On close examination of exactly what was measured, it is questionable
whether one of the studies which claims to measure expectations actually does
(Westbrook 1980). In reality, he measures performance relative to
expectations, or disconfirmation.

It should be noted that two additional studies claim to measure
expectations but do not and thus are not included here. Parasuraman,
Zeithaml and Berry (1988) and Cronin and Taylor (1992) actually measure
"desires," not expectations. Both studies use the SERVQUAL scale which
consistently used the word "should" in all items (i.e., "Please show the extent to

which you think institutions offering _.__ services should possess the

53

features described"). This could be interpreted as normative or idealized
expectations, but they clearly are not expectations in a predictive sense.

(9) On close examination, there are three kinds of "targets" of expectations,
i.e., expectations are measured with regard to three subtly different aspects of
product/ service performance. First, expectations are measured with regard to
perceptions of objective performance. Most of the studies using semantic
differential (or bipolar) scales fall into this category. For example, Cadotte,
Woodruff and Jenkins (1987) measure "speed of service" on a "fast - slow"
continuum. Second, expectations are measured with regard to overtly
subjective evaflltions of performance. Scales such as Oliver's (1980) "good -
ba " scale or the Tse and Wilton (1988) "very poor - very good" scale fall into
this category. For these scales, the anchors themselves are evaluative. Third,
expectations are measured with regard to the level of satisfaction expected
from the product/ service performance. In other words, respondents are asked
their expected reaction to an expected level of an attribute. See for example
the question wording used by Oliver and Linda (1981) as indicated on Table 4.
10) Finally, a couple of studies use a combination of types of ta_rge_t.
Churchill and Surprenant (1982) use evaluation of performance as an overall
expectation measure, while Swan and Martin (1981) use anticipated level of
satisfaction as an overall expectation measure.

It seems likely that as expectation measures move farther away from
objective performance as a target, individual responses become less and less
comparable across respondents. For example, one person may expect "poor
service promptness" and another may expect "good service promptness," and
both may expect to be served in 15 minutes. The problem is exacerbated when

expectations are compared with performance. For example, assume both

54

people were actually served in 20 minutes. The first person may be pleasantly
surprised if he dislikes feeling rushed. The second may be unhappy if he
dislikes waiting. Not only are expectations and performance confounded,
expectations and desires are also confounded in such scales.

It may be true that expected evaluations of and expected satisfaction
with performance is a superior predictor of satisfaction because predictive
expectations and expected responses are combined in these two measures.
Nevertheless, expectation of objective performance is closer to the original
conceptualization of expectations as a cognition-based construct.

Yet even commonly used measures of expectations regarding objective
performance are flawed. For example, using "fast" and "slow" to anchor
expected service promptness measures may be more "objective" than using
"very good" and "very poor." However, even the former set of anchors is
subject to most of the same issues regarding individual interpretation. In
addition, it is difficult to envision a truly objective expectations measure for
some attributes such as "food quality" in Cadotte, Woodruff and Jenkins (1987).

In summ expectation appears as a construct in about two-thirds of
the studies reviewed. Most studies used predictive expectation measures
obtained prior to performance, though several actually measured expectations
following performance. There is wide variation in question wording, the
number of measurement items and points per scale item. Expectations have
been measured with regard to (1) objective performance, (2) evaluation of
performance, and (3) level of satisfaction.

In short, operationalization of expectations has suffered from three
inconsistencies. mat, it has suffered from the problem of which attributes to

include. Even when attributes have been identified by extensive pretesting,

55

there is no guarantee that salient attributes are not omitted. When subjects
are allowed to state which attributes are salient to them personally, variation
between subjects may contribute to problems in fitting the model (e.g.,
Westbrook and Reilly, 1983). Second, operationalization of expectations has
also suffered from the variety of expectations targets, i.e., expectations
regarding the product service performance versus expectations regarding the
level of satisfaction likely to result from the product/ service performance.
Further, within the realm of expectations regarding performance,
expectations may be measured with regard to the performance itself or with
regard to subjective evaluations of performance. 111i_r_d_, operationalization of
expectation has suffered from lack of comparability of expectations across
subjects, or from one construct to another within a single study.

In this thesis all three of the problems apparent in the current

 

literature will be addressed. First, the problem of relevant attributes is solved
by asking subjects specifically what proportion of the popular vote they
expect each candidate to receive. In other words, the target event is so specific
that it is defined as possessing a single attribute. Expectations regarding this
single attribute can be measured exactly. Similarly, the target also possesses
an unambiguous measure of objective performance, and thus the expectation
measure is of predictive expectations. Finally, the measure is on the same
scale as two other relevant constructs: desires and actual performance.
Therefore, more than in any other study published to date, responses across

respondents will be comparable.

56

II-3.2. Desires

"Desires" is a general term indicating the direction and strength of
what a person wants to receive from a product or service. Although it seems
logical that "desires" would be modeled as an antecedent of satisfaction,
especially given early definitions of marketing, "desires" has only recently
become the focus of serious attention in the satisfaction literature (with one
early exception: Barbeau 1984). In at least three ways, however, "desires" has
indirectly entered the satisfaction literature.

First, desires appears to have explicitly entered the satisfaction
literature as a disconfirmation standard rather than as an independent
construct. Westbrook and Reilly (1983) introduced the concept of "value-
percept disparity" in a study which was marked by both design and
measurement problems. Value-percept disparity was operationalized as "....the
extent to which the respondents' automobile provided the features and
performance characteristics needed or desired" (p. 258). As such, it may be
thought of as a kind of "desires disconfirmation"; desires were not measured
separately (though expectations were).

Second, Sirgy (1984) manipulated desires in an experiment by asking
respondents to imagine an "ideal" level of an attribute. However, he did not
measure the effects of the manipulation on satisfaction.

Third, desires have been introduced to the satisfaction literature
through operationalizations of expectations. When respondents are asked to
state expectations on an evaluative scale (e.g., "very poor" -- "very good"), the
implication is that expectations occur relative to some "norm" or "ideal," which
appears to be closely related to a "desired" level. For example, Tse and Wilton

(1988) conceptualized "ideal expectations" as the "anticipated performance of a

57

player that 'has exactly the combination of attribute you would like to see....'"
(p. 206). Likewise, as we illustrated in the previous section, Oliver (1980)
operationalized expectations as the sum of the cross products of attribute
specific beliefs (degree to which product possesses an attribute) and
evaluation of the attribute (degree to which attribute is good/ bad). The
evaluation component reflects respondent desires relative to that attribute
(Spreng and Dixon 1992).

Woodruff, Cadotte, and Jenkins (1983) suggest a move away from pure
expectations to an antecedent which is more desires-based: desires as
constrained by realistic expectations. In reporting the results of an empirical
test of their ideas, Cadotte, Woodruff, and Jenkins (1987) summarize their
position as follows: "....consumers are likely to rely on standards that reflect
the performance a consumer believes a focal brand should provide to meet
needs/wants....These norms have two important characteristics: (1) they
reflect desired performance in meeting needs/wants and (2) they are
constrained by the performance consumers believe is possible..." (p. 306).

Since desires have appeared explicitly in so few studies, its specific
operationalizations will not be presented in a separate table, but will be
presented in the discussion which follows. In 1984, Barbeau published a study
which explicitly compared the effects of expectations and desires on
satisfaction. Both direct effects on satisfaction and indirect effects (via the use
of each as disconfirmation standards) were measured. Despite the promising
results when desires was used as a standard for disconfirmation, desires was
not included in a product-oriented model again until 1990, when Spreng uses
desires as the centerpiece of his satisfaction model. Spreng and Olshavsky

(1993) measure both "desires" and "desires congruency" in an experiment

58

involving subjects' responses to a new camera. Desires is operationalized
using both attribute and overall measures. The question is worded: "Given
your individual photographic needs and desires, what would be the level of
__ you desire?" "Clarity," "sharpness," "color," and "overall picture quality"
are the four items for which the question is asked.

In the service satisfaction literature, the construct labeled "expectation"
usually is measured as desires. In fact, Parasuraman, Zeithaml, and Berry
(1988) state " in the service quality literature, expectations are viewed as
desires or wants of consumers, i.e., what they feel a service provider gro_ulg_
offer rather than M offer" (p. 17).

There has been some discussion about whether desires is best
conceptualized as lying on a "more is better" continuum versus an "ideal
point" continuum (e.g., Spreng 1992). There are some attributes for which
"more is better," such as "quality" or "value." Other attributes, such as
"warmth" have an ideal desired point because it is logical for something to
possess either too much or too little warmth. Recently, Westbrook (1987) and
Westbrook and Oliver (1991) have investigated the dimensionality of affective
responses to performance. Yet neither of these studies looked at the pre-
performance construct of desires.

Problems involved with measuring desires on an evaluative scale are
evident. For example, Barbeau (1984) used scale anchors "extremely well" and
"extremely poorly" to measure desired level of attributes. If one attribute is of
the "more is better" variety (e.g., in that study, "development of marketing
skills"), why would someone desire "extremely poor" performance? Why
would someone desire any performance other than that which is "extremely

well?" If another attribute in the same study is of the "ideal point" variety,

59

(e.g., in Barbeau's study, "degree of challenge"), does a high rating indicate a
high degree of challenge or a level of challenge identical to the subject's ideal
level of challenge? Thus, when "desires" scales are evaluative, is difficult to
interpret responses logically. Also, it seems that there would be minimal
variance in evaluative desires as everyone would want attributes at me_ir ideal
level or at the maximum level of a "more is better" attribute. Further, it
appears highly questionable to create a single desires indicator by summing
across a mixture of "more is better" and "ideal point" measures.

One additional issue is the degree to which respondents take other
attributes into account when rating their desires. For example, an individual
may desire a car which is "highly luxurious" but may temper that desire with
the knowledge that they have $10,000 to spend on the car. Thus, the context in
which desires are measured is critical for understanding results.

In this dissertation these problems with desires measurement will be

 

minimized. Respondents will be able to specify a level of desire on a non-
evaluative, unambiguous scale whether they personally perceive desires in
this context to be of the "more is better" or "ideal point" variety. Desire is
operationalized as the degree to which a respondent wishes to distribute choice
points to one alternative relative to two other mutually exclusive choices.
Respondents can allocate all 100 points to their preferred alternative ("more is
better") or give the preferred choice some specific number of points less than
100 ("ideal point"). In other words, desire is measured with regard to a total
"package" of attributes relative to other "packages" of attributes rather than to
a single attribute. An additional benefit of this operationalization of desire is
that levels of desire can be directly compared with other relevant constructs

(e.g., expectations) and across individuals.

60

II-3.3. Attitudes

Attitude has been used as both an antecedent and consequent construct
in satisfaction formation models. Attitude is generally considered to be an
ongoing or persistent affective orientation to a particular "attitude object," or
a subjective overall evaluation of an object.

The most widely used attitude model is the Fishbein model (Fishbein and
Ajzen 1975). The model defines attitude as the sum of the cross products of the
strength of belief that an object possesses an attribute and the evaluation of
the attribute, across all salient attributes. In other words, attitudes possess a
cognitive component (very similar to long-term expectations; see Katona 1964)
and an affective component.

Only three studies have utilized attitude as an antecedent variable in
satisfaction formation models, so a table providing operationalization details
will not be presented. (In addition, all three studies used similar attitude
measures.)

In Oliver (1980), Oliver and Bearden (1983), and Oliver and Bearden
(1985), attitudes are measured using 9-item semantic differential scales. In
Oliver (1980), attitudes about receiving flu inoculations are measured, while in
the latter two studies, attitudes toward diet pills are measured. Exact
operationalizations of attitudes in these two studies are not presented.

Theoretical justification for including attitude as an antecedent in
satisfaction formation processes is that if attitude is a persistent orientation to
an object, attitude may impact satisfaction via its effect on perception of
performance and disconfirmation (Oliver and Bearden 1983). In addition, Yi

(1990) indicates that attitudes are hypothesized to affect expectations and

61

perceived performance judgments; this is thought to be especially true among
low involvement consumer, because low involvement consumers are thought
to be less attentive to actual performance and thus rely more on pre
performance factors in forming satisfaction.

In this dissertation, a multi-attribute conception of attitude is used. It is
operationalized using four attributes and is constructed using a modified
version of the Fishbein model, with attribute "importance" rather than
attribute "evaluation" as the factor with which belief is multiplied (Fishbein

1967)

II-3.4. Perceived performance

Churchill and Surprenant (1982) were the first to suggest that
perceived performance affects satisfaction. They hypothesize the influence to
be both direct and indirect (through disconfirmation). Tse and Wilton (1988)
extend the theoretical justification for including perceived performance in
the model. They suggest that sometimes consumers have a "learning motive"
for consumption, i.e., consumers want to learn about a new product; consumers
may thus be satisfied with the product "regardless of the levels of pre-
experience comparison standard and disconfirmation" (p. 205). In addition,
they hypothesize that the consumption experience M, especially if it is
extremely positive or negative, will affect post-consumption evaluation in
accord with cognitive dissonance theory, and will thus have a direct effect on
satisfaction.

In some previous studies, constructs which are labeled as "product
performance" actually are operationalized in a manner very similar to

disconfirmation. For example, Olshavsky and Miller (1972) asked subjects to

62

rate overall and attribute specific tape recorder performance relative to ".... an
ideal or perfect recorder" (p. 20). By providing subjects with a comparison
standard, the construct "performance" at best is confounded; at worst, it may
be considered to be a measure of disconfirmation.

Table 5 lists the studies which utilize performance in a satisfaction
formation model. It indicates whether performance is measured and/or
manipulated experimentally, whether performance is measured by overall
versus attribute-specific items, the number of items and scale points used,
question wording and scale anchors. Table 5 indicates the following:

1) Although the studies presented in Table 5 all claim to measure
performance, in at least two of the casesI the operationalizations of perceived
performance appears to be confounded. These studies, and the potential
confound in each, are:

a) Fisk and Young (1985) explicitly measured performance relative
to a standard ("waiting time" relative to what was "expected" and "fare paid"
relative to a "friend's"), and therefore it appears to be more like a
disconfirmation measure.

b) Olshavsky and Miller (1972) explicitly measured performance
relative to a standard ("....compared to an ideal or perfect recorder...."), and
therefore it appears to be more like a disconfirmation measure.

2) Cronin and Taylor (1992) and Parasuraman, Zeithaml, and Berry (1988)
both measured performance using the SERVQUAL scale. The Question wording
is somewhat confusing in that subjects are first told that it's their "feelings"
which are of interest (an affective cue), and then they are asked to rate how
strongly they agree or disagree that a provider possessed or exhibited a

specific attribute (a cognitive task).

63

80:. .mun 2 .o». 8.5 998.5 .m> .85. 33.3952
89: ...caoc N. 8 .3556 o. 2:; 0:36; 85098 .m> .25» 953;. 35:95: > > 33 33> pen in.

was. .2333 .... .8 8x2
..Eoesau 2: 3 39:83 838. 9: 8: ~>x
26:3 :0» 523 2 $068 05 :65 885
48363 er Son- moc=02 50»

.85» zucozm. - .8536 20:87.. 3 828 85888» 3 you. 3.26:8 2.... u mm > > mm? .033. new 5:20
fission. - 29:8. m. :5 o5 .0 £26 ...BoS. 3.208
1888.552: :93 ..cfi .58 Hosanna B x cm; > > > > 32 22293 2.. ...;236
2.. es 8. a S ta: _ _ > > 89 38.6
.835

098.an __o 5:26. $23395. 3695520.
Louooxsofimoes. .88....83 85%.?
8.8... 585 .858 .0 39%. 19.3.. 82. 68352 m o > x. > 32 255.. u... £3803 6:88

assets. 22%.: I 3.28% so: toss

585... .3 S .8. - .35... :9; be? ..85 5? £585? E96 .3 «co 5 a > > > 32 5.325 95 95.8
.98.... €585. - ..Es $585. $835.." w to“. k m > > 32 38.8
> 89 5:5 2.. 828 533
$3293. - Eng-.5. .5526 520 62k. 883.
.52. - .30.. .83 8% .Eaaa 8!. o. N > > 82 52:3 v:- 8885.
Snow 8.50:0 ES. .2. «£8. .0 __20>o 3:232 noun. v2.5.0! team-ox
3.. 2.8 3.52 .6 52:0: 32

oocwfiuotom no «2332 0:3on
m Ban...

64

.333 “5.85. .535 23%. :23? 2.8.3 F
- .933 88 .225? 128. ..mflceofi. .5220. «835: m

.83 88 ”828. 8:55: .33 .2 :93

38 582, 2.2. 6810 3:26 Jon 828

Bo. .u. 9:. so» is 258 68a .0 an:

”320:. 9:558 5:836? 38 .2 :98

23:28.6 ”5588 .335“. :08 co negated 93.5 a :6: 09.2

.8586 zocotm. - .35. bucobm. .526 P pco «93.53 9

..ucocasfi 2: B vacuums 238. as 8: 69
31.8 3» 62; 2 “Sea «5 32m 88:
.9388 S Sons 35.00.. 50»

.85u 29.8%. - .8585 29.9%. 8 822 3:068.“ ..o “8 3.32.0. on».

nzofi<z~=uzoom§ AI doc-gran =wB>o
.5882 SEE: .235 8.. £8: .85, .8 5:8: .6556

«o .33. 2( 9. 035.28.. ”952 in :35: ..o)! 3.0: 059.308 60:58
.Qnfimon ~20! - 03.38 «mom. -2“. Eat Eaton...— .omconm2 A8252. $3353

223.288 98. 6m 89 “5:28 532 :9:

G18... cat-“...“: 2. 3:82 225. 2.
2.8 u... no.8» 505.9... $3203 8:353
n_o_ucocotfi 332.com oucweuoton “cc .5080ng "2 con-3.90 “9:08..

8.2.8
.....uaona 0.253 .02 .... .58. «05.00.. 03>...» 9.3 E00303 8.28 .ucoEEoSco
pole 30: ..-... Son. 8293 .03 _. Evan .ucoEcoutEo .aufixza 603.05 8§tuu<

1.8. 8.880

poaczcou .m 03m...

n h\—_.

k v

53. .8 2.8. .0 .356 83.32

2.. 1.3. 33:52

\(

>0

93-. 923-0: 023-0:

-323!

\v

no.2 b.9220 8. 88%

v8. 5.328.. Ea 53.2..

mam — c8353.

39 Pom new _EE._oN .caecam8&

~02

wk: .55: u... b.9656

mmmp 096060 new 52.0

82 82.0 on. can:

002 302933 pco ucatno

65

..82 _ .2; 2.8.8 832..
- .muooc an. c2.» 80. .2 332a.

Loon. - 22.85.

.8386 39.25 - 85a .2395.

3:89. .8

Room

£3?de EwUxma-w:4<>.
fluz<2¢0umwa “.0 mung! 95a ...02 HPPOZ
.wuooc 805 SE 0.5023... 29.53 9 0803 3!.

co...» 9... 6.50833. :- c. 38: .392 38.33 m ~ \r
oocoEBton
uooo .m> oucactotoa .00.... 6323.5: \r >
....63 too. 05 20305 39A 26.. m: =8 3.8.... n .. \z

...: 3288 5.56.8 6:. 5.3 05.9% ....

c. Sn 558.8 .2. 5.3 8.38 ....

3208. . 62.8.8 5.; 3.38 s.
c. 5.. . 6.58.8 5.; 8.38 s. k 8 >

.553..sz ..Ego..... ...226

.05.? .88.... 35.22.. .82.. =2...
.BBocaciis. bias? .3... 2 :2.
L525... 3.... .05.?» 3:3... .8 $29.68....
.motoamouuohcozqo. 1.038 8.05.

Lots... .o 58... 39.8.... 63352 u m. > >
8.30:0 £3. .2. 2:3. .3 .....o>o 3.5.52
3.. 0.3m .363... Am

non—5:00 .m 035...

$30. 02:30: 3.330:

LEE-a

no:

32 »...3. .8. .6988;

comp 202.3 0:. 8*

—om_. x239; Uca :95

amp .025 0:9 :95

amp Eta! Dan cgm

66

3) In five of the studies, performance is manipulated; in two of these cases,

perceived performance is not measured, so actual performance becomes the
relevant independent variable. Performance and expectation are the two most
commonly-manipulated variables in satisfaction formation research because
they can be consistently and easily controlled. The motivation for
manipulating performance is not only to manipulate perceived performance,
but also to manipulate disconfirmation. Disconfirmation has been the focus of
so much attention in satisfaction research and there is no readily apparent
way to manipulate disconfirmation directly.

4) The mgioritv of studies mea_suredjattribute-specific performance or both
attribute—specific and overall performance. Only three studies measured
overall performance alone, and these were studies involving

products/ services that are simple to evaluate (e.g., pen, ROI, etc.)

5) Except for the three studies which use only an overall measure,
perceived performance is measured using 7-point scafis (range is from 5-

points to 8-points). Number of items used ranged from 4 to 28 items.

In this dissertation perceived performance is measured in one model

 

and not measured in the other. In the model which includes perceived
performance, it is measured conventionally, that is, using 7-point scales and
multiple attributes. In the other model, perceived performance is not
measured because actual performance (i.e., the election outcome) is so clear,
and so widely known and accepted, that an objective measure of perceived

performance is considered unnecessary.

67

II-3.S. Disconfirmation

Disconfirmation is a person's evaluation of product or service
performance relative to some standard. Performance above the standard is
traditionally referred to as "positive disconfirmation," performance below the
standard is referred to as "negative disconfirmation," and performance at the
standard is referred to as "confirmation."

Disconfirmation has been operationalized in a wide variety of ways.
Table 6 and Table 7 list all the empirical studies reviewed which utilized
disconfirmation, and certain relevant characteristics of how it was
operationalized. All operationalizations can be categorized along two
dimensions: 1) subjective versus derived (computational) measures, and 2)
standard which is being disconfirmed. The resulting matrix and the
classification of studies utilizing disconfirmation as an antecedent construct is
shown in Table 8.

The first dimension, subjective versus derived measures. refers to a
critical conceptual difference in how disconfirmation is ogrationalized.
Subjective disconfirmation refers to the direct measurement of a person's
evaluation of disconfirmation. Such measures are taken following the
subject's experience of the product or service. Typically, the subject would be
asked to rate whether the performance was "better than expected" or "worse
than expected" on some scale. Reporting overall disconfirmation in this
manner allows the subject to consider the entire experience relative to
expectations or whatever standard is used.

Derived measures of disconfirmation are indirect measures in that the
subject is not directly asked the degree to which performance meets the stan-

dard (e. g., expectations). Instead, measures of the standard are collected before

Measures of Disconfirmation

Am and Suan 1993
Babin, Darden and Griffin 1993
Barbeau 1984

Bearden and Tool 1983
Bitner 1990

Boiling and Woodruff 1988

Brown and Swartz 1989

Cadotte, Woodruff, and
Jenkins 1987

Churchill and Suprenant 1982
Cronin and Taylor 1992
Drape and Halstead 1991
Oliver 1980

Oliver 1993

Oliver and Bearden 1983

Oliver and Desarbo 1988

Oliver and Swan 1989
Olshavsky and Miller 1972

Parasuraman, Zeithaml, and
Barry 1988

Patterson 1993

Prakash and Lounsbury 1984
Richins and Bloch 1991
Spreng and Olshavsky 1993
Swan and Martin 1985

Tse and Wilton i988
Westbrook 1980

Westbrook and Reilly 1978

Derived or By
Subjective Attribute
subjective
subjective
derived V
subjective
7
subjective
derived V
subjective
subjective V
NOT MEASURED
subjective V
subjective
subjective
subjective
derived V
(manipulated)
subjective 7
subjective V
derived V
derived V
derived V
subjective
subjective V
derived V
subjective
subjective V
subjective V

68

Table 6

Overall

in Empirical Studies
.. ________ Standard Disconfirmad---_....-_-.
Product Best
Predictive Past type brand
Expectation Desire Experience Ideal Norm Norm
V
V
V V V
V
V V V
V
V V V
V
V
V
V V
V
V
V
V
V
V V V
V
V V
4
V" V
V
V

" predictive and equitable upectations
(2 treasures)

Anderson and Sullivan 1993

Babin. Darden and Griffin 1993

Barbeau 1984

BeardenandTeel 1983

Bitner1990

Bolfing and Woodruff 1988

BrovvnandSwartz 1989

Cadotte, Woodmfr, and
Jenkins 1987

Cardozo 1965

Churchill and Surprenant 1982

Cronin and Taylor 1992

Drbpemdllalstead 1991

Oliver 1980

Oliver 1993

Number Scale Pts.
of Items per Item

1

65

“42

69
Table 7

Specific Measures of Disconfirmation

10

7

Question
N/A

e.g., 'The product materials improved my score
by about as much as I expected it to.“

(perfomance minus standard) for 8 attributes

assumed constant; verified. but scales not provided

Askedoncel‘oreachstandard.

(performance minus expectation) for each

ttribute

Asked once for each standard.

Assumed to be manipulate: not measured.

wmiimmmmmizmm

rezplayermdonceforeverall
e.g.rorcost:

E.p. for overall: “My expectation regarding the
perform or the product was.-.“

Asked regarding cupet (woe. durfiility,
stain resistance) and service (wermty.
retailer service)

Asked of innoculation recipients:
Problems encountered

Benefitsmceived

Aslned of non-recipients:
'Decision not to get vaccine made me ....'

askedol'benefits.problemsmdaverall

Scale
'Quality much worse/better than expected'

'Liltert scales“

'Extremely well - Extremely poorly'

'Betterthci-Worsethan'

'Much better than - Much worse than' each
standard

'Agree - Disagree'

'Mrchbetter-better-aboutthesame-
worseomuchworse'

'Too high: Worsethm l thought-Accurate:
Justeslexpecsed-Toolow: Betterthan
Ithought'

'Worsethu'ilthoupht-justaslexpected—
betterthanlthought'

'notaspood asexpected -better than
expected'

'Mrchmoreseriousthanexpected-mch
Ieesseriomtnanexpected'

'Hichleasthanetpected—chhgreeter
thanexpected'

'Muchworseofftlmexpecsed—Aswellofl
asupecteduMich better offthan expected'

'better than expected—worse than expected'

Oliver and Bearden 1983

Oliver and Desarbo 1988

Oliver and Swan 1989

Olshavsky and Miller 1972

Patterson 1993

Parasuraman, Zeithaml. and
Berry 1988

PM did Lounsbury 1984

Richins lid Bloch 1991

Spreng andOishavsky 1993

Swan and Mutin 1981

Tse and Wilton 1988

Westbrook 1980

Westbrook and Reilly 1978

Number Scale Pts.
of items per item

4

10

22

ll-7

12

2.

29

7

70

Table 7, continued

Question

Asked re: overall. benefits only. problems only,
and positive/negative surprise over performance

Tvvo levels were experimentally manipulated.

Asked to rate recorder 'compared to an ideal or
perfect recorder.‘

' NOTE: They claimed to be measuring performance.
7 attributes were frequency response. freedom from

distortion. background noise level. flutter. speed
control. fidelity for voice. fidelity for music.
Also asked re: overall.

(performance - std) for 10 attributes

(performance minus desires). both measured on
Likert scales

(performance minus W)
All measured on semantic differential scales
Ldielled normative standards as 'normative
expectations' and 'mmparative expectations'
11 attributes for resurrents. 7 for beer

'Thinking about the benefits you have experienced
with thiscar.wouldyousaythesebenefitsvmi

'Thinkingmitmyproblemsyou havehadwith
thiscarsinceyoupurchasadit.wouldyousay
theseproblemswere..-‘

'ln comparison w/ the quality level you expected.
howwouldratetheperfomwiceofthiscemera

with regard to (clarity, sharpness. color. and
ovrall nudity) of the picture?‘ (asked for each)

(performance ratings minus expectations) for
12 attributes (e.g., prestige. room of interior. fun
to drive. engine power. etc.)

Expectationsandperformnemeaauredonsematic
differentialscalee

'....overail....how close did the [product] come to
your expectations.‘

'Subjectsflrstaskedtolistgoodandbadpoints
ofproducts. Askedonceforeachbenefit:

Askedonceforeachdrawbacic
'Subjectsl'lrstaskedtolistgoodandbedpolnts

ofproducts. Askedonceforeachbenefit:

Askedonceforeachdrawbeck:

Asked once for overalle aspects:

Scale

'Greater than expected-Worse than expected'

“Better than expected - Worse than expected'

'Worst possible - Best possible“

(pert ‘ ima) - (exp ° imp)

'Agree - Disagree'

'mch less than expected - Mich greater
thanexpected'

'Mlch less serious than expected - Mich
moreserious than expected'

Morse than i expected' - 'better than i
expected'

'Very much poored than expected - Very
muchbetterthanexpected'

'mchlesathanlexpected-Aboutwhatl
upected—Wchgreaterthanlexpected'

'mchmoreserious thanlexpected—About
whatlexpected—Mlchiesssen'ousthm
W

'Nuchbetterthanlexpected—Aboutasl
upected-Muchworsethnlexpected'

'Muchbetterthanlexpected-Aboutasl
expected-Wehwsetiwllexpected'

'Muchbetterthanlexpected-Aboutasl
upectednmchworsetimlexpected'

Classification of Studies which

DISCONFIRHATION
STANDARD

Predictive
Expectation

Desire

Ideal

Product

TYPO
Norm

Best
Brand

71

Table 8

Include Disconfirmation

CALCULATION IETHOD

Subjective

Anderson and Sullivan 1993
Babin, Darden, and Griffin 1993
Bearden and Teel 1983

Boiling and Woodruff 1988
Cadotte, Woodruff, 8: Jenkins 1987
Churchill and Suprenant 1982
Drdge and Halstead 1991

Oliver 1980

Oliver 1993

Oliver and Bearden 1983
Richins and Bloch 1991

Tse and Wilton 1988
Westbrook 1980

Westbrook and Reilly 1978

Spreng and Olshavsky 1993

Olshavsky and Miller 1972
Tse and W'Iton 1988

Boiling and Woodruff 1988

Cadotte, Woodruff, and Jenkins 1987

Boiling and Woodruff 1988

Cadotte, Woodruff, and Jenkins 1987

Derived

Barbeau 1984

Brown and Swartz 1989
Oliver and DeSarbo 1988
Patterson 1993

Prakash and Lounsbury 1984
Swan and Martin 1981

Barbeau 1984
Parasuraman, Zeithaml, and Berry 1988

Prakash and Lounsbury 1984

Prakash and Lounsbury 1984

72

exposure to the product or service; after exposure, the subject is asked to rate
performance, usually for the same attributes and using scales identical to
those used to measure the standard. Disconfirmation is then calculated as the
simple difference between performance and the standard. Positive
differences indicate performance exceeded the standard and thus were
positively disconfirmed. The reverse holds for negative differences. A few
scholars have suggested that derived disconfirmation measures should be
operationalized as the ratio of performance/standard rather than as a
difference score (e.g., Bonoma 1987; Lele and Sheth 1987). Cooper, Cooper,
and Duhan (1989) tested whether the ratio operationalization outperformed the
difference operationalization, and found the latter to be superior according to
a number of validity related criteria. Thus, derived disconfirmation is rarely
operationalized as a ratio.

Subjective disconfirmation has been criticized for one major reason.
This measure allows subjects to alter their remembered standards or
performance evaluations or both in a manner consistent with dissonance
theory (Festinger 1957) or assimilation-contrast theory (Sherif and Hovland
1961). Such measures may include a significant amount of bias (Prakash and
Lounsbury 1984), and thus may not accurately reflect the degree to which
performance met or failed to meet some pre-experience standard.

Derived measures have also been criticized for several reasons. First,
such measures typically do not account for disconfirmation of standards of
salient attributes for which no prior measurements were obtained. For
example, if a study of satisfaction with an airline flight neglected to obtain
pre-flight expectations regarding how smooth the flight is, post flight

measurements of performance likewise would not capture customer evaluation

73

of this attribute.

Second, derived measures are difference scores, which have been
strongly criticized for problems of reliability, discriminant validity, and
spurious correlations (Peter, Churchill, and Brown 1993).

Another problem with the derived disconfirmation measure is
empirical: such measures are only weakly related to satisfaction. This
criticism appears to hold true even when the reliability problem is controlled
(Prakash and Lounsbury 1984). The only studies which have found strong
relationships between derived disconfirmation and satisfaction are those
which do not include an independent construct for perceived performance in
the model (e.g., Swan and Trawick 1981).

The second dimen_sion of TJable 6 is the standard actually disconfirmed.
Numerous standards have been proposed and tested. The following is a list of
major disconfirmation standards proposed in the marketing literature:
predictive expectations (e.g., Miller 1976), normative expectations (e.g.,
Granbois, Summers and Frazier 1977), deserved expectations (Miller 1977),
ideal expectations (e.g., Tse and Wilton 1988), comparative expectations
(laTour and Peat 1980), best brand expectations (e.g., Prakash and Lounsbury
1984), etc.

Although most researchers seem to agree that disconfirmed
expectations are what explains satisfaction, consensus has not been reached
on this issue. For example, Woodruff, Cadotte, and Jenkins (1983) propose that
performance norms play an important role in disconfirmation; the subject
compares product performance with a normative ideal. According to that
article, norms may be either brand-based or product-based. If product

performance falls within a "zone of indifference," the normative explanation

74

is confirmed; if it falls outside the zone, disconfirmation occurs and leads to
satisfaction or dissatisfaction. The degree to which expectations and norms
relate, or whether one should be used rather than the other is open to debate
(Hunt 1977). A related idea is proposed in LaTour and Peat (1979); they
propose the Thibaut and Kelly concept of using a "comparison level" (C.L.) for
calculating disconfirmation. The C.L. is based on all salient outcomes of the
same or similar occurrences or products in the past; the C.L is continually
updated based on new experiences. Satisfaction thus becomes a relative, not
absolute, emotional state.

A few studies have focused on directly comparing the performance of
various standards of disconfirmation. These studies, and their results, are
summarized next. Prakash and Lounsbury (1984) compared predictive
expectations, normative expectations and comparative expectations in two
different usage situations (fast food hamburger restaurants and beer). All
disconfirmation constructs were operationalized as computational measures
(i.e., differences between the pre usage and the post usage evaluation of
performance). They concluded that normative and comparative expectation
measures outperformed predictive expectations as predictors of satisfaction,
though the evidence for the observed differential performance is weak.

Barbeau (1984) compared predictive expectations, desires, and past
experience as comparison standards in derived disconfirmation measures.
Disconfirmation measures using desires and past experience were found to be
a significant antecedent of satisfaction, while disconfirmation of predictive
expectations was not.

Swan and Trawick (1981) directly compared the explanatory power of

derived ("inferred") versus subjective ("perceived") disconfirmation. Using a

75

series of regression equations, they concluded that both types of
disconfirmation were significant, though derived disconfirmation appeared to
outperform subjective disconfirmation in terms of explanatory power.
However, it should be noted that the derived measure implicitly accounted for
perceived performance; it is impossible to determine whether the
operationalization of disconfirmation or the inclusion of performance
accounted for the superior explanatory power of the derived versus subjective
disconfirmation measure.

Cadotte, Woodruff, and Jenkins (1987) compared predictive expectations,
product type norms and best brand norms as predictors of satisfaction in three
different restaurant situations (fast food, family, and atmosphere).
Disconfirmation constructs were operationalized as overall subjective
measures. They concluded that product norm and best brand norms
outperformed predictive expectations as a predictor of satisfaction; however, it
should be noted that the evidence is equivocal given that none of the models
estimated fit the data well. They also concluded that different standards may
be relevant in different use situations or among different individuals in the
same usage situation.

Spreng and Olshavsky (1993) compared an expectations-based
disconfirmation measure with a desires-based measure, desires congruency.
Building on the work of Westbrook and Reilly (1983) and Barbeau (1984), they
demonstrated, and theoretically supported, the superiority of the desires
congruency measure as an antecedent of satisfaction. In their model, the

effects of disconfirmation of expectations on satisfaction was not significant.

76

As we have seen, the traditional model of satisfaction formation depends
heavily on the concept of disconfirmation. At its most basic level,
disconfirmation is the difference between what a person expects (expectation)
and what he or she receives (performance). Logically, only three things can
vary in this definition: the definition of expectancy, the definition of
performance, and the operationalization of the concept of "difference."

The four most-commonly referred to theories in the satisfaction
literature are utilized in an attempt to explain the relationship among these
three constructs. Anderson (1973) examines what the four theoretical
approaches say about the relationship of disconfirmed expectancy and
perceived performance:

1) Cognitive dissonance theory specifies that a consumer will alter the
perception of performance to make it more congruent with expectations.
Therefore, low expectations will yield perceptions of lower
performance; high expectations will yield perceptions of higher performance.

2) Contrast theory specifies that a consumer will intensify the disparity
resulting from disconfirmed expectations. Therefore, negatively disconfirmed
expectations will yield perceptions of higher performance; positively
disconfirmed expectations will yield perceptions of lower performance.

3) Generalized negativity theory states that any deviation from
expectations will result in a negative disconfirmation.

4) Assimilation-contrast theory specifies that slight-to-moderate
deviations will result in perception of slightly higher or lower performance
(higher for positive and lower for negative disconfirmation) as predicted by
cognitive dissonance. However, as disconfirmation becomes more severe, the

perceived difference between expectations and performance is magnified, and

77

the results predicted by contrast theory appear to hold. Anderson's own
experiment supported assimilation-contrast theory.

Oliver (1980) acknowledges that adaptation level theory appears to
apply to the process of disconfirmation. The theory states that perception
occurs in relation to an "adapted standard." Once established, this adaptation
level serves as a standard for future evaluations. Oliver states that
expectations, "however created" (p. 461), function as an adaptation level.
Disconfirmation results when actual performance is compared to the
adaptation level.

Those few studies which have examined results in relation to these
underlying theories have consistently found support for cognitive dissonance
explanations (Festinger 1957) and for the assimilation explanations offered by
assimilation-contrast theory (Sherif and Hovland 1961). For example, Olson
and Dover (1979) express a fairly typical conclusion when they report that
their results ".... support a dissonance theory prediction" and later, that their

results "....indicate a consistent pattern of assimilation-like effects...." (p. 186).

In summm, predictive expectation has been the most commonly used
standard for the operationalization of disconfirmation. In addition, subjective
measures have generally been found to be superior to derived measures of
disconfirmation. Single-item measures of disconfirmation, either overall or at
the attribute level, have been most commonly used in previous research.

In this dissertation consistent with previous research, a single item, 7-

 

point scale measurement of subjective disconfirmation will be used; the

standard disconfirmed will be predictive expectations.

78

lI-4. Principle relationships among constructs

In Sections II-2 and II-3, we examined individually the principle
constructs utilized in traditional satisfaction models. In this section,
relationships among the constructs are examined in detail. The section is
organized around the direct relationships between possible antecedent
variables and possible dependent variables. In Section II-4.1, the
relationships between perceived performance and its possible antecedents are
examined. In Section II-4.2, the relationships between disconfirmation and its
possible antecedents are examined. Finally, in Section II-4.1, the relationships

between satisfaction and its possible antecedents are examined.

II-4. l . Antecedents of perceived performance: Expectations,
Desires, and Attitudes

Researchers have been aware that the perception of performance (or
product quality) can be affected by prior expectations. For example, Anderson
(1973) demonstrates that the rating of product quality of a particular item is
higher if expectations of quality are high and lower if expectations of quality
are low. Olshavsky and Miller (1972) demonstrate similar results using a
different product. However, the link to customer satisfaction was not made.
This may be because the researchers think the link is obvious, or it may be due
to a lack of interest in customer satisfaction until the early 1970's (Hunt 1977).

Table 9 lists the studies which modeled antecedents of performance
along with the product/ service context, sample composition and analytical
methods. Columns to the right of the Method column refer to possible
antecedents of perceived performance. The numbers in these columns

indicated the relative strength of relationship between the antecedent and

79

Eozsaiwev F

Foucaeuotoq 8288a
co oucmctotoa
35:98.: Fouozv

F

jcasb a
fiauu<

 

ou35m<

macaw
mucaN

F

58

80329.8
80328.8

man-«Fa
£8

Foo—8.50:3 8288a

co cow—3898

38.33an ”3028

850

02.3..an

N

F
F

F
F

988 F
2.8 F
2.8 F

F

2,88...

<>Oz<

33am .>OU
.uuaum .>OU

88:. .8
88:. .8

32% $8
32% $8
82% $3
533.62

8.329..

53852

F563.“

3:33.»
8563.».

8085......
E38

muCOUSHm

82805
8259..."

353%
353%

ggugu
mac—Dug
”BF—Bug

353%
353%

gggu
£83»

.935 882

809
Jam: U02 any.

882.
25:

85¢ 32

mini 89>
new...

82.
neon

.umm. 008 emu.—
.quL 000.— umfl
am». too“— was;

..go 0958
83 08:3.

3883
0.3.2:

£80 gggllluflfiwlllfiuflfi

e 882:8»? 8;; as at
w ... .N 882:3»? b.5833 .28 5888
m .m .F tooz..?mmF £583.. 88 £885

88F 8888

8:25:58? 3580:. baa. Sn sausage
oucaEBtoa 2:82: 8 0058B 2058 ”302

at: .2... .28 39238

N 800213me 35:05.3 van ___FU.=EUV
F BEIANQmF pancake—Fm 0:5 55538

N 382:5? 8888
FavozsammF 8888

m .ouozstmmF 83:8 8a 5280; .8888
N 38:..FBmF 8:8 8a 5.88; .3888
F moozzFBmF 83:8 88 £285 6888
~ 382132 38.88
F .80stng 3.88

AmmmF 52:3 new combuc<v

 

gum

o>_ua_omloocmE._o.F._on_ Fo macoumoouc<

m «33

80

performance. If a relationship was tested and reported to be not significant, a
"0" appears in the appropriate column.

The following observations can be made regarding Table 9.
1) Studies which examined the expectations-perceived performance
relationship found the relationship to be significant. It should be noted that
in Cadotte, Woodruff, and Jenkins (1987), expectations was reported to be a
significant covariate of performance; it was not hypothesized to be an
antecedent in that study. Also, Olshavsky and Miller (1972) report a
significant effect of expectation on perceived performance, but the dependent
variable they actually measured was disconfirmation. Still, the evidence
linking expectations and performance presented in the table is strong.
2) Two sLtudies examined the strengh of predictive expectation_s on
performgnce relative to the strength of other potential antecedents on
performance. Prakash and Lounsbury (1984) investigated the relative
importance of predictive expectations versus other types of expectations. Only
in the case of Tse and Wilton (1988) was expectations found not to be the most
important antecedent. Not surprisingly, Tse and Wilton found that
manipulated performance was a more powerful antecedent of perceived
performance than was expectations.
3) Barbeau (1984) is the only study which examines the relationship
between desires and performance, and the relationship is reported to be
nonsignificant. It is difficult to make any generalizations about the desires -
performance relationship given that it has been tested so infrequently in the

literature.

81

In this dissertation perceived performance is measured in one of the

 

models and, consistent with the satisfaction literature, predictive expectation

is included as a potential antecedent.

II-4.2. Antecedents of disconfirmation: Expectations, Desires,
Attitudes, and Perceived Performance

Table 10 illustrates exactly which standards are being disconfirmed in
each of the studies reviewed. Table 11 illustrates how expectations compare
with desires, attitudes, and performance as antecedents of disconfirmation in
the studies reviewed. Table 1 1 lists studies which model antecedents of
disconfirmation along with the product/ service context, sample composition
and basic analytical methods used. Columns to the right of the Method column
contain possible antecedents of disconfirmation. The numbers in these
columns indicates the relative strength of relationship between the
antecedent and disconfirmation. Thus, for example, in Churchill and
Surprenant (1982)-—Model 1 (plant), expectation had a stronger relationship
with disconfirmation than did perceived performance, yet both relationships
were statistically significant. If a relationship was tested and reported to be
not significant, a "O" is used in the appropriate column.

The following observations can be made from Table 10 and Table 11.

1) With the exception of Churchill and Surprenant (1982)-Model 1,

whenever expectations and perceived performance appeared in a model

 

together, the relationship between expectations_and disconfirmation was
weaker than the relationship between perceived performance and
disconfirmation. Given that disconfirmation is an evaluative response to

performance, this finding should not be surprising.

82

Table 10
Antecedents of Disconfirmation -- Standards Disconfirmed

STANDARD DISCONFIRBED

 

 

 

 

Experience/
Product Best
Predictive Type Brand
Study Expectatia Desire Ideal Norm Norm OthL
(Anderson and Sullivan 1993) V
(Barbeau 1984) V V V
(Bolling and Woodruff 1988) V V V
(Cadotte, Woodruff, and Jenkins 1987)-Model 1 V
(Cadotte, Woodruff, and Jenkins 1987)-Model 2 V
(Cadotte, Woodruff, and Jenkins 1987)-Model 3 V
(Churchil and Surprenant 1982)—plant V
(Churchill and Surprenant 1982)-video player V
(Oliver 1980)-Model 1 V
(Oliver 1980)-Model 2 \/
(Oliver and Bearden 1983)-Model 1 V
(Oliver and Bearden 1983)-Model 2 V
(Olshavsky and Miller 1972)
Note: authors claimed to measure performance V
evaluation but really measured disconfirmation
(Patterson 1993)) V
(Prakash and Lounsbury 1984)-Model 1 V
(Prakash and Lounsbury 1984)-—Model 2 V
(Prakash and Lounsbury 1984)-Model 3 V
(Prakash and Lounsbury 1984)--Model 4 V
(Prakash and Lounsbury 1984)-—Model 5 V
(Prakash and Lounsbury 1984)-Model 6 V
(Spreng and Olshavsky 1993) V V

(Swan & Trawick 1981)-0V is derived discon.
(Swan 3. Trawick 1981)-0V is subjective discon. 1 (derived)

(T se and Wilton 1988)-Model 4 V V V

83

1.80.2...82 8...... 8.. at

.58... 3.8.33 m. 5:202 8.35 08 8:8

.88... 8.4.8 m. >0.-. 62 .935 8a 8:8

8.38:8... 8.9220 8.. 82%.
22868.6:88. 8.2.9.8 8.. 9.2%.

m .88..-.32 .0382: 8. .88....
m .88.--ng .5853 8a 89......
e 80.2.-.32 .5353 8a .88....
m 82:32 .9853 u... .88....
N 82:82 58...... 8a :88...
F 82:82 .38.... 8. 58.8...

.82 .8288...
8.8.6.28»... 8.88... .28. 5.. 8.09.26
8.8.52.8 0580... 8 88.30 2058 .802

.22 .2: 8a 8.9558

N 3821.82 $0.8m 832.8
F 802.82 ..888 82.2.8

N 382:.82 82.8
F .88....82 ......8

F 82:38. 5829... 82.258
288182 29.803 82.588

m 38.7.52 23...... 8.. £0.80; 63088

N 38.72.02 2.....2 8.. «3.825 .3888
F .ouozitmmF 23.8 8.. 53.08.... 65088

memF cog-am Pa c8093.

 

 

F o o 58 2.003... .93... 0.000.
N c0335... don duo .5383.
F 00.8060. don ....ou 2.9338.
F .80.... .>00 3503a 8058
F N 80.8 .>00 3503a p.068
.88... F 2.0.58.8 3:036 .08
.....o... F 25.8.8.8 8:83» .03. 8o. 03.
F535 F 2.0.52.8 3.003% .03
.....o... F 2.98.2.8 3588 .82 80. .8.
F «00.8.8.8 35030 .03
F 2.0.52.8 8:033 .30. poo. an!
F N 58 88205.. 288..
00:25.58... €225.58...
.0 8.88.8.0... :0 00330098
.8388... .202. 005.58... .802.
. N (52... 3888 .93... as
o 5.... 383...... .8 so...
0 £8 v.8: .... 8...
o 58 8:003... 8.89 ......
0 San 80...:ch 8.08.. 0..
F N .828 500 8080.7. .250 000$
N F .60.“ $00 2835 8a...
F .83... N .628 .>00 800.0030 .80. .000. an!
F .820 .8... N .628 .8 808890 .80. 000. .8.
F N .628 .>0u £00.83“. .80. 000. «no.
E2588 3032a
. N 8.93.8. 5.8.5 295...
Iguana: add. Jag: nag: flag I.. umflufl gala]. llmlxmucou
0028.8 .25 02.0.00...
|§§1

F F 03a...

a

02.3.09... 3.8.3.--c0.ua........00m.o ..0 30000035..

84

The Churchill and Surprenant (1982)-Model 1 (plant) finding that
expectation had a larger impact on disconfirmation than did performance may
be due to the magnitude of manipulations used in that experiment. In the
artificial environment in which the authors had complete control over both
the magnitude of expectation and performance, it is reasonable to believe that
expectations could be manipulated to dominate performance in the
relationship. The absence of similar findings in any field study or in any
other experiment indicates such is probably the case.

On the other hand, Cadotte, Woodruff, and Jenkins (1987) find a
significant causal path from expectation to disconfirmation for all three
restaurant settings they examine (p. 312), though it should be pointed out that
all of the LISREL models they constructed resulted in poor goodness of fit
measures for the entire model.

2) In most models tested, expectations were significantly related to
disconfirmation. Those studies which tested for this relationship but failed to
find it were Oliver (1980), Oliver and Bearden (1983), Bearden and Teel (1983)
and Tse and Wilton (1988). These three studies appear to differ from the others
in two regards. First, they are the only three studies to use path analysis as the
major analytical method. It is unclear why this particular analytical method
should be so strongly associated with this particular result. Second, all three
of these studies used presumably highly-involving products/services as the
study context. It may be possible that for highly involving targets,
expectations and disconfirmation are independent. However, two other
studies, Churchill and Surprenant (1982) and Olshavsky and Miller (1972), also
used durable goods as product contexts. (It should be pointed out that in the

later study, the authors claimed to measure "perceived performance" as the

85

dependent variable, not "disconfirmation.")

One observation is clear: none of the studies which used one-time
services or non—durable goods 23 study contexts failed to find a relationship
between expectjtion and disconfirmation. It may be that for product/ service
contexts in which involvement is likely to be relatively low, expectations
become a significant antecedent of disconfirmation. If involvement is low,
actual performance may be less important relative to expectations in affecting
disconfirmation because there is less perceived variance in performance or
because people care less about performance.

3) Note that desires is never modelew an antecedent of disconfirmation

probably because desires have seldomly appeared at all in satisfaction models
until recently. Desire is included in Spreng (1992), Spreng and Olshavsky

 

(1993) and Spreng, Dixon, and Olshavsky (1993). In all these studies, desire is
modeled as an antecedent of desires congruity rather than expectation
disconfirmation.

4) Attitude is tested in only two models as an antecedent of disconfirmation.
In both models, it is found not to be significantly related to satisfaction.

In this dissertation, expectations will be modeled as an antecedent of
disconfirmation in both models. Perceived performance will be modeled as an
antecedent in one model. Despite its absence in previous satisfaction models,
desires will be modeled as an antecedent of disconfirmation for reasons
discussed in Chapter III. Attitude will not be included as an antecedent of

disconfirmation.

86

Il-4.3 . Antecedents of satisfaction: Expectations, Desires,
Attitudes, Perceived Performance, and Disconfirmation

Tables 12, 13, and 14 illustrates the direct effects of all modeled
antecedents on satisfaction. In Table 12, the first four columns indicate study,
product/ service context, sample composition, and analytical method,
respectively. In Table 13, the first column again refers to the study. The next
three columns indicate the pre-performance antecedents: predictive
expectations, other expectations, and desires. The next column refers to
perceived performance. The next four columns all refer to disconfirmation,
with a separate disconfirmed standard listed for each column. As in previous
tables demonstrating relationships between constructs, numbers in columns
for each study indicate the relative strength of relationship between that
antecedent and satisfaction. A "0" indicates that the relationship between that
antecedent and satisfaction was examined and found to be non-significant. In
Table 14, following the column indicating study, the next two columns indicate
whether the disconfirmation measure used was derived or measured
subjectively. In the last column, other independent variables included in the
model are listed.

The following observations may be made with regard to direct
antecedents of satisfaction.
1) It apgars that perceived performance is consistently found to have the
largest impact on satisfaction. In all but four of the studies which examined
the relationship between perceived performance and satisfaction, perceived
performance had the largest direct effect on satisfaction. The three studies in
which perceived performance had the second strongest impact on satisfaction

were all experiments in which performance was manipulated. The reduced

87

 

 

Table 1 2
Characteristics of Studies Testing Antecedents of
Satisfaction

Study Contgrt Saer Lemod

(Anderson and Sulivan 199 3) multiple Sweden regression
promcts consumers

(Barbeau 1980—Model 1 course eval. students recession
(Barbeau 1980-Model 2 course eval. students recession
(Bearden and Teel 1983)—Model 1 auto repair gen. pop. cov. struct.
(Bearden and Teel 1983)—Model 2 auto repair gen. pop. cov. struct.
(Bitner 1990) travel agent gen. pop. path
(Bolling and Wooduff 1988) wine gen. pop. recession
(Cadotte, Wooctuff, and Jenkins 1987)-Model 1 fast food rest. gen. pop. cov. struct.
(Cadotte, Wooduff, and Jenkins 1987)-Model 2 fast food rest. gen. pop. cov. struct.
(Cadotte, Wooduff, and Jenkins 1987)-Model 3 fast food rest. gen. pop. cov. struct.
(Cardozo 1965)—Model 1 pens students (if. means
(Cardozo 1965)—-Model 2 pens snidents (if. means
(Churdiill and Surprenant 1982)-Model 1 plant gen. pop. cov. struct.
(Churdiil and Surprenant 1982)—Model 2 video player gen. pop. cov. struct.
(Cronin and Taylor 1992)—Model 1 banking gen. pop. cov. struct.
(Cronin and Taylor 1992)-Model 2 pest control gen. pop. cov. struct.
(Cronin and Taylor 1992)—Model 3 dy deaning gen. pop. cov. struct.
(Cronin and Taylor 1993-Model 4 fast food gen. pop. cov. struct.
(Cronin and Taylor 1992)-Model 5 banking gen. pop. cov. struct.
(Cronin and Taylor 1992)—Model 6 pest control gen. pop. cov. struct.
(Cronin and Taylor 1992)-Model 7 try cleaning gen. pop. cov. struct.
(Cronin and Taylor 1992)-Model 8 fast food gen. pop. cov. struct.
(Drbge and Halstead 1991)—Model 1 carpet non-complainers cov. struct.
(Drbge and Halstead 1991)—-Model 2 carpet complainers cov. struct.
(Fisk and Young 1985)-Modd 1 air travel gen. pop. ANOVA
(Fbk and Young 1983—Model 2 air travel gen. pop. ANOVA
(Fisk and Young 1985)—Model 3 air travel student ANOVA
(Fisk and Young 1985)—Model 4 air travel student ANOVA
(Garland and Westbrook 1989) library svc. students cov. struct.

88

Table 12, continued

 

 

Stlfiy Comm Sam e Method
(Giver 1980)—Model 1 flu vaccine gen. pop. path
(Oliver 1980)-Model 2 flu vaccine students path
(Oliver 1993)-Model 1 automobile gen. pop. cov. struct.
(Oliver 1993)-Model 2 class student cov. struct.
(Oliver and Bearden 1983)—Model 1 Get pill users path
(Oliver and Bearden 1989-Model 2 Get piI non-users path
(Oliver and Desarbo 1988) stock returns students ANOVA
(Oiver and L'nda 1981)-Model 1 pajamas gen. pop. - males path
(Oliver and Linda 1981)-Model 2 pajamas gen. pop. - males path
(Oliver and Linda 1981)-Model 3 pajamas gen. pop. - females path
(Oliver and Linda 1981)—Model 4 pajamas gen. pop. - females path
(Oliver and Swan 1989)
(same data as Swan and Oliver. 1991 - see below)
(Olshavsky and filler 1972) tape player students ANOVA
(Patterson 1993) home remnt path

heaters purchasers

(Prakash and Lounsbury 1980—Model 1 fast food rest. snidents correlations
(Prakash and Lounsbury 1980-Model 2 beer students correlations
(Prakash and Lounsbury 1984)-Model 3 fast food rest. students correlations
(Prakash and Lounsbury 1980—Model 4 beer students correlations
(Prakash and Lounsbury 1980-Model 5 fast food rest. students correlations
(Prakash and Lounsbury 1980-Model 6 beer students correlations
(Richins and Bloch 1991 )—Model 1 automobile gen. pothi involv. correlations
(Richins and Bloch 1991)-Model 2 automobile gen. pop./lo involv. correlations
(Richins and Bloch 1990-Model 3 automobile gen. pop/hi involv. correlations
(Ridiins and Bloch 1991)—Model 4 automob'de gen. pop/lo involv. correlations

Table 1 2, continued

89

 

 

Study Contfl Saer Mpthod
(Spreng and Olshavsky 1993) camera students cov. struct.
(Swan and Martin 1981) automobile student regeesion
(Swan and Oliver 1991) automobile new purchasers correlations
(Swan and Trawidr 1981) restaurant gen. pop. regession
(Tse and Wilton 1988)-Model 4 record player student path
(Westbrook 1980)—Model 1 automobiles student regession
(Westbrook 1980)—Model 2 footware student regession
Note: author claimed to measure expectation
but really measured disconfirmation
(Westbrook 1981) stores gen. pop. factor analysis
(Westbrook and Newman 1978) shopping process gen. pop. regression
(Westbrook and Reilly 1983) automobile students cov. struct.

90

o F o 8.3.9.68. e585...» 9... ...5558

P
N
m

 

€29.6me Eases e5 .....osco.

N aging, 88.8.
Faeozlfimm. 2838

 

_ 22811.32 2.58.. 23 £38; .8638
F N 3351.32 2:54 2... £38; .3638
P . 3221.89 2.35% 213583.388.
P o o as... .82 .32.. 9182 :38: e5 9.508
o P o 9.2.. .5 .32.. 3.8»? :88; ...... 9539
P N o :5: .85 62.3182 :58: Ba 96.8.
N P o 9.8.. .88 .62.. $182 :58; use 9.58.
P N o 9.8.. .5 .32.. #182 .258; e5 95.8.
N P o :8: .8... .32.. #182 .2583 Ha 95.8.
N F N wooing. .8» Ba 828$
N P F wooing. .9: us. 8288
N F o N 38.132 8258
F o o P muesli... 388$
N P .82 52...“ 2a 885v
682 :32 «would. a? x .3: mi .314]... :2 jam an an“ 53W
22m a...» 3.6.8... a??? 850 Eden;
63 88a:
SEEN
3%... .u. J33

oocmtoag o>3£oxlc030£m3am Co mucouooeuc<
mp 03a...

91

 

.88 88535.8

.9... e .o N
2
P
F
F
2
p N
P o
5.62 lEbz jun flawglxm 3.5 a... I loulzuqu < '38 g g
9.2.. 82 28.8... 8.83.. .25 .288...
8m 632..
85:85
3 J3

cozczcoo .mp 033.

N 8851.82 82.0.
. 82.1.82 32.0.

58.88 - .82 82.0.
28 - .82 82.0.

. 82 88......

.82 92:82 Ba 8898...

Swap .8953; vs Una—.30.

e 88182 9.8» 9... 8...
m 882.82 9.8» 9... 8....
N 8.182 9.8» 9... 8....
2 821.82 9.8» 9... 8....

.82 288.2.

N .2811. 82 888...: .95 «9.5.
P 802:. 32 988.2. 9... 985.

 

 

a

92

 

.Eoz Ilegll .980. 8.4g
9...... 9.... 288...
“men 630k.

805.0805
3

PFFP

F
_.

NNNN

 

 

lleolllllicngba 80332 2.80 liege xm an.“ x

03.88.. 850 85888.5

3%

08:59.00 .m_. 038..

.82 888.8...

.22 8.....-9... gfifi.

.88.. 8... - .8. 82.0 98 .88 9. 38 8.8.
.82 .82.. 98 82.0.

a 8...... 52 8...: 9... 82.0.
m 8821.62 8.... 98 82.0.
N 8821.62 «9.... 98 82.0.
F 88:... 82 8...: 9... 82.0.

.82 8.88 98 82.0.

~ BqulAmomp 50.8w new L62.0.
— .ouoc‘lammmp cooeaom new 82.0.

N Bootlammmp 83.0.
.. SOIIAmmmP 32.0.

 

 

Ham

93

e .300... 8......

N ”22930 8:8
N "8:93 8......
.. .330... 8.3...
m N
. 0 N
88.88888...
N\_
_\N
N\.
~\_
N .
N .
N 2
N P
m P
m P

 

6.02 6.02 .8. .|_8q leogfixd I 8. . emcee; ll.

9...... a...» 2.8.8.. 9288..
88 88..

80:8..on
1%: .. z

......8. m
......8. m
....8... m
.58... m
N
N

 

 

803...... 8.80 lava xm Idem—'00.... .0
850 2,88...

4%

08:52.00 .m_. 83....

.82 82.0 9... 52w.

.59 ...th 05 Sam.

.82 598.0 9... 988w.

e 882... 82 .8... 9... 8.8....
m 8......82 .8... 98 8.8....
N 821.82 8... 9... 8.8....
p 8.182 .8... 98 2.6....

e $.82 8828.. 9... .89.“...
m 8.21.82 2:88.. 98 28.2..
.. 8821.82 2888.. 98 .898...
m 821.82 289.8. 9... 82......
N 8:...82 8398.. 9... .88...
2 821.82 289.8. 9... .88....

 

63w

94

m
—

 

 

5.02 8.02 8.000 Illlglcozfiu .6

988 25 2.8.9:.
.80 828..
00000096

03332 .0

llll00ca§¢

 

 

.8388.

00:50:00 .mp 030...

to

N

0030?. 0.000 :0. ....filxm lung... xm
0050

02.8.60...

.89 .....8. .28 ..8.£8.s.
0008:3080! 00.0080... 2.80. 50
00030098 0.300... 8 008.20 .05.... .802

N $802 .8983.
. $.00... .8553.

8 09.5188. 8...; 98 at

Camp .930... 0:0 ..ng

 

063$

95

> > \z

> > >

\z > >

2850202. > > >

0:003:58 9.5302828 > >

> \z >

> > >

> > >

> > >

 

 

 

......z lqlg ...0 02 null» .8280 2.80.4.0. 0
28.. .0 ... 88.0880 8..

 

0.). .050

0_000_2 :0503033 :_

an< .30 Ed
ZO.hU<u_m-h<m 20.._.<1¢EZOUm .0

88.. 08......N02 28.8.93. 98 .....o.=..0.
988:.N02 2.8.88.3 98 .....o.....0.

m 83:...32 2.22 98 .8208; .8088.
N 89.2.32 28.8. 98 0508.... .8388.
. 89.2.82 2.22 98 5:88; .3088.

.002 0.58.... 98 98......

.82 85....

N $.82 .2: 98 .828...
. 09.2...82 83 98 .828...

N 32152 8388
2822.....02 838...

.82 82.80 98 80805...

 

203m.

0003 00.3080: 33080.03 0:8 :0_um:.._._:000_o .0 0030030226

¢ _. 030...

96

3.... 2.0.2.3...
.8... 9808....

...:o 22.8.... .302 Jam

.....0 2.0.2.0.... A- .8

2.8.... 8.28
2.8.... 8.28
2.8.... 8.28
2.8.... 8.28
.28.... 8.28
2.8.... 8.28
2.8.... 8.28
2.8.... 8.28

 

A8.2 .050

> > >
> > >
> >
> >
~. ~ .... ...
w .... .... ~
~ ~ ~ ~
~ ~. ~ N..
> >
> >
> >
> >
> >
> >
> >
> >

 

 

2... I199... 02 In.» 9.2.8 In...>_8...m..|m
88.. ... ... 88.9.8 8..

 

Hm? 5w am
ZOFU<um§<m ZOF<2¢EZOU£O

80:50:00 .8» 838...

N Elana» 82.0.
— Unoiomm» 83.0.

.52 .6882.

8mm» b.2352 0.8 2.038...

.82 .8582. 98 98.80.

.. 8002.82 9.3» 98 .8...
8 882.82 0......» 98 .8...
N 882.82 0......» 98 .8...
. 89.2.82 0......» 98 .8...

N 89.2. 82 9888: 98 8......
P 89.2. 32 0888.. 98 80.0.

m 89.2.82
» 89.2.82
0 882.82
m 89.2.82
.. 89.2.32
m 89.2.32
N 89.2.2.2
» 89.2.82

.98» 98 98.0.
.92.» 98 98.0.
.98» 98 2.8.0.
9....» 98 2.8.0.
.98» 98 98.0.
.9...» 98 98.0.
9....» 98 98.0.
.98» 98 9.8.0.

 

.635

97

4.88.038 3.5.5.0 ”200:0 02.002.83.00.
88.3.8 8.5.5... .38... 8.8.2.8580.

>>>>

> >>>>>>

550020.305... .. . ...

>
>
\z
>

>7» >>

 

Ea: 065w m2 00>
053. .0 9 «002055 00:

 

> >>>>>> >>>>

w

 

0.? .050 20....U<n_m.._.<m

>>>>

>>>>>>

00.93.20...

>> >>

a

ZOF(!¢IZOUw.o

80:50:00 .3. 030..

8 882... S2 .89.. 98 8.8....
m 89.2.82 .89.. 98 8.6....
N 89.2... 82 .89.. 98 8.6....
. 882. 82 .890 98 9....9...

0 89.2.82
... 89.2.52
8 89.2.52
8 89.2.52
N 80.2.82
. 89.2.9.2

238.9... 98 .898...
299.9... 98 .898...
298...... 98 .898...
298...... 98 .898...
298...... 98 89.9..
239...... 98 .898...

.82 8.8.8..

.320... 000 - Pam» £025 new 22.5 an 38 OER...

3mm» 2026 9.0 .025.

3mm» 00.0000 9.0 82.0.

......88.82 .8980 98 82.0.
8.8.9.2 .8980 98 82.0.

N BUOZIAmmmP $2.0.
_. BOOIIAmmmF .025.

 

.635

98

2 8...: ...
.8 8:... 0:0 :ucuucoomn... 8.. 820.380.... ".8... .58....
0> 803.50 03:22.80 0... 8.3. 2,202.0 83.2....

2.5...ch 8.680 62.80

> >
> >
> > >
> > >
> > >
> > >
> >
> > >
> > >

 

 

 

.....2 Imlg ...... .z in... 8...... Illfll.>..... ...
88.... .. 88.9.... 8:

 

m.). .050

mar—.0 = 6 Eu
20.5(uw_h<w 20_h<1¢.u200w.o

88:52.00 .8. 830...

3.2 2.... .98 .8958)...

8.2 8......82 98 8.5.83.

.82 8.5.83.

5005.....886 08.308... 2...... 50
503080.... 2:808... 8 02...... .05.... .302
N 88.2.82 8.8.83.

9 8002.000. 30.533.

¢ 80021862 002.? 0:0 on...

:82 8......» 98 .83....
2.2 82.0 98 .83..

.52 ....82 98 .83....

.82 28.8.8.0 98 9.8....

...2 .....8.

 

53w

99

impact of perceived performance in these instances may have been due to
either 1) an artificially attenuated range of actual performance (making other
variables relatively more important) or 2) a lower than usual level of
involvement among respondents leading them to care less about actual
performance.

Note that in none of the studies was perceived performance found to
have a nonsignificant impact on satisfaction, except for the low involvement,
focal brand treatment of Bolfing and Woodruff (1988). However, also note that
in every study except one the objective level of performance is difficult or
impossible to assess. In other words, measuring perceived performance is a
highly subjective endeavor as the scale anchors often imply (see II-S). It
seems logical, therefore, that the subjective evaluation of performance and
satisfaction with performance should be highly related. One definition of
satisfaction as "....an emotional response to the experience provided by, or
associated with, particular products or services..." (Westbrook and Reilly 1983,
p. 256) clearly illustrates why the two should be closely linked.

Oliver and DeSarbo (1988) was the one study for which performance
could be unambiguously assessed. In that study, performance of a stock
investment was stated in terms of percentage return on investment; the
performance level was manipulated in the experiment, but not measured.

2) The second most influential direct antecedent of satisfaction appears to
be disconfirmation. For those models in which perceived performance was not
included, some form of disconfirmation was rated as the first or second most
influential variable. For those models which did include perceived
performance, expectations disconfirmation was still found to be significant in

all studies except for Churchill and Surprenant ( l982)-—Model 2 (video player),

100

and Spreng and Olshavsky (1993). However, in both Churchill and Surprenant
(1982)—Model 1 and Oliver and DeSarbo (1988), disconfirmation was found to
have a stronger relationship with satisfaction than perceived performance
had. Some studies tested multiple standards of disconfirmation and found at
least one to be significant (e.g., Barbeau 1984).

Churchill and Surprenant (1982) suggested that the reason for finding
no disconfirmation - satisfaction relationship in their Model 2 may be because
disconfirmation is less important in predicting satisfaction of durable goods.
However, numerous studies involving automobiles (Oliver 1993; Richins and
Bloch 1991; Swan and Martin 1981; Westbrook 1980; Westbrook and Reilly
1983) and other durable goods (e.g., Droge and Halstead 1991) found the
relationship to be significant.

It does not appear to matter which disconfirmation standard is used.
Disconfirmation of predictive expectations, desires, product type and best
brand norms all proved to be significant antecedents of satisfaction.

In summary, with few exceptions (e.g., Barbeau 1984, Spreng and
Olshavsky 1993), it appears that the empirical evidence in support of

disconfirmation as an antecedent of satisfaction is quite strong, regardless of

 

the product/ service context, sample composition, or standard used.

It is important to note that for several studies, examining the
disconfirmation-satisfaction relationship is of secondary importance. For
example, in Oliver and Swan (1989), the primary focus is on the importance of
perceived equity and fairness on exchange satisfaction. Thus the study is
designed to highlight the effects of these variables. Likewise, Bolfing and
Woodruff (1988) focus on the situational context of product usage, and Brown

and Swartz (1989) focus on dyadic relationships and "gaps" between expected

101

and perceived performance using a 65 item instrument. In these studies, too,
disconfirmation is found to compete with or exceed the explanatory power of
the primary study variable.

Interestingly, in both studies reviewed which do not examine
satisfaction as the primary dependent variable (Anderson 1973; Olshavsky and
Miller 1972), disconfirmation is shown to significantly effect the perceived
performance or perceived quality of the product being evaluated.

3) The role of expectaiions in saflfgction formation has received extensive
attention in the literature. Empirical support has been found for direct effects
and indirect effects. via disconfirmation and perceived performance. The
indirect effects of expectations on satisfaction are especially complex as each
route (via disconfirmation and perceived performance) are thought to have
opposing impacts on satisfaction. As Yi (1990) succinctly states, "Thus, raising
consumer expectations may yield both an increase (due to the increase in
perceived product performance) and a decrease in CS (due to the increase in
disconfirmation)" (p. 79). .

Predictive expectation was frequently modeled as an antecedent of
satisfaction. At times it was found to have no direct effect on satisfaction,
though it usually was found to have at least a moderate direct positive impact
on satisfaction. In two studies, Swan and Martin (1981) and Bearden and Teel
(1983), it was found to be the antecedent most strongly related to satisfaction.
In one study, Droge and Halstead (1991), retrieved expectations was found to be
negatively related to satisfaction.

There is no apparent pattern in these results. Among the studies which
found expectations to be significant, some involved durable goods and others

involved non-durable goods or services. Some of these models included

102

perceived performance as an antecedent, and others did not. In some,
expectations "outperformed" disconfirmation, while in others the reverse
occurred. No patterns based on how expectations were measured are apparent
either (see Table 3 and Table 4).

On the other hand, all the studies which found expectations not to be an
antecedent of satisfaction involved either a durable product or an ongoing
service with the exception of the Bolfing and Woodruff (1988) study involving
wine. In several studies involving durables or ongoing services, expectations
was a significant antecedent. In two cases (Drbge and Halstead 1991; Oliver
1980) the same study found expectations to be significant in one model but not
in another model. One thing is clear: in no studies involving non-durable
goods or short-lived services has expectations been found to be unrelated to
satisfaction, except for Bolfing and Woodruff (1988).

In addition, five of the eight models in which expectations was found
not to be a significant antecedent occurred in field studies, the exceptions
being Churchill and Surprenant (l982)-—Model 2 (video player), Spreng and
Olshavsky (1993), and Bolfing and Woodruff (1988).

In short, when non—durables and short-duration services provide the
context, expectations are always directly related to satisfaction. When the
context is durable goods or ongoing services, sometimes expectations is
unrelated to satisfaction and sometimes it is not. There are two obvious
questions raised. First, what differences between durable versus non-durable
goods and between ongoing versus short-lived services may be relevant to the
role of expectations in satisfaction formation? Second, under what
circumstances in durable good/ ongoing service contexts does the

expectations— satisfaction relationship become important?

103

To summarize the important direct antecedents of satisfaction are

 

perceived performance, disconfirmation and expectations. Attitude has been
found to be significant in one model; desires is found to be directly related to
satisfaction in one model, and its use as a disconfirmation standard has proven
to be significant.

In this disserta_ti_g1_, all major antecedents will appear in at least one of
the two models to be used for testing the hypotheses developed in Chapter III.
The major deviation from the traditional satisfaction literature is the omission
of perceived performance as an antecedent in one of the models; the reasons
for this unique specification is that the objective measure of performance is
universally known and accepted, thus obviating the need for a measure of

perceived performance.

II-S . Other relevant topics

In addition to the constructs and relationships comprising the
traditional satisfaction formation model, additional factors regarding model
complexity are discussed next. It is precisely because these complexities have
not been well integrated into the traditional model that the research
hypotheses offered in this dissertation can make a contribution to the
literature. The complexities fall into two categories and are discussed below,

structural (Section II-5.l) and process-related (Section II-S.2).

11-5. 1 . Model complexity -- Structural
Several researchers have noted that current models of satisfaction
formation are structurally simplistic. They generally conclude that additional

constructs are required to adequately model satisfaction formation. Such

104

observations usually result from a combination of theoretical and empirical
evidence linking other variables to the traditional model. The following

sections discuss some of these crucial issues.

Need for additional complexig in the dependent variable

As indicated in Chapter II-2 above, different targets of satisfaction have
been successfully used in satisfaction models. For example, Churchill and
Surprenant (1982) measured satisfaction with the product itself, Cadotte,
Woodruff and Jenkins (1987) measured satisfaction with the consumption
experience, and Oliver (1980) measured satisfaction with the purchase decision
itself. Yi (1990) enumerates a complete list of targets. In their review of
satisfaction literature, Peterson and Wilson (1992) summarize the situation as
follows: "Without reiterating the specific findings reported above, it would
appear that satisfaction ratings observed to date likely represent more than
satisfaction with the product or service being studied" (p. 68). Indeed, in
Chapter I, it was demonstrated that cognitive dissonance theory (Festinger
1957) and regret theory (Loomes and Sugden 1981) support the notion that
both choice and nonchoice alternatives are relevant satisfaction targets.

In short, using a single, specific target of satisfaction appears to be
overly simplistic given the realities of consumer decision-making within a
competitive environment. Rather, satisfaction with each of multiple specific
targets (both choice and nonchoice alternatives) and overall satisfaction, a
summary evaluation or judgment of the relevant consumption experience,
appear to be necessary to model satisfaction formation processes in a realistic

manner.

105

In this dissertation satisfaction with multiple specific targets as well as

 

overall satisfaction are included.

Need for giditional complexity in antecedent variables

Numerous potential antecedent variables have been found to be related
to satisfaction other than those presented here as comprising the traditional
model. Some have been studied in conjunction with traditional antecedents,
such as those listed in the final columns of Table 14 (e.g., "attributions" in
Bitner 1990, "service quality" in Cronin and Taylor 1992, etc.). Peterson and
Wilson (1992) review literature indicating that there are weak relationships
between demographic variables and satisfaction, there are moderate but
inconsistent relationships between size of choice set and satisfaction, and
significant relationships between satisfaction and "life satisfaction."

Other researchers have mentioned "non-traditional" antecedent
possibilities. For example, Swan and Trawick (1981) state " we feel strongly
that the observation of what other people received is part of the satisfaction
process ...." (p. 56). (Despite this clear indication that what B experiences has
an effect on A's satisfaction, the authors did not explicitly model the impact or
in any way pursue the possibility of such a relationship.)

In this dissertation recognizing that other antecedent variables may be

 

relevant, traditional antecedents of alternatives not selected ("nonchoice")
will be included as possible antecedents of satisfaction formation. Thus, in this
dissertation, the satisfaction model will be enhanced using additional

antecedent constructs and additional satisfaction "targets."

106

lI-S.2 Model complexity -- Processes

With few exceptions, it has been assumed that satisfaction processing is
uniform across consumers. In other words, researchers have been searching
for the satisfaction formation model without explicitly considering that
satisfaction formation processes may differ systematically among groups of
consumers. Duhaime (1988) and Kennedy and Thirkell (1988) are two
exceptions to this generalization. In addition, Oliver and DeSarbo (1988)
mention that consumers may differ with regard to the relative importance of
certain antecedent variables. Two other exceptions to the assumption of
uniform processing are researchers who find satisfaction processing varies
by level of involvement and those who find satisfaction processing varies by

level of ultimate satisfaction.

Involvement

Richins and Bloch (1991) hypothesize and find differential decay rates
in satisfaction among high versus low involvement automobile owners. They
find that among low involvement owners, satisfaction increases over time,
while the opposite occurs among high involvement owners. They did not
gather pre-purchase data and thus do not analyze pre-experience antecedents
such as expectations. However, their-logic, that highly involved owners
process more intensively. would also be expected to hold for pre-experience
variables.

In addition, Westbrook and Newman (1978) concluded that dissatisfied
consumers may become more highly involved because they are dissatisfied.

This study is discussed in the following section.

107

Level of Satisfaction/ Dissatisfaction

In the satisfaction literature, a few researchers have recognized that
consumers who are dissatisfied may process differently than consumers who
are satisfied. Westbrook and Newman (1978) found that consumers who are
dissatisfied with previous appliance purchases are more likely to be
dissatisfied with future appliance purchases. They conclude that some of this
tendency may be due to individual personality factors. However, they also
conclude that consumers who have been dissatisfied in the past may be more
highly involved in a subsequent purchase than are previously satisfied
consumers (who have routinized the previously successful purchase process).
Significantly, this higher involvement may serve " to amplify sensitivity to
less favorable experiences in shopping and choosing" (p. 464).

As detailed above in II-2.1, Maddox (1981) also found that dissatisfied
consumers appear to process more intensively than do satisfied consumers.
Specifically, he finds that dissatisfied consumers are more likely than satisfied
consumers to focus on "instrumental" factors, i.e., the Mg of their negative
experience.

Droge and Halstead (1991) find strong empirical evidence of differential
processing among consumers who are dissatisfied ("complainers") versus
those who do not complain about their newly purchased carpets. Specifically,
they find that complainers are more likely to be expectations-influenced
while non-complainers are found to be more disconfirmations-influenced. In
fact, they hypothesize that by assuming satisfaction processes are not
differentiated, " past research may have masked differences in the
fundamental interrelationships among these key constructs, leading to

contradictory results across studies" (p. 319). Droge and Halstead (1991)

108

hypothesize that complainers are more expectations-driven because of
significantly greater cognitive processing than non-complainers, resulting
in suspension of the normal "decay of expectations" (Oliver 1981).

In addition, the theory of cognitive dissonance (Festinger 1957) may
once again be relevant. If dissatisfaction results because performance does
not live up to desires or expectations, cognitive dissonance results. Dissonance
always results in some level of psychological discomfort which the subject
tries to reduce. Satisfied consumers experience no dissonance and thus are not
driven to dwell on relationships between constructs. On the other hand,
dissatisfied consumers attempt to reduce the dissonance by reexamining each
construct in each dissonant relationship, looking for a way to reduce the
dissonance. Again, the increased processing of dissatisfied consumers may
result in stronger paths from preperformance to postperformance constructs.

In this dissertatiin, dissatisfied subjects are hypothesized to process
more extensively than are satisfied subjects, because dissatisfied subjects feel
the need to reconcile their pre-performance expectations, desires, and

attitudes with the election outcome and their level of satisfaction.

II-6 Summary of literature review

The following is a summary of the major points presented for each
concept and each relationship utilized by the traditional satisfaction model.
Constructs
0 Satisfaction

- The studies reviewed exhibited wide variation in conceptualization.

- The studies reviewed exhibited wide variation in scale wording, but

109

usually included the word "satisfaction" or a variation of that
word,

- Satisfaction was generally measured as a unidimensional construct,
often using a single item,

- The studies reviewed indicated that satisfaction has been

operationalized using a variety of "targets."

0 Expectation

Most of the studies reviewed included expectations as a potential

antecedent variable.

Most studies measured expectations prior to product/ service

exposure, but three did not.

There are three kinds of "targets" of expectations: objective
performance, overtly subjective evaluations of performance,

level of satisfaction expected.

Even commonly used measures of expectations regarding objective

performance are flawed.

0 Desire
- Desire has only recently become the focus of serious attention in
the satisfaction literature
- In at least three ways "desires" has indirectly entered the
satisfaction literature: as a disconfirmation standard rather
than as an independent construct, manipulated as the "ideal"

level of an attribute, and through evaluative expectations.

110

- In the service satisfaction literature, the construct labeled
"expectation" usually is measured as desires.

— Desires is conceptualized sometimes as a "more is better"
continuum and sometimes as an "ideal point" continuum.

- When scales are evaluative, it is difficult to interpret

responses logically.

0 Attitude

- Attitude has been used as both an antecedent and consequent
construct in satisfaction formation models.

- Only three studies have utilized attitude as an antecedent variable in
satisfaction formation models.

- Attitudes are hypothesized to affect expectations and perceived
performance judgments; this is thought to be especially true
among low involvement consumers, because low involvement
consumers are thought to be less attentive to actual
performance and thus rely more on pre—performance factors

in forming satisfaction.

0 Perceived performance
- Operationalizations of perceived performance sometimes appear to
be confounded with other constructs.
- In five of the studies reviewed, performance was manipulated; the
motivation for manipulating performance was not only to
manipulate perceived performance, but also as a way to

manipulate disconfirmation.

111

- The majority of studies measured attribute-specific performance or

both attribute-specific and overall performance.

0 Disconfirmation

All operationalizations can be categorized along two dimensions: 1)
subjective versus derived (computational) measures, and 2)

standard which is being disconfirmed.

Numerous standards have been proposed and tested.

Studies have found support for cognitive dissonance explanations

and for the assimilation explanations of disconfirmation.

Predictive expectation has been the most commonly used standard
for the operationalization of disconfirmation.
- Subjective measures have generally been found to be more strongly

related to satisfaction than are derived measures.

Relationships
0 Antecedents of perceived performance

- Studies which examined the expectations-perceived performance
relationship found the relationship to be significant; the
evidence linking expectations and performance is strong.

- Only a few studies examined the effects of expectations on
performance relative to the effects of other potential
antecedents on performance.

- It is difficult to make any generalizations about the desires -
performance relationship given that it has been tested so

infrequently in the literature.

112

0 Antecedents of disconfirmation

— With one exception, whenever expectations and perceived
performance appeared in a model together, the relationship
between expectations and disconfirmation was weaker than the
relationship between perceived performance and
disconfirmation.

- In most models tested, expectations were significantly related to
disconfirmation. Those studies which tested for this
relationship but failed to find it used presumably highly-
involving products/ services as the study context. (None of the
studies which used one-time services or non-durable goods as

study contexts failed to find a relationship between expectation

 

and disconfirmation.)
- Desires was never modeled as an antecedent of disconfirmation, and
attitude was only tested in two models as an antecedent of

disconfirmation.

0 Antecedents of satisfaction
- It appears that perceived performance was consistently found to
have the largest impact on satisfaction. In none of the studies
was perceived performance found not to have a significant
impact on satisfaction.
- The second most influential direct antecedent of satisfaction
appears to be disconfirmation, regardless of the

product/ service context, sample composition, or standard used.

113

- Empirical support has been found for direct effects and indirect
effects (via disconfirmation and perceived performance) of
expectations on satisfaction.

- When non-durables and short-duration services provide the
context, expectations are always directly related to satisfaction.
When the context is durable goods or ongoing services,
sometimes expectations is related to satisfaction and

sometimes it is not.

The following chapter develops the research hypotheses of the
dissertation and makes frequent reference to the material covered in this

chapter.

Chapter 111
Key Theoretical Issues Addressed

III- 1 . Introduction

The purpose of this chapter is to develop the research hypotheses and to
link the research hypotheses with testable hypotheses. Recall that all
proposed enhancements to satisfaction theory presented in this thesis are
developed and tested using data collected regarding the 1992 U.S. presidential
election. Because most hypothesis testing will be conducted within the context
of two models, these models are briefly presented first. Next, each of the four
research themes and the associated hypotheses are described. Finally,
comprehensive overviews of each model are presented. Specific research

methods are described in Chapter IV.

Election Outcome Satisfaction Model -- Overview

The Election Outcome Satisfaction Model (EOSM) describes the
relationship between satisfaction with the election outcome and its
antecedents. Expectations, desires, and attitude towards the candidate were
measured before the election; disconfirmation and satisfaction were measured
following the election. Perceived performance was not included because

actual performance was unambiguous and widely known (see Chapter II-4.4).

Campaign Process Satisfaction Model -- Overview
The Campaign Process Satisfaction Model (CPSM) describes the
relationship between satisfaction with the campaign process and its

antecedents. Two elements of the campaign process are included: "advertising

114

115

effectiveness" and "running an unfair campaign." These were judged to be
two elements (1) which people would recognize as being relevant to the
campaign process, and (2) about which people would form expectations and
perceptions of performance. Measures of expectations were collected early in
the campaign and measures of perceived performance, disconfirmation, and

satisfaction with the process were gathered following the election.

III-2. Thesis Themes, Research Hypotheses, and Testable
Hypotheses
Each of the thesis themes described in Chapter I corresponds to
research hypotheses of importance to the satisfaction literature. First, each
research hypothesis is described. Next, it is contrasted with the relevant
hypothesis derived from the traditional satisfaction model. Finally, testable

hypotheses within the framework of the CPSM and/or the EOSM are presented.

III-2.1. Theme No. 1: Multiple Targets

Marketing scholars appear to be reaching consensus that the
appropriate phenomena of study are exchange relationships, consumption
experiences, integrated strategies, etc.; that is, broadly defined phenomena
(Hunt 1990; Sheth, Gardner, and Garrett 1988). Likewise, satisfaction scholars
appear to be moving toward broadly defined targets of satisfaction (see
Chapter II-Z). The traditional model of satisfaction formation itself appears to
be in a transition from focusing on a single, specific target of satisfaction
(Figure 4-a) to focusing on an overall target (Figure 4-b). As described in
Chapter II-2, some authors have included both targets (Figure 4—c). A few

studies have even included satisfaction with specific attributes of the

116

Satisfaction with
Specific Target

Overall
Satisfaction

 

 

 
    
 

  

Satisfaction with
Specific Target

  

Overall
Satisfaction

  
 
 
   

 

(C)

  

Satisfaction w/
Attribute

  
 

Satisfaction with
Specific Target

    
 
 

 
 

  

Satisfaction w/
Attribute

 
 

(d)

Figure 4
Targets of Satisfaction: Variations in the
Traditional Model

The tradtiond model most commonly utilizes a specific target of satisfaction formation
processes (a); examples of other targets include overall satisfaction (b), both specific
and overall targets (c), and multiple attributes of a single specific target and an overall

target (d).

117

target as antecedents of overall satisfaction (Figure 4-d).

Given this trend, and the conceptualization of satisfaction formation as
being intimately related to consumer decision-making (Chapter I), it seems
logical to expand the concept "target of satisfaction" to include both overall
satisfaction and multiple targets, where the multiple targets include
satisfaction with a choice selected (hereafter, "choice") and satisfaction with
alternative members of the choice set which were not selected (hereafter,

" nonchoice" ).

Figure 5 illustrates the research hypothesis in its most basic form:
satisfaction with the alternative chosen and satisfaction with an alternative
not chosen each contribute independently to overall satisfaction. In other
words, an individual experiences some level of satisfaction/dissatisfaction with
alternatives not selected and this, too, contributes to overall satisfaction. Not
only is this hypothesis a logical extension of trends already evident in the
satisfaction literature, it also has theoretical support from decision theory and
reflects common experiences (e.g., see example in Chapter I).

This research hypothesis recognizes explicitly that satisfaction
formation is a complex process which occurs in a competitive environment.
In such an environment, the act of selecting one alternative does not
immediately and absolutely remove the alternatives not selected from one's
consciousness. It is hypothesized here that ultimately, in some sense, one is
satisfied or dissatisfied with nonchoices, alternatives not selected, and that
satisfaction/dissatisfaction with these other targets affects overall satisfaction.

It should be noted that although Figure 5 appears to be structurally

similar to Figure 4-d, the two are fundamentally distinct. Figure 4—d

118

 
    
 

Satisfaction with
Alternative Chosen
("CHOICE")

 
   
   
 

Overall

   

Satisfaction

  
 

Satisfaction with

  
 

Alternative Not Chosen

  

("NONCHOICE")

Figure 5
Hypothesized Relationships: Satisfaction with Choice,
Satisfaction with Nonchoice, and Overall Satisfaction

Satisfaction with two specific targets, "choice" and "nonchoice" altematives, are hypothesized to
contribute independently to overall satisfaction.

119

implies that overall satisfaction is somehow the summation of satisfactions
with individual attributes of a single target. Figure 5 illustrates that

overall satisfaction is the result, at least in part, of satisfaction with different
targets.

In the Election Outcome Satisfaction Model, one target of satisfaction is
Clinton's performance on election day and another target of satisfaction is
Bush's performance. Although an individual could only select one alternative
on voting day, it is hypothesized here that satisfaction with the performance
of each cfiandidate affects an individuals overaL satisfaction with the election
outcome regardless of which candidate that individual supported. Figure 6
depicts this research hypothesis among both Clinton and Bush voters.
Traditional satisfaction models would predict that satisfaction with the
performance of one's 932p candidate (i.e., the candidate voted for) is the only
relevant target of satisfaction.

Specific testable research hypotheses are the following.

Hla: Among Clinton voters, paths from satisfaction with Clinton's
performance to overall satisfaction and from satisfaction with Bush's
performance to overall satisfaction will each be positive.

Hlb: Among Bush voters, paths from satisfaction with Bush's performance to
overall satisfaction and from satisfaction with Clinton's performance to
overall satisfaction will each be positive.

An alternative conceptual model for investigating the multiple target
hypothesis is developed if we omit the satisfaction with target A and
satisfaction with target B and focus on the direct effects of traditional -
antecedent variables on overall satisfaction. Again, as depicted in Figure 4-b,

the traditional model specifies that an antecedent variable regarding a

120

    
   
   
    
     
 

Satisfaction with
Clinton's Performance

Overall

Satisfaction

Satisfaction with

Bush's Performance

Figure 6
Election Outcome Satisfaction Model:
Theme No. 1 Research Hypothesis

Satisfaction with both Clinton's performance a_id Bush's performance affects overall satis-
faction, regardless of which candidate an individual supported.

121

single target affects overall satisfaction. As described in Chapter II, such
antecedent variables often include expectations, perceived performance, and
disconfirmation.

The multiple target research hypothesis advanced here indicates that
the antecedent variables related to multiple targets affects overall satisfaction
as indicated in Figure 7. Specifically, perceived performance and
disconfirmed expectations of multiple targets affect overall satisfaction.
Again, the implication is that these variables regarding alternative choices
contribute independently to the satisfaction formation process.

The Campaign Process Satisfaction Model provides a forum for testing
this hypothesis. Perceived performance and disconfirmed expectations
regarding advertising effectiveness and campaign unfairness of both
candidates is hypothesized to affect overall satisfaction with the campaign
process. Obviously, the performance of both candidates is relevant to
satisfaction formation for supporters of either candidate.

The specific paths of interest for testing the hypothesis of multiple
targets are illustrated in Figure 8. Previous research has established that
disconfirmed expectations and perceived performance each affect satisfaction
(Chapter II-4.4). Therefore, both links are hypothesized for both campaign
elements and for both candidates. Specific testable hypotheses include the
following:

ch: Both paths from Clinton's perceived performance to overall
satisfaction are positive and both paths from Bush's perceived performance to
overall satisfaction are negative for Clinton supporters.

Hld: Both paths from Clinton's disconfirmed expectations to

disconfirmation are positive and both paths from Bush's disconfirmed

122

   
     
 
     
   
     

Antecedents
related to
Target A

  

Overall
Satisfaction

Antecedents
related to
Target 8

Figure 7
Hypothesized Effects of Satisfaction Antecedents
Regarding Separate Targets on Overall Satisfaction

lt is hypothesized that antecedents regarding alternative targets contribute
independently to overall satisfaction.

123

 

 

 

 

 

 

 

 

 

 

Figure 8
Hypothesized Paths in the Campaign Process Satisfaction Model for
Testing the Multiple Targets Research Hypothesis

Expectations, perceived performance and disconfirmation for campaign elements relative to
each candidate are hypothesized to affect overall satisfaction, for both Clinton supporters
and Bush supporters.

124

expectations to disconfirmation are negative for Clinton supporters.

Hle: Both paths from Clinton's perceived performance to overall
satisfaction are negative and both paths from Bush's perceived performance to
overall satisfaction are positive for Bush supporters.

Hlf: Both paths from Clinton's disconfirmed expectations to overall
satisfaction are negative and both paths from Bush's disconfirmed
expectations to overall satisfaction are positive for Bush supporters..

If support for the multiple target hypothesis is found in both the EOSM
and the CPSM, it can be argued that such support is especially strong

considering the significant differences in the structure of the two models.

III-2.2 Theme No. 2: Parallel Paths with Cross-Over Effects

If satisfaction with the choice and satisfaction with the nonchoice both
contribute to overall satisfaction, it is reasonable to consider the processes
which lead to satisfaction with each alternative. In addition to assuming a
single target of satisfaction, the traditional model literature posits that all
relevant antecedents of satisfaction have a single target (though possibly
consisting of multiple relevant attributes). The most commonly referred to
antecedent of satisfaction is disconfirmation (see Chapter II—3.S). The
traditional model, then, would suggest that if one wanted to describe the
antecedents of satisfaction with each of two potentially related targets, one
would need two independent paths (see Figure 9).

The traditional satisfaction literature has investigated in great detail the
impact of external standards on the satisfaction process (Chapter II-3.S), but
such standards are assumed to be contained within the concept of

disconfirmation. Thus, when Cadotte, Woodruff and Jenkins (1987) talk about

125

Satisfaction with

D’ f' t'
Iscon Irma Ion TargetA

Disconfirmation Satisfaction With
Target B

 

Figure 9
Traditional Model: Different Targets Require
Parallel and Unrelated Paths

126

"best brand norms" or when Tse and Wilton (1988) talk about "ideal standards,"
they are referring to a set standard against which the person judges the
target's performance. While such models recognize that other targets have
relevance to the satisfaction process, they are assumed to have relevance at a
single, isolated step in the process.

In this thesis, processes relevant to satisfaction formation with each of
two competing alternatives are hypothesized to be related structurally.
Specifically, as illustrated in Figure 10, disconfirmation of expectations with

regard to the choice alternative is hymthesized to affect satisfaction with the

nonchoice alternative the reverse is also hypothesized to hold. In other

 

words, the parallel paths of satisfaction formation processing are expected to
exhibit cross-over effects.

The logic for hypothesizing these cross-over effects is straightforward.
As others have argued theoretically and demonstrated empirically, non-target
standards are relevant in the disconfirmation - satisfaction link. In the
special case of directly competing alternatives, the judgment that "target A
performed better/worse than expected" affects a person's satisfaction with
target B is clear. Judgment of target A's performance as being "better/worse
than expected" cannot be isolated from the reality of the expectations or
performance of target B, target A's alternative. In the big football game, that
the opposing team performed better than expected obviously affects
satisfaction with the home team's performance.

Among closely competing alternatives, the term "better than expected"
refers not only to the expectations regarding A's performance, but also to the
reality that expectations regarding A's performance were formed in

conjunction with expectations regarding B's performance. Similarly,

127

 
     
   
 

Overall
Satisfaction

Antecedents Satisfaction with
. TargetA

Antecedents D isconfirmation Satisfaction With

Target 8

 

Figure 10
Hypothesized Model: Parallel Paths and Cross-over Effects

Processes relevant to satisfaction formation with each of two competing alternatives are
hypothesized to be related structurally as indicated.

128

judgment of A's performance is made in conjunction with judgments
regarding B's performance. So it is logical that disconfirmed expectations
regarding A's (B's) performance affects satisfaction with B (A).

Note that the critical point of this hypothesis is that the processes
themselves are entwined, not just the individual constructs.

In the Election Outcome Satisfaction Model, disconfirmed expectations
regarding Clinton's performance are hypothesized to affect satisfaction with
Clinton's performance (in accord with the traditional model) and satisfaction
with Bush's performance (due to the hypothesized cross-over effects).
Likewise, disconfirmation of Bush's performance is hypothesized to affect
satisfaction with Bush's performance and satisfaction with Clinton's
performance.

Specific testable research hypotheses, therefore, are the following:
HZa-i: The path from disconfirmed expectations regarding Clinton's
performance to satisfaction with Clinton's performance will be positive among
Clinton supporters.

HZa-ii: The path from disconfirmed expectations regarding Bush's
performance to satisfaction with Clinton's performance will be negative
among Clinton supporters.

HZb-i: The path from disconfirmed expectations regarding Bush's
performance to satisfaction with Bush's performance will be positive among
Clinton supporters.

HZb-ii: The path from disconfirmed expectations regarding Clinton's
performance to satisfaction with Bush's performance will be positive among
Clinton supporters.

H2c-i: The path from disconfirmed expectations regarding Bush's

129

performance to satisfaction with Bush's performance will be positive among
Bush supporters.

H2c-ii: The path from disconfirmed expectations regarding Clinton's
performance to satisfaction with Bush's performance will be negative among
Bush supporters.

H2d—i: The path from disconfirmed expectations regarding Bush's
performance to satisfaction with Clinton's performance will be positive among
Bush supporters.

H2d—ii: The path from disconfirmed expectations regarding Clinton's
performance to satisfaction with Clinton's performance will be negative

among Bush supporters.

III—2 .3 . Theme No. 3: Differential Processing Due to Ultimate
Level of Satisfaction

The traditional satisfaction model does not differentiate between
satisfied and dissatisfied consumers with regard to the satisfaction formation
process each follows. Stated another way, consumers are presumed to have
arrived at their satisfied/ dissatisfied state via the same processing paths
regardless of the ultimate level of satisfaction: the strength of the relations
between constructs is hypothesized to be the same in the traditional models.
Figure 11 illustrates this process in a model which includes expectations,
desires, attitudes, and subjective disconfirmation as antecedent variables.
Paths are presumed to apply both to those who are ultimately satisfied and to
those who are ultimately dissatisfied.

Some researchers have begun to question the traditional model on this

point based on inconsistencies observed in empirical studies (e.g., Droge and

130

Halstead 1989). Other indications of differential processing among satisfied
versus dissatisfied consumers, along with possible theoretical reasons why
this may be true, are discussed in Chapter II-S.2.

In this thesis, it is hypothesized that dissatisfied consumers process
more extensively than do satisfied consumers. Satisfied consumers do not feel
the need to rethink or analyze the factors which are responsible for their
largely positive state, probably because there has been "closure" on the
process. On the other hand, dissatisfied consumers do not reach closure as
easily. They are in a negative affective state and seek to understand why,
perhaps because they need to rationalize their feelings (Maddox 1981), or
perhaps because they want to avoid being dissatisfied in the future, or perhaps
because dissatisfaction somehow "amplifies" their "sensitivity" to the process
(Westbrook and Newman 1978). Whatever the motivational source of this
increased processing, it is hypothesized to occur among dissatisfied consumers.

The manifestation of increased processing among dissatisfied
consumers is expected to be evident in more and stronger links between
constructs measured at different points in time. Consider Figure 11, for
example. Expectations, desires and attitudes were all measured before exposure
to performance while disconfirmation and satisfaction were measured after
exposure. Because satisfaction/ dissatisfaction occurs only after exposure, the
relationship among variables measured pre-exposure should be unaffected by
the ultimate satisfied/ dissatisfied state. Once exposure occurs, consumers
reach some level of satisfaction very shortly after disconfirmation. Therefore,
the link between disconfirmation and satisfaction is also expected to be
significant for both satisfied and dissatisfied consumers.

The difference between satisfied and dissatisfied consumers is expected

131

Pre-performance Post-performance

 

 

Expectations

Disconfirmation

Satisfaction

 

Deskes

Figure 11
Hypothesized Relationships Among Key Antecedent Variables
and Satisfaction

Expectations, desires, and attitudes are formed preperformance. Expectations and desires
are hypothesized to have direct and indirect (via disconfirmation) effects on satisfaction;
attitudes are hypothesized to have only direct effects on satisfaction. Disconfirmation and
satisfaction are formed post-performance, and disconfirmation is hypothesized to affect

satisfaction.

132

to show up in differential relationships between pre— and post-exposure
constructs. Satisfied consumers are hypothesized to not feel the need to
reconcile or even examine pre-exposure expectations, desires, and attitudes in
light of post-exposure disconfirmation or satisfaction. Dissatisfied consumers
are hypothesized to attempt to reconcile their feelings by re-examining pre-
exposure expectations, desires and attitudes, and attempting to make them "fit"
post-exposure feelings. The only way to do this is by adjusting either
subjective disconfirmation, satisfaction, or both. Given that measurement
occurs after such attempts at reconciliation have been made, closer
relationships between pre-exposure and post—exposure constructs are
hypothesized.

In the Election Outcome Satisfaction Model (EOSM), this research
hypothesis predicts that the links from pre-election constructs to post-
election constructs will be weaker for Clinton voters than for Bush voters
(assuming, of course, that Clinton voters are satisfied and Bush voters are
dissatisfied with the outcome of the election). Specifically, for Clinton voters,
desires and attitude are expected to impact satisfaction directly, as is
disconfirmation (see Figure 12-a). These relationships are expected to hold for
Bush voters. However, for Bush voters, other relationships will also be
significant (see Figure 12-b). The basic rationale for this hypothesis is that
satisfied voters are expected to be largely "disconfirmation—driven" while
dissatisfied voters are expected to be driven by both expectations and
disconfirmation (see Section II-S.2 and Droge and Halstead 1991).

Though these relationships are expected to hold for both Bush and
Clinton as targets among both Bush and Clinton voters, only a single target is

presented in these figures for ease of presentation.

1 33
Clinton (Satisfied) Voters

Expectations

Disconfirmation

Satisfaction

 

 

Bush (Dissatisfied) Voters

 

Expectations

Disconfirmation

 

Satisfaction

Desires

Figure 12
Hypothesized Paths for Satisfied Versus
Dissatisfied Voters

it is hypothesized that the links from pre-election constructs to post-election constructs will
be stronger for Bush supporters than for Clinton supporters.

134

The major testable hypothesis, therefore is:

H3a: The total number of significant paths from pre-election to post-
election constructs, for both Bush and Clinton as targets, will be greater
among Bush voters than among Clinton voters.

Specific testable hypotheses are:

H3b: Among Clinton voters, the following paths will be significant and
positive in explaining satisfaction with Clinton: Desires -> Satisfaction,
Attitude -> Satisfaction, and Disconfirmation -> Satisfaction.

H3c: Among Clinton voters, the following path will be significant and
positive in explaining satisfaction with Bush: Disconfirmation -> Satisfaction.

H3d: Among Bush voters, the following paths will be significant in
explaining satisfaction with Clinton: Expectations -> Disconfirmation
(negative), Desires-> Disconfirmation (positive), Expectations -> Satisfaction
(negative), Desires -> Satisfaction (positive), Attitude -> Satisfaction (positive),
and Disconfirmation -> Satisfaction (positive).

H3c: Among Bush voters, the following paths will be significant in
explaining satisfaction with Bush: Expectations -> Disconfirmation (negative),
Desires-> Disconfirmation (negative), Expectations -> Satisfaction (negative),
Desires -> Satisfaction (negative), Attitude -> Satisfaction (negative), and
Disconfirmation -> Satisfaction (positive).

In the Campaign Process Satisfaction Model, only one relevant pre-
election construct is measured: expectations. Therefore, the research
hypothesis prediction is straightforward. Expectations will be significant
predictors of disconfirmation only among those from each voter group who
are dissatisfied with the process (Figure 13); among those from each voter

group who are satisfied with the process, expectations will be unrelated to

135

 

 

 

 

 

 

 

 

 

Figure 13
Hypothesized Paths in the Campaign Process Satisfaction Model for
Those Dissatisfied with the Process

Expectations, perceived performance and disconfirmation for campaign elements relative to
each candidate are hypothesized to affect overall satisfaction, for dissatisfied Clinton
supporters apg Bush supporters.

136

post-performance constructs (Figure 14).

Notice that the relevant subgroups to test the research hypothesis are
not Clinton versus Bush voters. Unlike the EOSM, in which there should be a
close relationship between voting behavior and satisfaction with the outcome,
satisfaction with the process is less related to voting behavior. It is quite
reasonable for a Clinton supporter to be dissatisfied with Clinton's advertising
effectiveness and running an unfair campaign, yet still be satisfied with the
outcome of the election. Because the research hypothesis is stated in terms of
satisfied versus dissatisfied consumers, it is necessary to test the hypothesis
using Clinton versus Bush voters who are satisfied versus dissatisfied with the
process as a classification variable.

Specific testable hypotheses using the CPSM, therefore are the
following:

H3f: Among those who are satisfied with the process, expectation will
not be a significant predictor of perceived performance or subjective
disconfirmation for any campaign element of either target's campaign.

H3g: Among those who are dissatisfied with the process, expectation
will be a significant predictor of perceived performance (positive path) and
subjective disconfirmation (negative path) for both campaign elements of
each target's campaign.

Alternatively, if small group sizes preclude analysis based on two two-
group analyses as planned, hypotheses will be tested using an analysis of
correlation coefficients.

Note that this research hypothesis does not state that it is satisfaction or

dissatisfaction per se which causes the increased processing; the question of

 

causation is explicitly addressed in the discussion (Chapter VI), but it is not

137

 

 

 

 

 

Figure 14
Hypothesized Paths in the Campaign Process Satisfaction Model for
Those Satisfied with the Process

Perceived performance and (isconfirmation (but M expectations) for campaign elements
relative to each candidate are hypothesized to affect overall satisfaction, for satisfied
Clinton supporters M Bush supporters.

138

addressed in this hypothesis. Rather, level of processing is hypothesized to be
associated with level of satisfaction. Likewise, the issue of using level of
satisfaction for segmentation purposes is not addressed here, as key elements

of segments (e.g., stability) are not investigated in this research.

III-2.4 Theme No. 4: Importance of Desires versus Expectations in
Satisfaction Formation

Until recently, the traditional satisfaction literature has focused
primarily on cognitive antecedents of satisfaction formation. Thus,
expectations are hypothesized to affect satisfaction both directly and
indirectly through disconfirmation. Recent authors have suggested that
desires, too, are important antecedents of satisfaction (see Chapter II-3.2). The
conditions under which one or the other might dominate the satisfaction
formation process has not been resolved.

It could be argued that on issues of great importance, the effects of
desires would exceed the effects of expectations on satisfaction. Individuals
have more at stake on issues of great importance and it is logical that
performance congruent with desires would be a major antecedent of
satisfaction. If someone who has just bought a house wants quiet neighbors
but expects to have noisy neighbors, and the neighbors are quiet, the person
is likely to be satisfied. If the neighbors are noisy, the person is likely to be
slightly dissatisfied even though the expectations are being exactly confirmed.
See Chapter II—3.2 for a complete discussion of recent developments on this

issue.

139

Figure 1 1 illustrates the traditional model modified to include desires. It
specifies relationships among expectations, desires, subjective
disconfirmation, and satisfaction. It is hypothesized that desires are
negatively related to disconfirmation. This hypothesis may appear to be
unreasonable because "disconfirmation" is really "disconfirmation of
predictive expectations." The question arises of why "desired performance
level" can affect the degree to which observed performance was "better/worse
than expected performance level." Consider subjective disconfirmation and
how it is usually operationalized. If disconfirmation truly was the difference
between actual performance and predictive expectation, then the hypothesis
would be unreasonable. However, subjective disconfirmation encompasses a
judgment of the degree to which remembered predictive expectations are met
by perceived performance and an evaluation of any difference perceived. It
seems reasonable to think that desires might affect a consumer's memory of
what was expected, the perception of performance, and the evaluation of the
disparity between perceived performance and expectation. In other words,
desires can affect memory, perception, and judgment - three key elements in
how subjective disconfirmation is operationalized.

The research hypothesis proposed in this thesis is that the direct and
total effects of desires on satisfaction will exceed the direct and total effects of
expectations on satisfaction, given that the target is generally perceived to be
important. When individuals have strong desires, it seems obvious that their
ultimate level of satisfaction will at least be somewhat dependent on these
desires. As demonstrated throughout Chapter II, when expectations is the only
pre-performance construct measured, it is found to play a major role in the

satisfaction formation process. These findings are especially strong in

140

laboratory studies in which variance in desires is likely to be minimal, and/ or
when the target is of relatively slight importance. However when desires are
included along with expectations as potential antecedents of satisfaction, the
disconfirmation of desires is found to be significant (e.g., Barbeau 1987;
Spreng 1990; Spreng and Olshavsky 1993), and disconfirmation of
expectations is not significant.

Further, Spreng (1990) has pointed out that one possible reason for
failure to find support for the effects of desires on satisfaction is the difficulty
involved in attempting to manipulate consumer desires in a laboratory setting.
Without the ability to introduce variance in desires, one popular traditional
avenue of conducting satisfaction research has been unproductive in
investigating this relationship. Likewise in field studies, variance in desires
has been difficult to establish in traditional consumer satisfaction studies,
probably for reasons similar to those behind finding minimal variance in
measures of satisfaction (Peterson and Wilson 1992). Just as there is slight
variance in satisfaction measures in empirical studies because most consumers
are satisfied with most purchases, there may be slight variance in desires
because most people desire the same things: high quality, good service, low
price, etc.

For these reasons, the Election Outcome Satisfaction Model provides a
uniquely favorable set of circumstances for testing the research hypothesis.
Specifically, variance in both satisfaction and desires is virtually guaranteed
in a presidential election: large proportions of people are as likely to be
dissatisfied as satisfied with the election outcome. Since it is likely that most
voters consider the presidential election to be an important event, the

research hypothesis indicates that desire should dominate expectation as an

141

antecedent of satisfaction; this should be true for both supporters of Bush and
supporters of Clinton.

Specific testable hypotheses, therefore, are:

H4a: The magnitude of the desires -> satisfaction path will be greater
than the magnitude of the expectation -> satisfaction path for both targets
among Clinton voters.

H4b: The magnitude of the desires -> satisfaction path will be greater
than the magnitude of the expectation -> satisfaction path for both targets
among Bush voters.

H4c: The total effects, direct and indirect, of desires on satisfaction will
exceed the total effects of expectation on satisfaction for both targets among
Clinton voters.

H4d: The total effects, direct and indirect, of desires on satisfaction will
exceed the total effects of expectation on satisfaction for both targets among

Bush voters.

III-3. Models to test hypotheses

These four research hypotheses are to be tested in one or both of two
models. Testable hypotheses for each research hypothesis and each model
have been presented throughout Section III-2. An overview of each complete

model is presented next. Construct operationalizations appear in Chapter IV.

III-3.1. Election Outcome Satisfaction Model
The complete Election Outcome Satisfaction Model is presented in Figure
15. Where applicable, hypothesized signs of each path are indicated. Also, the

related testable hypotheses are identified for each path.

     
 

Attitude
toward
Clinton

 

Expectations

142

     
 

 

Satisfaction w/

 

  

re: Clinton

Desires re:

        
 

  

Clintons's
Performance

    

   

D .. nflrrnatlon w/
Clintons‘s
Performance

 
   
     

 

Clinton

Desires re:

Bush
0,-

Overall

Satisfaction

 

  

Disconfirmation
w/ Bush's
Performance

 
   
     
    

Satisfaction
w/ Bush's

 

 

 

Expectations
re: Bush

Attitude

Bush

Performance

   

0,‘

c,-

Figure 1 5

Complete Election Outcome Satisfaction Model

Note: First sign is the hypothesized sign for Clinton supporters and the second sign is the
hypothesized sign for Bush supporters.

143

III-3.2. Campaign Process Satisfaction Model

The complete Campaign Process Satisfaction Model is presented in
Figure 16. Where applicable, hypothesized signs of each path have been
indicated.

Also, the related testable hypotheses have been identified for each path.
Construct operationalizations and complete model specifications are presented
in the next chapter.

For ease of reference, the four major research hypotheses are
summarized once again:

(1) Multiple targets of satisfaction are relevant to the overall satisfaction
formation process.

(2) Processes relevant to the formation of satisfaction with competing
alternatives are structurally interrelated.

( 3) Dissatisfied consumers process more extensively than do satisfied
consumers.

(4) When the target of satisfaction is relatively important, the effects of
desires will dominate effects of expectations on the entire satisfaction

formation process.

144

 

 

 

 

 

 

 

 

 

Figure 16
Complete Campaign Process Satisfaction Model

Expectations, perceived performance and disconfirmation for campaim elements relative
to each candidate are hypothesized to affect overall satisfaction, for both Clinton
supporters and Bush supporters.

Chapter IV
Method

This chapter is divided into three sections. In the first section, data
collection methods and the critical issue of timing are discussed. The second
section presents operationalizations of key constructs for the Election Outcome
Satisfaction Model. The third section contains similar information for the

Campaign Process Satisfaction Model.

IV—l. Data Collection and Timing

To test the research hypotheses developed in the previous chapter, data
were collected at two points in time: pre-election information was collect in
September, 1992, and post-election information was collected in November,
1992, immediately following the election.

Data were collected by MBA students at Michigan State University and
the University of Kentucky for course credit and under the supervision of
course professors. Each student was directed to contact five acquaintances and
administer the pre-election questionnaire in- person or by telephone.
Following the election, students recontacted each respondent and administered
the post-election questionnaire. A total of 331 pre-election questionnaires
were completed while 303 usable post-election questionnaires were completed.
The discrepancy of 28 is due to respondents of the pre-election sample not
voting in the November election. Only respondents who completed both pre-
and post-election questionnaires are included in this analysis.

Although questionnaires administered via personal contact are not new

to consumer research in general, they have not been utilized in field studies of

145

146

satisfaction formation. For this particular context, personal contact is the
preferred method of data collection. In the 1992 presidential campaign
process, significant developments occurred weekly and it is likely that history
effects could have jeopardized data from a mail-out survey due to the typically
wide variance in response times. Personally contacting respondents enabled
tight control of the timing of data collection to minimize the chance that
respondents would be differentially affected by media stories.

The most obvious illustration of this potential confound is the behavior
of candidate Ross Perot. Mr. Perot was a candidate for president throughout
early 1992, withdrew from the race in the summer and re-entered the race in
September. Had a mailout survey been distributed during the period Mr. Perot
was not a candidate, it is almost certain that some respondents would have
answered while he was out of the race and some would have responded after he
re-entered the race. Measures of expectations and desires would have been
rendered noncomparable as the dynamics of the campaign radically shifted
with the single event of Mr. Perot's re—entering the race. Similarly,
expectations regarding advertising effectiveness and running an unfair
campaign would likely change on a weekly basis as the campaign entered it
final month. likewise, the possibility of major events occurring following the
election (e.g., increasing U.S. involvement in Somalia or Bosnia) illustrates the
need to complete all post-election data collection in a short time period, too.
Therefore, minimizing the variance in data collection timing is important to
addressing the research questions, and personally contacting respondents to
collect data is the most appropriate technique for accomplishing this

objective.

147

Some may question whether the snowball sample selection methodology
used here resulted in a "representative sample." Kerlinger (1986) defines a
representative sample as one which "....has approximately the characteristics
of the population relevant to the research in question" (p. 11 1). If one was
interested in predicting population voting patterns or in describing the
demographics of the population of voters, then this sample would not be
considered representative. However, testing the research hypotheses in this
dissertation does not require a representative sample in this sense. All
research hypotheses involve relationships among key constructs and thus the
absolute magnitude of population proportion or mean values is not relevant.
All that is required is that the relationships between theoretical constructs not
vary systematically between the sample and a relevant population, and there
is no reason to suspect that such systematic differences might exist here. As
long as external validity is of secondary importance in a particular research
endeavor, as it is here, convenience samples are considered to be adequate for

theory testing purposes (Calder, Phillips and Tybout, 1982).

IV-2. Operationalization of Key Constructs

In this section, the operationalization of key constructs for each model
is presented and discussed. Table 15 lists all constructs in a matrix so that the
appropriate model and the timing of measurements is clear. Issues related to
the operationalizations of constructs are discussed in Section 4.2.1 for the
Election Outcome Satisfaction Model (EOSM), and in Section 4.2.2, for the

Campaign Process Satisfaction Model (CPSM).

Election
Outcome
Satisfaction
Model

Campaign
Process
Satisfaction
Model

148

Table 15
Operationalization of Constructs
Timing of Measurements for Each Model

Measured Before Measured After
the Election the Election

- Expectations of popular - Disconfirmation related

vote percentages to actual election outcome
percandidate

- Desired distribution of - Overall satisfaction with
100 points across results
candidates

- Beliefs about each - Satisfaction with results
candidate on (a) foreign obtained by each candidate

policy, (b) domestic
economy, (c) other
domestic issues, and
(d) personal integrity

- importance of the four
issues listed above

- Expectations re: - Perceived performance
advertising effectiveness re: advertising
effectiveness and unfair
campaign
- Expectations re: unfair - Disconfirmation re: ad.
campaign effectiveness and unfair
campaigns

- Disconfirmation of
overall process

- Satisfaction with
overall process

149

IV-2.1. Election Outcome Satisfaction Model - Operationalization of
Key Constructs
The complete EOSM is depicted in Figure 15. Construct
operationalizations are presented here in the order in which they appear in
the model, from left to right. A table summarizing these operationalizations,

Table 16 appears at the end of this section.

Expectations
In Section II-3.l, the range of expectations measurements was described .
and discussed. It was concluded that the most reasonable operationalization of
expectations was "predictive expectations," that is, the level of performance
actually anticipated by the respondent as measured before exposure to
performance. Further, measurement of expectations appeared to be most
consistent with its conceptualization when the evaluative nature of

expectations was held to a minimum. Therefore for the election outcome

 

model, expectations is operationalized as the percent of mpular votes each

candidate is actually expected to receive on voting day.

Several advantages result from measuring expectations in this manner.
First, it guarantees that expectations are measured on exactly the same scale as
is the ultimate objective performance of each candidate. Thus, direct and
objective comparison of performance with a priori expectations of
performance is possible. Second, expectations and desires are also measured
on exactly the same "percentage" scale, enabling direct comparisons of
expectations and desires. Third, it enhances the competitive dimension of the
model by forcing explicit trade-offs among candidates in a zero-sum context

(i.e., percentages must add to 100).

150

Desires

As discussed in Section lI-3.2, desires has been infrequently included as
a separate antecedent variable. As is true of expectations, desires is most
useful when operationalized in terms of objective performance levels rather
than evaluative performance levels. Therefore for the election outcome

model, desires are operationalized a_s the percentage of votes (that is. the

number of points out of 100) a subject would want to assign to each candidate
on voting day.

 

There are two major advantages to this operationalization of desires.
First, as discussed above with regard to expectations, the measurement for
desires is directly comparable with measures of both expectations and ultimate
performance. No previously published study has included such directly
comparable measures of these three constructs. Second, this
operationalization of desires is unambiguous enough to allow responses of
either the "more is better" or the "ideal point" variety (see Section II-3.2).
Previously used scales left this distinction open to interpretation of both the
subject and the researcher.

One possible disadvantage of this operationalization is that it may not
match the manner in which voters would normally conceptualize or express
their desires in an election context. Specifically, some voters may simply
desire that "Bush win" or that "Clinton win," and may not care by how much.
A related potential problem is that someone may desire a candidate to receive
35% of the vote and win (in a three way race) while someone else may desire a

candidate to receive 49% of the vote and lose.

151

The second of these potential problems is addressable, in part, through
empirical analysis. In a three way race, it is clear that a candidate who
receives 32% of the vote or less is a loser and a candidate who receives 51% or
more of the vote is a winner. Between 32% and 51%, however, a candidate may
be either a loser or a winner. To address this issue, respondents were also
asked who they would vote for.

The first potential problem, however, is less easily addressed. If
someone only desires that someone win or lose, is that person constrained by
the format of the measurement from adequately expressing his or her desires?
The answer is "probably not" given the characteristics of the sample. The
sample is generally well-educated and it is doubtful that a respondent would be
unable to express his or her desires in the format requested. A more serious
question may be whether the measurement format imposes an unrealistic
level of exactness on the results. In other words, might someone desire only
that Bush win, but, because of the format of the question, be forced to assign
somewhere between 34 and 100 points to Bush thereby implying more finely-
tuned desires than is actually felt? Although this may have occurred to some
degree, there is no evidence from interviewers or from response rates to this
specific question that any confusion existed. Further, because the analysis is
focused on relationships rather than on means, it is unlikely this potential
problem has major implications for the analysis proposed here. Though a
respondent may have only desired that candidate A win, it is likely that they
could rank order the three candidates in some sort of preference order; the
assignment of points is a more detailed way of specifying the relative desires

associated with each candidate.

152

Finally, the task of assigning 100 points to reflect desired candidate
performance in this particular context seems especially appropriate given
that the media presented polling results in a similar format (proportion of
likely voters desiring each candidate) almost daily. Thus, in the context of a
political election asking respondents to assign 100 points to reflect level of
desire for each candidate appears to be a reasonable strategy.

One potential drawback of the operationalization of both expectations
and desires is the use of a single item scale to measure each. Although
multiple item scales are generally perceived to be superior, it is difficult to
imagine how items could be added to the measurement of both constructs
which 1) were not simply a reworded version of the current item and 2) would

not jeopardize the direct comparability of the two variables with each other

and with actual parformance.

Attitudes

A subject's attitude towards a satisfaction target has been shown to be an
antecedent of satisfaction processing (see Section II-3.3). Attitude is
operationalized in a manner consistent with the Fishbein (1967) expectancy-
value model, that is, as the summed cross products of beliefs (that the target
possesses a specific characteristic) times the importance (of that

characteristic) in a manner which is both context-specific and personalized.

For the election outcome model, a similar belief times immrtance
operationalization is used, incorporating the following four characteristics:
domestic economy, other domestic issues, foreigp policy, and personal

integrity. These four characteristics were consistently shown in public

opinion polls to be salient candidate attributes. Likewise, media coverage of

153

the election seemed to focus on these candidate attributes. Beliefs were
measured on a 10 point scale anchored by "will do an extremely poor job" (= 1)
and "will do an extremely good job" (= 10). Importance of each attribute was
measured on a 10 point scale anchored by "extremely a_nimportant" (= 1) and

"extremely important" (= 10).

Disconfirmation

The concept of disconfirmation has been central to most of the
empirical satisfaction literature since about 1980. Various attempts at
conceptualizing and measuring disconfirmation are discussed in detail in
Section II-3.5. The two major decisions to make regarding the
operationalization of disconfirmation are 1) what standard should be used and
2) whether the measure should be subjective or derived. In order to make the
measure of disconfirmation consistent with the measure of expectations, the
standard disconfirmed is predictive expectations.

Further, in this dissertation, the measurement of disconfirmation will
be subjective rather than derived for two reasons. First, subjective measures
appear to be the most consistently significant as antecedents of satisfaction,
despite some theoretical shortcomings. Second, given that ultimate
performance is objective, it is uniform across all subjects. Any derived
measure would require the subtraction of expectations from a "constant"
(actual performance), and thus would explain no variance in satisfaction in
addition to the variance already explained by expectations. Thus, a derived
measure would be completely meaningless in the model proposed here.

Therefore, for the election outcome model disconfirmation is the

 

subjectively expressed degree to which a candidate's actual performance was

 

154

worse (= 1) or better (= 7) than expected. This operationalization is most
consistent with other subjective measures of disconfirmed predictive

expectations.

Satisfaction

The broad range of satisfaction measures, from non-verbal symbolic
scales through complex multiattribute summated scales, is reviewed in detail in
Section II-2. Here, satisfaction is measured for the performance of each

candidate and for the election outcome overall. In this dissertation

 

satisfaction with Bush's performance, satisfaction with Clinton's performance,

 

and overall satisfaction with the outcome of the election are each measured on

identicai scales ranging from "extremely satisfied" (= 7) to "extremely
dissatisfied" (= l).

 

Variable

Expectations

Desires

Attitudes

Disconfirmation per
Caiddate

Overall Satisfaction

Satisfaction with
Carddate

155
Table 16

Wording

Measured before the election

Regardless of who you personally support,
what do you think the outcome of the election
will actually be? Please indicate the percent
of the popular vote that you expect each
candidate will actually get.

Sippose for the moment that you codd
distribute 100 points among the candidates
instead of having just one vote. You can give
all 100 to one cmdidate and zero to the
others, or you can distribute the points any
wayyouwish. Howmanypointswouldyou
give to each candidate?

Please evaluate the three candidates on (A)
foreign policy, (8) the domestic eccnorny,
(C) domestic issues excluding the economy,
and (D) personal 'ntegrity....Give each
candidate between 1 and 10 points.

How important are the move sets of issues to
you?

Operationalization of Constructs --
Election Outcome Satisfaction Model

Scale

0 through 100
toeechcanddate
(summing to 100
across all
cmddates)

0 through 100
toeachmnddate
(summing to 100
across all
cmddates)

1- 'Will do an
extremely poor
job“ though

10- 'Will do an
extremely good job'

1 -" Extremely
Unimportant'
through

1 O-' Extremely
Important“

(Note: the belief and importance measuernents for each attribute
are multiplied then summed across all attributes to create a

formative construct)

Measured after the election

Were the results better or worse than you
expected? Note that whether your canddate
woncrlost, hecouldhave donebetteror
worse that you expected he would.

The actual results of the election were:

3896 for Bush, 4396 for Clinton, and

1996 for Perot. How satisfied or dissatisfied
are you with the results?

Please mswer the same question for each
of the candidates separately.

7-pt. semantic

differential from

" much worse than
Iexpected' to

" mudi better than

Iexpected'

7-pt. semantic
differential from
'extremeiy
dissatisfied' to
'extremeiy
satisfied'

7-pt. semantic
differential from,
'extrernely
dissatisfied“ to
'extremeiy satisfied“

156

IV-2.2. Campaign Process Satisfaction Model - Operationalization
of Key Constructs
The complete CPSM is depicted in Figure 16. Again, construct
operationalizations are presented in the order in which they appear in the
model, from left to right. A table summarizing these operationalizations, Table

17 appears at the end of this section.

Expectations

A complete discussion of the range of expectations measurement
appears in Section II-3.l. As in the EOSM, expectations is most reasonably
operationalized as "predictive expectations." However, unlike the EOSM,
measurement of expectations cannot utilize an "objective" scale as there is no
widely understood and accepted means of measuring advertising effectiveness
or campaign unfairness. Therefore. in t_hi_§ model, eapectations regarding

each candidate's performance with regard to advertising effectiveness and

running an unfair campaigp is measured on a "will definitely hapmn" (= 10)
to "will definitely not happen" (= 1) scale.

Perceived Performance

Again, because there are no widely understood and accepted means of
measuring advertising effectiveness and/or running an unfair campaign,
perceived performance is measured in this model. To maintain comparability
with the expectations measure, perceived performance is measured on similar
scales. In this model, perceived performance with rem to both aavertisigg
effectiveness and running an unfair campaign are measured on a scale

anchored with "False-didn't ha en" = l to "True--ha ened a lot"(= 10).

 

157

Disconfirmation

As in the EOSM, disconfirmation is measured as a subjective construct in
which predictive expectations are the standard disconfirmed.
Disconfirmation. therefore. is measured as the degree to which performance is
Mged to be "more than expected" (= 10) or "less tLan expected" (= 1) for ear;

of the two included process attributes.

Overall Disconfirmation

Since "overall process satisfaction" is the ultimate target of this model,
an intervening, general measure of disconfirmation is included. This is
necessary because the term "campaign process" could reasonably include a
very large number of possible campaign attributes. Because only two of the
possible campaign process attributes are considered in this model, an overall
measure of disconfirmation is included so that other non—specified attributes
salient to individual respondents (which may be more important than the two
included attributes) can contribute to explaining variance in overall process
satisfaction. Overall disconfirmation is measured on a seven point "much

better than expected" (= 7) to "much worse than expected" (= 1) scale.

Satisfaction

The range of possible satisfaction measures and a discussion of
satisfaction "targets" appears in Section 11.2. In the CPSM, satisfaction with the
campaign process is measured. The scale used is identical to the seven point
satisfaction scale used the EOSM, ranging from "extremely dissatisfied"

(= l) to "extremely satisfied" (= 7).

158

Table 17
Operationalization of Constructs
Campaign Process Satisfaction Model

Variable Wording Scale

Measured before the election

Expectations Please indcate how you expect the election 1-"No, this will
campaign will be conducted [asked for definitely not
"advertising will be extremely effective" and happen" to
"will run an unfair campaign" for 10-"Yes, this
eachcanddate] will definitely
happen"

Measured after the election

Perceived Performance We would like you to make two evaluations 1-"false (didn't
each on a ten point scale. The first is an happen)" to
evaluation of how true or false a statement 10-"true
is....[asked for "advertising being extremely (happened a
effective" and "running an unfair campaign" lot)"
for each canddateJ

Disconfirmation The second is an evaluation of whether some- 1-"less than 1
thing happened more or less that you expected expected" to
it would. [asked for "advertising being 10-"more than
extremely effective" and "running an unfair lexpected"
campaign" for each candidate]

Overall disconfirmation [Overall] Was the process better or worse 7-pt. semantic
than you expected? differential from

"rruch worse than
lexpected" to

"much better
than I expected"

Overall Satisfaction Overall, how satisfied or dissatisfied are you 7-pt. semantic
with this year's presidential election differential from
campaign process (i.e., the way the carmaigi "extremely
was condicted)? dssatisfied" to

"extremely

satisfied"

159

IV-3. Other Relevant Variables

In addition to the variable used in each of the models, other relevant
variables were collected and used during data analysis. These variables are
enumerated in Table 18. First, both level of certainty and level of involvement
were thought to have the potential to compete with the independent variables
of interest in the two models; collecting data on them would allow their effects

to be controlled for, if needed.

Choice Certainty

Choice certainty was measured to determine whether Bush versus
Clinton voters differed in the degree to which they were confident of their
selection. This variable was collected because systematic differences in level
of certainty could logically be related to amount of processing undertaken by
certain respondent groups. Certainty was measured both before and after the

election.

Involvement
Involvement was measured using a reduced version of Zaichowsky's
(1985) involvement scale. It, too, was measured because differential levels of

involvement might be expected to affect processing intensity.

Political Affiliation and Demographics
Political affiliation and demographic variables were collect for
classification and comparison purposes. Interviewers were instructed to

gather data from acquaintances which reflected the population of adult voters.

160

Collecting this information allowed comparisons to be made to the general
electorate.

In addition, political affiliation variables, including past voting
behavior, were collected to determine the extend of "brand switching" which

occurred among respondents.

Operationalization of Constructs
Other Relevant Variables

Variable

 

Certainty
(before election)

Certainty
(after election)

Involvement

Political Affiliation

Demographics

161
Table 18

Wording

How certain are you of your choice?
Putacheck markinoneoftheseven
spaces to reflect how certa‘n you are.
(asked after determining who subject
would vote for " ....if the election were

held tomorrow....')

How certain were you of your choice?
Put a check mark in one of the seven
spaces to reflect how certain you were

at the time you were voting.
How do you feel about the coming

presidential election? Please put a

check mark in one of the seven
spaces to reflect how you feel.
(5 semantic differential scales)

Are you a registered voter?
Are you a registered Democrat?
Are you a registered Republican?

Did you vote in the last election?
If so, for whom?

Sex:

Age

Please check the highest level of
emcation you atta'ned:

Please check off your total family
inccrne per year. By 'family' we
mean 81 people related to you and
living in your household.

Scale

1 - Not at all certain
to
7 - Extremely certain

1 - Not at all certain
to
7 - Extremely certain

1 - unimportant to me -
7 - important to me
1 - irrelevant to me --
7 - relevant to me
1 - uninteresting to me --
7 - interesting to me
1 - bor'mg to me --
7 - fascinating to me
1 - means nothing in my life-
7-meansalotinmylife

Yes-No
Yes-No
Yes-No

Yes - No
Bush - Dukakis

Male - Female
... years

_less than high school

_graduated from him school

..some college or post high

school education

_graduate from 41-year college
(BA/BS or similar)

-Master's level degree or
equivdent

_Ph.D.

-under $5,000
-35,000 - $9,999
.310,000 - $14,999
.315,000 - $19,999
-320,000 - $24,999
_SZS.000 - $29,999
_S30,000 - 3 39,999
.340,000 - $49,999
.350,000 - $59,999
-Over $60,000

Chapter V
Results

This chapter is divided into six sections. The first section presents a
general description of the Clinton and Bush voter subsamples used in the
various two-group analyses. Sections two through five of this chapter each
deal with a different analytical tool - subsample combination. The second and
third sections each deal with one of the main hypothesized models (Election
Outcome Satisfaction Model and the Campaign Process Satisfaction Model)
comparing Clinton versus Bush voters. The fourth section presents results of
an analysis of satisfied versus dissatisfied voters within the Election Outcome
Satisfaction Model. The fifth section contains a correlation analysis of satisfied
versus dissatisfied voters on variables collected for the Campaign Process
Satisfaction Model. The sixth section summarizes research findings which
specifically relate to the four research hypotheses.

Within sections two through five of this chapter, results relating to the
four research hypotheses are presented. Table 19 summarizes the research
hypotheses and identifies which section contains results related to which

research hypothesis.

V-l. General Description of the Subsamples

For each of the two models, a two-group analysis is performed. The two
groups analyzed are Clinton supporters and voters versus Bush supporters and
voters. "Supporters and voters" are defined as those who reported supporting
a particular candidate pre-election and voting for that same candidate: those

who supported and voted for Clinton are in the first group, and those who

162

Organization of Chapter 5--Sections Addressing
Results Relating to Specific Research Hypotheses

Research Hypotheses

#1: Multiple targets of satisfaction
are relevant to the overall satisfaction
formation process.

#2: Processes relevant to the formation
of satisfaction with competing alternatives
are structurally related (via cross-over
effects).

#3: Dissatisfied consumers process more
extensively than do satisfied consumers.

#4: Effects of desires will dominate the
effects of expectations on the satisfation
formation process in high involvement
situations.

163
Table 1 9

Analytical
Complete §ample
EOSMl PSMZ
Section Number....

V.Z.3 V.3.3
v.2.4 - -
v2.5 - -
V2.6 - -

Tool

Partial
EOSM3

V.4.Z

V.4.3

V.4.1

am le
CPSM4

V.S

 

1) Clinton voters versus Bush voters in Z-grcup analysis
2) Clinton voters versus Bush voters in Z-group analysis

3) Satisfied Clinton voters versus Dissatisfied Bush voters in Z-group analysis

4) Satisfied Clinton voters versus Dissatisfied Bush voters in correlation analysis

164

supported and voted for Bush are in the second group. For convenience, these
groups will be referred to as "Clinton voters" and "Bush voters" in the
remainder of this dissertation. Note that those who supported one candidate
before the election but voted for a different candidate are omitted from both
groups (as are those who voted for Perot).

Table 20 presents means of the variables for each group and significant
differences between the groups on these variables. Differences between the
groups with respect to variables used in each model are discussed in the
relevant sections to follow.

From Table 20, note especially the following:

1) There is some evidence of allegiance-switching prior to the pre-
election measurement. First, over one fourth of Clinton voters (28.2%) voted
for Bush in the previous election, though only 396 of Bush voters voted for
Dukakis. Second, 14.796 of Bush voters are registered Democrats while 7.796 of
Clinton voters are registered Republicans.

2) Voters for each candidate appear to be equally involved in the election
as measured by all five involvement items: importance, relevance, interest,
fascination, and meaning.

3) Voters for each candidate reported equal levels of certainty about their
choice before the election, while Clinton voters reported slightly higher
levels of certainty following the election.

4) There is no difference in the age of Clinton versus Bush voters in this
sample; Clinton voters have received more education, and Bush voters reported
higher household incomes.

5) In this sample, significantly more Bush voters versus Clinton voters are

male (52.6% versus 29.5%).

165
Table 20
Description of the Sample
Differences Between Clinton and Bush Subsamples
on Selected Variables

 

Significantly

Clinton Bush Different?
19.9221 yams: -vaIue 2
Political packggound
Percent registered Democrats 61.5 14.7 <.053
Percent registered Republicans 7.7 64.2 <.053
Voting pattern in 1988 election:
a) voted for Bush 28.2 68.0 <.053
b) voted for Dukakis 38.5 3.1 <.053
c) rid not vote 33.3 29.9 - -
lnv 1v t d ertain
Relevance of the 1992 election4 (7 pt. scales):
a) "important to me" 6.35 6.41 - -
b) "relevant to me" 6.06 6.22 - -
c) "interesting to me" 6.06 6.05 - -
d) "fascinating to me" 5.55 5.33 - -
e) "means a lot in my life" 5.71 5.75 - -
Certainty of choice (7 pt. scales)5:
a) before the election 6.14 6.18 - -
b) after the election 6.32 6.00 .09
Other Characgeristics
Demogaphics6
a) age 31.5 29.8 - -
b) income category? 6.66 7.69 .03
c) education (96 college grad or higier) 69.2 50.5 <.053
d) gender (96 male) 29.5 52.6 <.OS3

 

(1)
(2)
(3)
(4)

Includes respondents who supported the cmdidate before the election and voted for the canddate.

Using a difference of means t-test unless otherwise noted.

Difference of proportions test.

Opposite scale labels were " uninportant to me,‘ 'irrelevmt to me,“ 'minteresting to me,‘ “boring to me“

and 'meens nothing in my life.‘ Measurements taken before the election.

(6)

(5)
(7)

Demographic characteristics of all U.S. voters (source: New York Times November 5, 1992, p. 89)

 

Clinton Votgs Bush Voters

median age: 41.3 41.8
medan 'ncome: $35,024 $42,407
96 college graduate or higher. 40.3 41.6
gender (96 male): 43.2 46.7
Endpoints labeled 1 - ”not at all certain,‘ 7 - 'extremeiy certain."
Categories:

1 - raider $5,000 6 - $25,000 - $29,999

2 - $5,000 - $9,999 7 - $30,000 - 3 39,999

3 - $10,000 - $14,999 8 - $40,000 - $49,999

4 - $15,000 - $19,999 9 - $50,000 - $59,999

5 - $20,000 - $24,999 10 - Over $60,000

166

V-2. Results of the Election Outcome Satisfaction Model

Results of the two-group analysis using the Election Outcome
Satisfaction Model (EOSM) are presented in this section. First, differences
between the two groups with respect to the variables included in the model are
highlighted. Then, results of the LISREL analysis are presented. Finally,

results associated with each of the research hypotheses are described.

V-2.1. Differences Between the Two Groups (EOSM)

Although research questions specifically address the relationships
between key variables, means and standard deviations of model variables are
presented in Table 21 for completeness and to highlight differences between
the groups.

Note that Clinton and Bush voters differ significantly on all variables
included in the model (for all differences, p-values < 0.001 for two-tailed
difference of means t-tests, except as indicated). Specifically, Clinton voters
expressed a higher level of desires, expectations and attitude for Clinton than
did Bush voters, while Bush voters indicated higher levels than Clinton voters
on these same variables with regard to Bush. Regarding Clinton's
performance, Bush voters had a slightly higher level of disconfirmation (p =
0.124 for two-tailed t-test) than did Clinton voters. Regarding Bush's
performance, Clinton voters indicated higher disconfirmation than did Bush
voters. Finally, Clinton versus Bush voters expressed higher levels of
satisfaction with Clinton's performance, Bush's performance, and the overall

election results.

167

Table 21
Election Outcome Satisfaction Model
Means and Standard Deviations of Each Group

 

 

Clinton Bush p-values
Voters Voters from
Mean Std. Dev. Mean M M1
For Clinton:
Desires 76.179 18.910 11.381 14.576 0.000
Expectations 49.051 7.243 44.490 7.823 0.000
Attitudes 243.909 44.512 1 54.660 68.250 0.000
Disconfirmation 4.645 1.334 4.979 1.472 0.124
Satisfaction 5.705 1.250 1.794 1.020 0.000
For Bush:
Desires 12.808 14.667 78.948 19.325 0.000
Expectations 37.676 6.340 42.552 9.055 0.000
Attitudes 178.779 56.104 266.221 63.918 0.000
Disconfirmation 4.263 1.237 2.854 1.179 0.000
Satisfaction 4.551 1.593 2.010 1.075 0.000
Overall:
Satisfaction w/ election results 5.910 1 .1 86 1 .979 1.1 1 8 0.000

 

1 p-values for t-test for Ho: u1- ”2-

168

When groups are compared on the degree of pre-election desires with
regard to "their candidate," a similar level is observed. That is, Clinton voters
had a mean desire of 76.2 for Clinton while Bush voters had a mean desire of
78.9 for Bush (p value = .358). However, Bush voters had a more positive
attitude towards Bush than Clinton voters had towards Clinton (p value =.008),
and Bush voters had a poorer attitude towards Clinton than Clinton voters had
towards Bush (p value = .012). The two groups differed on their expectations of
the relative performance of each candidate: Clinton voters showed a
difference of l 1.4 in the expected performance of Clinton over Bush, while
Bush voters showed a spread of only 1.9 for Clinton over Bush.

Because attitude towards each candidate is measured by combining the
belief times importance of four candidate attributes, a separate table (Table 22)
is presented to provide details of this variable. Again, note that the mean
values of all importance component variables differ significantly between
groups, as do the mean values of all belief component measures. While both
groups give highest importance ratings for "domestic economy" and lowest
ratings to "foreign policy," there are differences between the groups. Clinton
voters rate "domestic economy" and "other domestic issues" as being more
important than do Bush voters (p = .069 and p = .020, respectively). Bush voters
give "foreign policy" and "personal integrity" higher importance ratings
than do Clinton voters (p = .010 and p = .005, respectively). If the attributes are
rank-ordered based on these importance ratings, "personal integrity" would
rank above "other domestic issues" for Bush voters, while the reverse order

would hold for Clinton voters.

169

Table 22
Election Outcome Satisfaction Model
Components of Attitude Towards the Candidate

 

Significantly
Clinton Bush Different?
Voters1 Votersl -value 2

MW
knportance to respondents.

(10 pt. scales)3:

a) foreign policy 7.10 7.88 .010

b) domestic economy 9.36 9.02 .069

c) other domestic issues 8.71 8.16 .020

d) personal integrity 7.85 8.62 .005
Belief Cornponmt
Evaluation of candidate Clinton.

(10 pt. scales)4:

a) foreign policy 6.45 4.02 .000

b) domestic economy 8.06 5.27 .000

c) other domestic issues 7.77 5.23 .000

d) personal integrity 6.92 3.97 .000
Evaluation of candidate M.

(10 pt. scales)4:

a) foreign policy 7.55 9.05 .000

b) domestic economy 4.30 6.94 .000

c) other domestic issues 4.19 7.01 .000

d) personal integrity 6.16 8.40 .000

 

( 1 ) lndudes respondents who reported supporting the candidate before the election and voted for the canddate.
(2) Using a difference of means t-test with Ho: u1-u2.

(3) Endpoints labeled 1 - 'extremeiy animportant' to 10 - 'extrernely important to me.“

(4) Endpoints labeled 1 - "wil do an extremely poorjob' to 10 - 'will do an extremely goodjob.‘

170

V-2.2. Results of the EOSM Two-Group LISREL Analysis

First, the correlation matrix for each group is presented. Then, overall
and group-specific indicators of fit are provided.

The top half of Table 23 contains the correlation matrix for Clinton
voters. The bottom half of Table 23 presents results for Bush voters in a
similar format. Covariance matrices were used as input for the analyses.

Overall, the model fit the data moderately well. The appropriate fit index
of the two group model overall is chi square goodness-of-fit. For EOSM, the chi
square value is 75.60 with 46 degrees of freedom (p = .004). In addition, the
comparative fit index (CFI) is .95.

Within group indicators of fit are provided in Table 24. The goodness of
fit (GFI) index for Clinton and Bush voters are .938 and .927, respectively.
Squared multiple correlations for endogenous variables in the Clinton voter
group range from .001 to .232, the total coefficient of determination of the
structural equations is 0.127. These statistics are all higher for the Bush voter
group: squared multiple correlations range from .024 to .415, and the total
coefficient of determination for the structural equation is .330.

There are two modification indices greater than five for the Clinton
voter group and five for the Bush voter group. A total of five standardized
residuals lie outside the -2.0 to 2.0 range for the Clinton voters, while eleven
lie outside that range for the Bush voters. Although there are several
standardized residuals outside the -2 to +2 range in each group, an examination
of their distribution indicates they are roughly normally distributed (see
Figure 17).

Of the five high modification indices indicating the need for additional

structural paths, four of the suggested paths lead directly to "overall

Clinton Voters

dd

”OOOVQU‘AMNd

-0.194
-0.034
-0.234
0.1 51
0.091
-0.1 70
0.179
-0.1 39
0.244

Bush Voters

dd

”OUOVOU'IéwN-l

-0.231
-0.031

-0.026
0.130
-0.142
0.012
0.050
-0.001
0.105

0.375
0.324
0.260
0.075
0.1 70
0.1 78
-0.1 34
-0.Z30
-0.237

01¢de

171

Table 23

Election Outcome Satisfaction Model

Correlation Matrices

3 4 5
0.494 - - - -
0.376 0.329

0.177 0.209 -0.003
0.036 0.004 -0.083
0.092 0.020 0.262
-0.061 -0.078 -0.360
-0.096 0.008 0.087
0.075 0.057 -0.1 06

3 4 5
0.467 - - - -
0.617 0.477

0.355 0.303 0.353
0.085 0.062 0.049
0.405 0.440 0.442
-0.363 -0.432 -0.459
-0.333 -0.112 ~0.175
-0.207 -0.194 -0.343

En eno Variabl
Clinton Disconfirmation
Bush Disconfirmation
Clinton Satisfaction
Bush Satisfaction
Overall Satisfaction

0.080 - - - -
-0.093 0.333 - - - -
0.135 -0.1 13 -0.827
-0.200 -0.372 -0.027
0.508 -0.230 -0.518

0.198 - - - -

0.507 0.160 - - - -
-0.282 -0.120 -0.786
-0.181 -0.623 -0.114

0.1 14 -0.056 -0.242

Exo e ous Var les
Clinton Attitude
Clinton Expectations
Clinton Desires
Bush Desires

dd

0.105
0.519

0.224
0.446

-0.178

10 11

0.069 - - - -

172

Table 24
Election Outcome Satisfaction Model
Within Group Indicators of Fit

 

Clinton Bush
Voters Voters
Goodness of Fit Index 0.938 0.927
Total Coefficient of Determination
for the Structural Equations 0.127 0.330
Squared Multiple Correlations for:
Clinton Disconfirmation 0.001 0.024
Bush Disconfirmation 0.073 0.064
Clinton Satisfaction 0.232 0.269
Bush Satisfaction 0.083 0.277
Overall Satisfaction 0.170 0.415
Modification lndices Over 5 twol fiveZ
Standardized Residuals:
less than -2 two3 six4
greater than 2 three5 five6
Root Mean Square Residual 2.851 4.729

 

1 Both of them are for structural paths: C. Desire -> Overall Sat. (5.007) and B. Desire -> Overall Sat. (9.948).

2 Three of them are for structural paflrs: B. Desire -> Overall Sat. (5.914), B. Expectation -> C. Sat. (12.049),
and B. Attitude -> Overall Satisfaction (6.201 ).

3 They are: B. Disconfirmation - Overall Sat. (-2.023) and B. Desires - Overall Sat. (-3.225).

4 They are: 8. Disconfirmation - B. Attitude (-2.235), B. Desires - C. Sat. (-3.225), B. Expectations - C. Sat
(-4.076), B. Attitude - 8. Sat. (-2.023), B. Desire - Overall Sat. (-3.27S), and B. Att - Overall Sat.
(-3.080).

5 They are: 8. Attitude - C. Sat. (2.431), C. Attitude - 8. Sat. (2.452), and C. Desire - Overall Sat. (2.274).

6 They are: B. Desire - C. Disconfirmation (2.014), C. Desire - B. Disconfirmation (2.489), C. Desire - 8. Sat.
(2.323), C. Attitude - Overall Sat. (2.123), C. Desire - Overal Sat. (2.815)

173

 

 

20-— -
184” -
16w ‘

Number of ._ .

Residuals
in each

Category

 

Bush Voter Group

Magnitude of Standardized Residuals

 

 

 

 

 

20in '
18" '
16" '
149- -

Number of 12 " '
Residuals 10-- -

in each
Category

C N
l r
I

#051:
111

 

Clinton Voter Group '

 

Magnitude of Standardized Residuals

 

 

Figure 17

Election Outcome Satisfaction Model
Distribution of Standardized Residuals

 

 

174

satisfaction" from antecedent variables. For Clinton voters, these paths are
Clinton Desire to Overall Satisfaction (5.007) and Bush Desire to Overall
Satisfaction (9.948). For Bush voters, these paths are Bush Desire to Overall
Satisfaction (5.914) and Bush Attitude to Overall Satisfaction (6.201).

Likewise, of the sixteen large standardized residuals, nearly half (seven) are
associated with overall satisfaction. For Clinton voters, these paths are Bush
Disconfirmation - Overall Satisfaction (-2.023), Bush Desire - Overall
Satisfaction (—3.225), and Clinton Desire - Overall Satisfaction (2.274). For Bush
voters, these paths are Bush Desire - Overall Satisfaction (-3.275), Bush Attitude
- Overall Satisfaction (-3.080), Clinton Attitude - Overall Satisfaction (2.123),
and Clinton Desire - Overall Satisfaction (2.815) . Of the remaining nine, seven
are "cross-over paths," that is, the residuals are associated with antecedent
variables of one candidate and post-election variables of the other candidate.
For Clinton voters, these paths are Bush Attitude - Clinton Satisfaction (2.431)
and Clinton Attitude - Bush Satisfaction (2.452). For Bush voters, these paths
are Bush Desire - Clinton Satisfaction (-3.225), Bush Expectation - Clinton
Satisfaction (-4.076), Bush Desire - Clinton Disconfirmation (2.014), Clinton
Desire - Bush Disconfirmation (2.489), and Desire Clinton - Bush Satisfaction

(2.323).

V-2.3. EOSM Two-Group Results Related to Research Hypothesis #1
The test of the research hypothesis that multiple targets are relevant is
very straight-forward. Paths from satisfaction with each candidate to overall
satisfaction were hypothesized to be positive for both Clinton and Bush voters.
Because path signs were hypothesized, all significance results reported in this

section are for one-tail tests.

175

Table 25 contains unstandardized path estimates and their associated
standard errors for Clinton versus Bush voters. Note that paths from
satisfaction with each candidate to overall satisfaction are significant for both
groups. For Clinton voters, the path estimate from Clinton Satisfaction to
Overall Satisfaction is .269 (significant at .05) and the path from Bush
Satisfaction to Overall Satisfaction is .141 (significant at .10). For Bush voters,
the path estimate from Clinton Satisfaction to Overall Satisfaction .554 and the
path estimate from Bush Satisfaction to Overall Satisfaction is .250 (both paths
significant at the .01 level).

The two paths of special interest are those which would pat be modeled
in the traditional satisfaction model, namely Clinton Satisfaction to Overall
Satisfaction for Bush voters and Bush Satisfaction to Overall Satisfaction for
Clinton voters. Note that among Bush voters, the path from Clinton
Satisfaction (i.e., the agr-choice) to Overall Satisfaction is of greater
magnitude than the path from Bush satisfaction (i.e., the choice) to Overall
Satisfaction. This difference was not hypothesized but is extremely interesting
because it clearly demonstrates the importance of satisfaction formation
processes relative to the non-choice among Bush voters. Among Clinton

voters, there is no significant difference between the two paths. Thus,

research hypothesis #1 is strongly supported by the EOSM two-group analysis,
as the path from satisfaction with the non-choice is as important (Clinton
voters) or more important (Bush voters) than is the path from satisfaction

with the choice.
It is possible to compare differences between Clinton voters and Bush
voters in terms of the relative magnitudes of these paths. To make such a

comparison, the common metric standardized paths must be compared (see

176

Table 25
Election Outcome Satisfaction Model
Unstandardized Estimates and Standard Errors of Structural Paths

 

Clinton Voters Bush Voters
Unstd. Std. Unstd. Std.
Estimates Errors Estimates Errors

Structural Path
To Overall Satisfaction from:

Clinton Satisfaction .269" * .1 16 .554" * * .098

Bush Satisfaction .141* .091 .250" * * .090
Clinton Disconfirmation —>

Clinton Satisfaction 376* * * .095 .009 .066

Bush Satisfaction .297* * .132 .066 .070
Bush Disconfirmation —>

Clinton Satisfaction -.101 .1 12 .288" * * .077

Bush Satisfaction .054 .1 58 .307’ * * .085
Clinton Desire -—>

Clinton Disconfirmation -.001 .009 .006 .010

Clinton Satisfaction .008 .007 .017* * * .007
Clinton Expectation —>

Clinton Disconfirmation -.005 .023 -.028* .019

Clinton Satisfaction .001 .018 -. 007 .012
Clinton Attitude —>

Clinton Satisfaction .003 .003 .003“ .001
Bush Desire —->

Bush Disconfirmation .016* * .009 -.004 .006

Bush Satisfaction ‘ -.01 8 .014 -.024* * * .006
Bush Expectation —>

Bush Disconfirmation -.040* " .021 -.031* * .014

Bush Satisfaction .025 .028 .014* .01 1
Bush Attitude —>

Bush Satisfaction .003 .004 .001 .002

 

* Signifcant at .10 for 1-tail tests
* * Significant at .05 for 1-tail tests
* * * Significant at .01 for 1-tail tests

177

Table 26). The common metric standardized solution produces parameter
estimates which have been standardized across the two groups so that direct
comparisons of path magnitude are possible. First, the Clinton Satisfaction to
Overall Satisfaction path for Bush voters is of greater magnitude than the
Clinton Satisfaction to Overall Satisfaction path for Clinton voters (difference
significant at .05 level, two tail test). Second, the Clinton Satisfaction to Overall
Satisfaction path for Bush voters is also significantly greater than the Bush
Satisfaction to Overall Satisfaction path for Clinton voters (difference
significant at the .01 level, two tail test). Third, the path magnitudes related to
each group's choice, i.e., Clinton Satisfaction to Overall Satisfaction for Clinton
voters and Bush Satisfaction to Overall Satisfaction for Bush voters, are equal
(difference not significant at .10 level, two tail test). Fourth, there is no
difference between Clinton voters and Bush voters with regard to the
magnitude of the path from Bush Satisfaction to Overall Satisfaction. From
this analysis, we can conclude that the single most important determinant of
Overall Satisfaction is Clinton Satisfaction among Bush voters, i.e., satisfaction

with a non-choice. Figure 18 and Figure 19 also present EOSM results.

V-2.4. EOSM Two-Group Results Related to Research Hypothesis #2
To test the research hypothesis that cross-over effects connect the
processes relating to both choice and non-choice, the information in Table 27
is once again examined. It was hypothesized that, in both groups, the cross-
over effects would take the form of significant paths from disconfirmation
with the choice to satisfaction with the non-choice, and from disconfirmation

with the non-choice to satisfaction with the choice.

178

Table 26
Election Outcome Satisfaction Model
Common Metric Standardized Solution

 

Path n.s.=nonsi nificant

Clinton Satisfaction —>0veral| Satisfaction
Bush Satisfaction —->Overall Satisfaction

Clinton Disconfirmation —>
Clinton Satisfaction
Bush Satisfaction

Bush Disconfirmation —>
Clinton Satisfaction
Bush Satisfaction

Clinton Desires —>
Clinton Disconfirmation
Clinton Satisfaction

Clinton Expectations -->
Clinton Disconfirmation
Clinton Satisfaction

Clinton Attitude —>
Clinton Satisfaction

Bush Desires —->
Bush Disconfirmation
Bush Satisfaction

Bush Expectations —>
Bush Disconfirmation
Bush Satisfaction

Bush Attitude —>
Bush Satisfaction

Clinton Voters

.263
.165

.474
.314

-.109 n.s.
.049 n.s.

-.017 n.s.
.116 n.s.

-.024 n.s.
.008 n.s.

.170 n.s.

.234
-.228 n.s.

-. 260
.147 n.s.

.119 n.s.

Bush Voters

.542
.293

.011 n.s.
.070 n.s.

.312
.278

.070 n.s.
.255

-.149
-.050 n.s.

.139

-.054 n.s.
-.31 S

-. 204
.086

.068 n.s.

 

179

      
  
   
  

  
  

Expectations

Satisfaction w/
re: Clinton

Clintons's
Performance

 

   

  
    
 
    

Clintons's
Performance

.263
(.116)

Desires re:
Clinton

 

 

 
     
 
        
   

 

-.109
(112)
Overall
Satisfaction
.314
(.132)
-.2 :
Desires re: 23‘ £014) .165
Bush (‘ 009) Disconfirmation ( . 09 1)
’ w/ Bush's
Performance . 9
-.26 ‘
.031) (J 1 Satisfaction
Expmtatioru .147 W/ Bush's
re: Bush (.028) Performance
.119
(.004)
Attitude
towards
Bush
Figure 18

Results: Election Outcome Satisfaction Model
Clinton Voters

The first nunber for each path is the common metric standardized solution parameter estimate, aid the
second nunber is the associated standard error. Simifieent paths mpec in bold.

180

 

 

 

 
   
   
   
 

 

  
   
   
     

 

 

 
   
      

 

   
   
   
 

 

 

.139
(.001) +
Expectations -.050 Satisfaction w/
re: Clinton | 0012) 011 Clintons's
' ' Performance
6019) ' confirmation w/ (.066)
Clintons's 5“
.070 Performance (.090
.010
Desires re: ( ) .
Clinton (-007) _312
(.017)
Overall
Satisfaction
.070
(.070)
Desires re: 3,,
Disconfirrnaticn (.09 o)
w/ Bush's
Performance ‘78
(.08 ) Satisfaction
w/ Bush's
Performance
Figure 19

Results: Election Outcome Satisfaction Model
Bush Voters

The first number for each path is the common metric standardized solution parameter estimate, aid the
second number is the associated standard error. Significant paths appear in bold.

181

Among Clinton voters, the path from Clinton Disconfirmation to Bush
Satisfaction (.297) is significant and positive as hypothesized (p < .05, one-tail
test); while the path from Bush Disconfirmation to Clinton Satisfaction is
negative as hypothesized, it is not of sufficient magnitude to be significant.
Among Bush voters, the path from Bush Disconfirmation to Clinton
Satisfaction (.288) is significant and positive (p < .01, one-tail test), as
hypothesized; the path from Clinton Disconfirmation to Bush Satisfaction is
neither significant nor negative. Thus. results from the EOSM provide
moderate support for research hypothesis #2. ag cross-over effects are foun_d_
from disconfirmation with choice to satisfaction with non-choice for both
groups.

To compare differences in path magnitudes between the two groups, we
again turn to the common metric standardized solution found in Table 26.
Among Clinton voters, the path from Clinton Disconfirmation to Bush
Satisfaction is .314, while among Bush voters, the path from Bush
Disconfirmation to Clinton Satisfaction is .312. There is no difference in the
magnitudes of these two paths. Thus, the cross-over effects appear to be of
equal size for both groups.

In addition to the cross-over effects evident in the structural paths,
errors between Clinton Disconfirmation and Bush Disconfirmation, and
between Clinton Satisfaction and Bush Satisfaction were allowed to covary.
These error covariances, along with the error variances associated with the
prediction of each of the endogenous variables is reported in the psi matrix
results, Table 27. Note that the two covariances along with each of the

variances are significant at the .01 level.

182

 

 

 

Table 27
Election Outcome Satisfaction Model
Psi Matrix Results
Unstd. Std. Std.
Estimate Error Estimate
Psi Matrix Elements (1)
(A) Clinton Supporters
Clinton Disconfirmation 1.967' * " 0.217 0.986
Bush Disconfirmation 1369* " ' 0.151 0.932
Clinton Satisfaction 1.206' * * 0.199 0.961
Bush Satisfaction 2.348" " * 0.388 1.311
Overall Satisfaction 1.169‘ ' * 0.193 0.895
Disconfirmation: Clinton - Bush -0.443* * * 0.184 -0.259
Satisfaction: Clinton - Bush 0743‘ ' ‘ 0.215 0.496
(8) Bush Sumrters
Clinton Disconfirmation 1.967' " * 0.217 0.986
Bush Disconfirmation 1.364' * * 0.151 0.932
Clinton Satisfaction 0.733" * * 0.108 0.584
Bush Satisfaction 0.849' " * 0.125 0.474
Overall Satisfaction 0.717* f " 0.106 0.549
Disconfirmation: Clinton - Bush -0.392* * 0.168 -0.229
Satisfaction: Clinton - Bush 0.221 * * * 0.086 0.148

 

(1)

* Signifcant at .10
* * Significant at .05
* * * Significant at .01

All other psi matrix elements were set to zero in both groups.

183

V—2.5. EOSM Two-Group Results Related to Research Hypothesis # 3

Using results of the two group analysis of EOSM, it is possible to perform
a weak test of research hypothesis #3, that dissatisfied people process more
extensively than do satisfied people. A stronger test of research hypothesis #3
using EOSM and a subsample of voters is provided in Section V.4.3.

As demonstrated in Table 21, Bush voters are less satisfied than Clinton
voters on all three satisfaction measures. According to research hypothesis
#3, then, one would expect to observe evidence of more extensive processing
among Bush voters relative to Clinton voters. The results presented here are a
"weak" test of the hypothesis because the samples are confounded: some
Clinton voters were not satisfied and some Bush voters were not dissatisfied. In
this analysis, amount of processing is indicated by the number of significant
paths connecting variables measured pre-election and variables measured
post-election (hereafter called "pre-post" paths) in the EOSM two—group
analysis.

From Table 25, we see that two such pre-post paths are significant for
Clinton voters: Bush Desire to Bush Disconfirmation and Bush Expectation to
Bush Disconfirmation. Both paths are significant at the .05 level (one-tail
test). On the other hand, six pre-post paths are significant for Bush voters:
Clinton Desire to Clinton Satisfaction, Clinton Expectation to Clinton
Disconfirmation, Clinton Attitude to Clinton Satisfaction, Bush Desire to Bush
Satisfaction, Bush Expectation to Bush Disconfirmation, and Bush Expectation
to Bush Satisfaction. We conclude that Bush voters did exhibit more extensive
processing than did Clinton voters. Therefore, research hypothesis #3 is

supported.

184

One additional observation is that throughout Table 25, Bush voters in
the sample exhibited smaller standard deviations for all but two of the paths
than did Clinton voters in the sample. The two exceptions are the Clinton
Desire to Clinton Satisfaction path for which the standard errors are equal, and
the Clinton Desire to Clinton Disconfirmation path for which Clinton voters

had a standard error of .009 while Bush voters had a standard error of .010.

V-2.6. EOSM Two-Group Results Related to Research Hypothesis #4

Finally, to address the issue raised by research hypothesis #4, the
relative impact of desires versus expectations on satisfaction, Table 28 is
presented. It shows the total direct and indirect effects of expectations and
desires on satisfaction with each candidate and on overall satisfaction. We see
that only one expectations - endogenous variable path has significant total
effects (Bush Expectation to Clinton Satisfaction). However, four desires -
endogenous variable paths have significant total effects: Clinton Desire to
Clinton Satisfaction and to Overall Satisfaction, and Bush Desire to Bush
Satisfaction and to Overall Satisfaction. Therefore. research hypothesis #4 is
supported.

Parenthetically, it is worth noting that all five of the significant paths
from expectations or desires to endogenous variables occurred among Bush
voters (while none occurred among Clinton voters). This is further evidence
in support of research hypothesis #3, that dissatisfied people process more

extensively than do satisfied people.

185

Table 28
Election Outcome Satisfaction Model
Total Effects of Expectations versus Desires

 

Clinton Voters Bush Voters
Total Std. Total Std.
Effects Errors Effects Errors
For Clinton
Expectation ———>
Clinton Satisfaction -.003 .020 -.008 .018
Bush Satisfaction -.001 .001 -.002 .004
Overall Satisfaction -.001 .006 -.005 .010
Desire —-—>
Clinton Satisfaction .007 .008 .017*** .007
Bush Satisfaction .000 .003 .000 .001
Overall Satisfaction .002 .002 .010"** .004
For Bush
Expectation —>
Clinton Satisfaction .004 .005 -.009* .007
Bush Satisfaction .023 .027 .005 .028
Overall Satisfaction .004 .005 .004 .009
Desire ——->
Clinton Satisfaction -.002 .002 -.001 .003
Bush Satisfaction -.017 .014 -.025*" .015
Overall Satisfaction -.003 .003 -.007*" .005

 

* Signifcant at .10 for 1-tail tests
* * Significant at .05 for 1-tail tests
* * * Significant at .01 for 1-tail tests

186

V-2.7. Summary of EOSM Results

To summarize, the EOSM appears to fit the data relatively well overall.
In the Clinton voter group, there were fewer high modification indices or
standardized residuals than in the Bush voter group. However, for the Bush
voter group, the proportion of variance explained for each of the endogenous
variables was greater than in the Clinton voter group. Most of the large
modification indices and standardized residuals could have been eliminated
had direct structural paths between antecedent variables and overall
satisfaction been specified.

All four research hypotheses were supported to varying degrees.
Strong support was found for research hypothesis #1, that multiple targets of
satisfaction are relevant in the overall satisfaction formation process.
Moderate support was found for research hypothesis #2, that the processing
paths associated with each target are structurally related via cross-over paths.
Moderate support was found for research hypothesis #3, that dissatisfied
voters process more extensively than do satisfied voters. Finally, strong
support was found for research hypothesis #4, that desires have greater total

effects on satisfaction formation than do expectations.

V-3. Results of the Campaign Process Satisfaction Model

Results of the two-group analysis using the Campaign Process
Satisfaction Model (CPSM) are presented in this section. First, differences
between the two groups with respect to the variables included in the model are
highlighted. Then, results of the LISREL analysis are presented. Next, results
associated with research hypothesis #1 are described. Finally, additional

observations about the model results are presented.

187

V-3.l. Differences Between the Two Groups (CPSM)

Although research questions specifically address the relationships
between key variables, means and standard deviations of model variables are
presented in Table 29 for completeness and to highlight differences between
the groups.

Note that Clinton and Bush voters differ significantly on most variables
included in the model. Clinton and Bush voters had similar expectations
regarding Clinton's running an unfair campaign, and both groups made
similar disconfirmation judgments regarding Clinton's performance on both
campaign elements. The two groups differed on all measures regarding Bush's
campaign except on the perceived performance of Bush's advertising

effectiveness.

V-3.2. Results of the CPSM Two-Group LISREL Analysis

This section contains results of the Campaign Process Satisfaction Model
(CPSM) two-group LISREL analysis. First, the correlation matrix for each
group is presented. Then, overall and group-specific indicators of fit are
provided.

The first half of Table 30 contains the CPSM correlation matrix for
Clinton voters. Note that levels of significance appear below the correlations.
The second half of Table 30 presents results for Bush voters in a similar
format.

Overall, the model fit is acceptable. The appropriate overall fit index is
chi square goodness-of-fit statistic. For this model, the chi square value was

117.63 with 72 degrees of freedom (p = .001)

188
Table 29
Campaign Process Satisfaction Model
Means and Standard Deviations of Each Group

 

Clinton Bush p-values
Voters Voters from
Mean Std. Dev. Mean Std. Dev. t-testsl

For ClintonZ:
Expectations: ad. effectiveness 7.20 1.67 . 6.49 1.80 .010
Performance: ad. effectiveness 7.01 2.11 6.27 2.03 .020
Disconfirmation: ad. effectiveness 6.29 2.09 6.18 2.02 .707
Expectations: unfair campaign 4.70 2.26 5.26 2.65 .141
Performance: unfair campaign 3.84 2.38 4.63 2.96 .054
Disconfirmation: unfair campaign 4.55 2.17 4.61 2.20 .854
For Bush:
Expectations: ad. effectiveness 5.55 2.02 6.53 1.84 .001
Performance: ad. effectiveness 4.26 2.33 4.69 2.20 .212
Disconfirmation: ad. effectiveness 4.17 2.06 3.54 1.96 .042
Expectations: unfair campaim 6.18 2.50 3.16 2.12 .000
Performance: unfair campaign 6.08 2.63 2.84 2.29 .000
Disconfirmation: unfair campaign 5.50 2.70 4.36 2.67 .000
Overall3:
Disconfirmation: overall process 4.50 1.43 3.68 1.33 .000
Satisfaction: overall process 5.80 1.63 . 3.40 1.62 .001

 

1 p-values for t-test for Ho: u1- ”2-
2 10-p0int scales.
3 7-point scales.

189

Table 30
Campaign Process Satisfaction Model
Correlation Matrices

Clinton Voters

1

2

10

11

12

13

14

1.000

-0.203 1.000
(0.079)

0.046 0.276 1.000
(0.691) (0.016)

0.093 0.224 0.069 1.000
(0.426) (0.053) (0.553)

0.302 ~0.269 0.086 -0.375 1.000
(0.007) (0.019) (0.454) (0.001)

0.076 -0.142 -0.061 -0.494 0.615 1.000
(0.508) (0.220) (0.595) (0.000) (0.000)

0.004 0.122 0.070 -0.005 0.281 0.374 1.000
(0.973) (0.302) (0.547) (0.969) (0.014) (0.001)

0.461 -0.062 -0.003 0.028 0.093 0.248 0.303 1.000
(0.000) (0.592) (0.976) (0.812) (0.421) (0.029) (0.008)

0.017 0.369 0.043 0.266 -0.200 -0.415 0.122 -0.161 1.000
(0.885) (0.001) (0.710) (0.022) (0.083) (0.000) (0.295) (0.164)

-0.198 0.587 -0.006 0.130 -0.252 -0.270 0.062 -0.294 0.468 1.000
(0.087) (0.000) (0.958) (0.263) (0.028) (0.018) (0.601) (0.010) (0.000)

-0.308 0.027 0.215 -0.064 0.012 -0.082 0.386 -0.060 0.180 0.009 1.000
(0.007) (0.821) (0.062) (0.590) (0.920) (0.483) (0.001) (0.605) (0.119) (0.937)

-0.200 0.235 0.442 -0.075 -0.093 -0.069 0.090 0.134 -0.034 0.087 0.348 1.000
(0.080) (0.041) (0.000) (0.520) (0.416) (0.546) (0.439) (0.243) (0.770) (0.453) (0.002)

0.234 0.052 0.009 0.155 0.023 -0.116 0.157 0.081 0.506 0.164 -0.161 -0.076 1.000
(0.042) (0.659) (0.940) (0.187) (0.844) (0.317) (0.175) (0.489) (0.000) (0.116) (0.165) (0.501)

-0.056 0.350 0.116 0.644 -0.322 ~0.290 0.083 0.028 0.259 0.258 -0.086 0.102 0.318 1.000
(0.626) (0.002) (0.316) (0.000) (0.004) (0.011) (0.482) (0.809) (0.025) (0.024) (0.462) (0.379) (0.005)

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4
W Wrinkles
1 Dbconfimtlon re: Chton Advertbhg Effectiveness 7 Expectation rs: Cliton Ad. Effectivursss
2 Disconfirnstion rs: Chton Unfas Canpsbnhg 8 Performance re: Chton Ad. Effectiveness
3 Disconfirmation rs: Bush Advertlshg Effectiveness 9 Expectation re: Chton Unfair Campaignhg
4 Disconftmation re: Bush Unfair Canpsignirg 10 Performance re: Cirton Unfas Cannabnhg
5 Overal Disconfirmation 11 Expectation rs: Bush Ad. Effectivenes
6 Overal Satisfaction with Certain Process 12 Performance re: Bush Ad. Effectivensu

13 Expectation rs: Bush Unfa's Canvabnhg
14 Performnce rs: Bush Unfak Carmalgnhg

Bush

11

12

13

14

190

Table 30, continued

Voters

1300

-0.062 1.000
(05531

-0.033 0.115 1.000

(0.753) (0.266)

-0.061 0.259 0.080 1.000
(0.558) (0.011) (0.443)

-0.113 -0.132 0.112 -0.051 1.000
(0.272) (0.200) (0.277) (0.623)

-0.077 -0.117 -0.012 -0.021 0.708 1.000
(0.454) (0.255) (0.907) (0.843) (0.000)

0.070 0.036 -0.151 0.114 0.239 0.227 1.000
(0.496) (0.725) (0.143) (0.270) (0.018) (0.026)

0.567 0.062 -0.133 -0.005 -0.102 -0.038 0.149 1.000
(0.000) (0.549) (0.197) (0.965) (0.323) (0.715) (0.149)

-0.057 0.219 -0.090 0.141 -0.014 -0.152 -0.110 0.135 1.000
(0.584) (0.034) (0.387) (0.175) (0.894) (0.141) (0.289) (0.193)

-0.007 0.351 -0.055 0.263 -0.200 -0.250 -0.065 0.147 0.497 1.000
(0.949) (0.001) (0.596) (0.010) (0.054) (0.014) (0.529) (0.152) (0.000)

0.209 0.081 -0.015 0.002 -0.045 0.085 0.348 0.301 0.097 0.115 1.000
(0.041) (0.432) (0.884) (0.985) (0.659) (0.405) (0.001) (0.003) (0.349) (0.263)

0.112 0.082 0.444 0.181 0.061 0.146 0.086 0.076 -0.079 0.033 0.211 1.000
(0.275) (0.432) (0.000) (0.079) (0.555) (0.156) (0.406) (0.463) (0.450) (0.749) (0.039)

-0.041 0.048 -0.117 0.274 0.096 -0.017 -0.003 0.049 0.302 0.166 -0.165 -0.264 1.000
(0.698) (0.643) (0.263) (0.008) (0.356) (0.871) (0.980) (0.642) (0.003) (0.108) (0.111) (0.010)

-0.119 0.144 0.087 0.531 0.051 -0.004 0.007 -0.125 0.114 0.408 -0.021 0.006 0.389 1.000
(0.247) (0.162) (0.398) (0.000) (0.622) (0.966) (0.943) (0.224) (0.273) (0.000) (0.937) (0.952) (0.000)

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 14
W transmissible:
1 Disconfirmtion re: Chton Advertissrg Effectiveness 7 Expectation rs:CI1tor1 Ad. Effectiveness
2 Disconfirnation rs: Chton Unfak Cambnhg 8 Performance re: Clinton Ad. Effectiveness
3 Disconfimation re: 8M Advertlshg Effectiveness 9 Expectation re: Chton Unfa‘r Canvaignhg
4 Disconfirmation re: Bush Unfai Cannabnhg 10 Performance rs: Chton Unfai Cambnhg
5 Overal Disconfimation 11 Expectation re: Bush Ad. Effectivenes
6 Overal Satisfaction with Canvabn Process 12 Performance re: Bush Ad. Effectiveness

13 Expectation rs: Bush Unfak Canneignhg
14 Performance re: Bush Unfas Carmslgnhg

 

 

 

191

Group specific indicators of fit are provided in Table 31. The goodness of
fit (GFI) index for Clinton and Bush voters are .893 and .945, respectively.
Squared multiple correlations for endogenous variables in' the Clinton voter
group range from .224 to .531, and the total coefficient of determination of the
structural equations is 0.866. In general, these statistics are slightly lower for
the Bush voter group: squared multiple correlations range from .049 to .546,
and the total coefficient of determination for the structural equation is .697.

There are six modification indices greater than five for the Clinton
voter group and two for the Bush voter group. A total of eleven standardized
residuals lie outside the -2.0 to 2.0 range for the Clinton voters, while seven lie
outside that range for the Bush voters. Although there are several
standardized residuals outside the "acceptable" range in each group, an
examination of their distribution indicates they are roughly normally
distributed, with the exception of one outlier for the Clinton group (see Figure
20). The outlying residual and the large number of high modification indices
and residuals indicate that the model may be misspecified for the Clinton

group.

V-3.3. CPSM Two-Group Results Related to Research Hypothesis #1
The research hypothesis that multiple targets are relevant to the
formation of overall satisfaction may be tested by examining the significant
paths in the satisfaction formation process for Clinton and Bush voters. Of
particular interest is the number of significant paths relating to candidate
Bush among Clinton voters and relating to candidate Clinton among Bush
voters. These are paths which would not be modeled in traditional satisfaction

models.

192

Table 31

Campaign Process Satisfaction Model
Within Group Indicators of Fit

 

Goodness of Fit Index

Total Coefficient of Determination
for the Structural Equations

Squared Multiple Correlations for:
Disconfirmation: Clin. Ad. Effectiveness
Disconfirmation: Clin. Unfair Campaign
Disconfirmation: Bush Ad. Effectiveness
Disconfirmation: Bush Unfair Campaign
Overall Process Disconfirmation
Overall Process Satisfaction

Modification lndices Over 5

Standardized Residuals:
less than -2
greater than 2

Root Mean Square Residual

Clinton Bush
Voters Voters
0.893 0.945
0.866 0.697
0.262 0.326
0.354 0.119
0.224 0.215
0.419 0.292
0.291 0.049
0.531 0.546
sixl twoz
six3 three4
five5 fourG
0.367 0.230

 

1 Five were for structural paths: BADDIS -> CADDIS (5.780), BUCDIS -> PROCSAT (12.372), PROCSAT
->BUCDIS (9.736), CADEX -> OVERDIS (9.114), and BADEX -> CADDIS (6.602).

(”N

They were: CADEX -> OVERDIS (7.637) and BADPER -> BUCDIS (5.069).
They were: CADDIS - BADDIS (-3.421), CADDIS - BUCDIS (-3.938), CADDIS - PROCSAT (-3.748),

CADDIS - BADEX (-2.206), CADDIS - BADPER (-2.209), and BUCDIS - PROCSAT (-1 2.022).

(ll-b

OVERDIS - CADEX (2.600), PROCSAT - CADEX (2.564).

a)

PROCSAT - BUCEX (2.488).

ABREVIATIGNIS: CADEX Expectation re: Clinton Ad. Effectiveness

BADEX

CADPER Performance rs: Clinton Ad. Effectiveness BADPER
CADDIS Disconfirmtion re: Clinton Ad. Effectivensss BADDIS

CUCEX Expectation re: Clinton Unfair Campaign
CUCPER Performs re: Clinton Unfair Corrosion

BIKZEX
CUCPER

CUCOIS Disconfirrmtion re: Clinton Unfair Carrpaign BUCDIS

They were: CADDIS - BADDIS (-5.412), BADDIS - PROCSAT (-3.065), and PROCSAT - CUCPER (-3.288).
They were: CUCDIS - BUCDIS (4.393), CUCDIS - PROCSAT (3.446), OVERDIS - PROCSAT (2.569),

They were: BADDIS - BUCDIS (3.124), OVERDIS - CADEX (2.374), PROCSAT - CADEX (3.950),

Expectation re: 81-h Ad. Effectiveness
Perforrnancs rs: Bush Ad. Effectiveness
Disconfirmation rs: Bmh Ad. Effectiveness
Expectation rs: Bush Unfair Carrpaign
Performance re: Bah Unfair Carmaign
Disconfirmation re: Bush Unfair Campaign

OVERDIS Overall Disconflrrr-tion rs: overall process
PROCSAT Overall Process Satisfaction

193

 

 

Clinton Voter Group

30-- -
27-— -
24L -

21 «i- -
Number of 18.- .
Residuals
in each

Category 12" ‘

 

 

    

\\\\\\\ \\\\\\\‘
-40r '31!) -2to '1“) a9” Oto 1w Zto 3m 4or
less -3.9 -2.9 -1.9 0 .9 1.9 2.9 3.9 greater

Magnitude of Standardized Residuals*

 

*NOTE: Value of residual in '-4 or less' category is -1 2.02.

 

 

Bush Voter Group
30-- '
27+ -
244k -

    
 

Number of 21
Residuals 18* '
in each ‘5“ '
Category 124- ~

9r -

 

 

. \ a
-40r -3to -2to -lto -.9to Ono 1to 2m 3to 4or
less -3.9 -2.9 -l .9 0 .9 1 .9 2.9 3.9 greater

Magnitude of Standardized Residuals

 

Figure 20
Campaign Process Satisfaction Model
Distribution of Residuals

 

 

194

Table 32 contains unstandardized path estimates and their associated
standard errors for Clinton versus Bush voters. Error variances associated
with the prediction of each of the endogenous variables is reported in Table
33. In addition, errors were expected to covary among the disconfirmation
measures associated with each of the campaign elements. Therefore, the
values of these covariances are also included in the table. Note that all the
variances in both groups are significant at the .01 level, while 4 of the
covariances among Clinton voters and 1 among Bush voters are significant.

Among Clinton voters, two paths from antecedent variables regarding
Bush have a direct effect on Overall Satisfaction with the campaign process:
Expectations of Bush's Advertising (p < .01, one tail test) and Bush's Unfair
Campaign Performance (p < .10, one tail test). Further, among Clinton voters,
both paths from disconfirmation regarding Bush's campaign performance
have direct effects on overall disconfirmation: Bush Advertising
Disconfirmation (p < .10, one tail test) and Bush Unfair Campaign
Disconfirmation (p < .01, one tail test). Thus, four of six potential paths to
overall measures from Bush-related measures are significant for Clinton
voters. Among Bush voters, one path from an antecedent variable regarding
Clinton has a direct impact on Overall Satisfaction: Expectations of Clinton's
Unfair Campaign (p < .10, one tail test). In addition, the path from
Disconfirmation of Clinton's Unfair Campaign to Overall Disconfirmation is
significant (p < .10, one tail test). Therefore, two of six potential paths to

overall measures from Clinton-related measures are significant for Bush

voters. We conclude that research hxmthesis #1 receives strong supmrt from

 

the CPSM analysis among Clinton voters, and moderate support among Bush

 

voters.

195

Table 32
Campaign Process Satisfaction Model
Unstandardized Estimates and Standard Errors of Structural Paths

 

 

Clinton Voters Bush Voters
Unstd. Std. Unstd. Std.
Estimates Errors Estimates Errors

Structural Path
To Overall Satisfaction from:

Overall Disconfirmation .468*** .093 .847*** .086

Expectation: C. Ad. Effectiveness .363‘” .091 -.007 .069

Performance: C. Ad. Effectiveness .033 .070 .018 .062

Expectation: C. Unfair Campaign -.206*** .077 -.069" .054

Performance: C. Unfair Campaign -.005 .065 -.045 .050

Expectation: 8. Ad. Effectiveness -.155* * .082 .104" .073

Performance: 8. Ad. Effectiveness -.002 .061 .052 .056

Expectation: B. Unfair Campaign -.022 .064 -.003 .067

Performance: 8. Unfair Campaign -.075* .054 .009 .062
To Overgfl Disconfirmation from:

Disconfirmation: C. Ad. Effectiveness .204*** .072 -.079 .068

Disconfirmation: C. Unfair Campaign -.111* .070 -.089* .065

Disconfirmation: 8. Ad. Effectiveness .101" .072 .086 .070

Disconfirmation: B. Unfair Campaign -.200*"* .053 -.018 .062
To Disconfirmation: C. Ad. Effectiveness from:

Expectation: C. Ad. Effectiveness -.155 .124 -.005 .097

Performance: C. Ad. Effectiveness 542*“ .098 .570*** .087
To Disconfirmation: C. Unfair Campaign from:

Expectation: C. Unfair Campaign .140" .095 .027 .091

Performance: C. Unfair Campaign .47?” .091 242*“ .081
To Disconfirmation: 8. Ad. Effectiveness from:

Expectation: 8. Ad. Effectiveness .153" .106 -.128* .101

Performance: 8. Ad. Effectiveness 362*” .091 .423*** .084
To Disconfirmation: B. Unfair Campaign from:

Expectation: B. Unfair Campaign -.110 .101 .080 .101

Performance: 8. Unfair Campaign .696"** .096 .502*** .093

 

* Signifmnt at .10 for l-tail tests
* * Significant at .05 for l-tali tests
* * * Significant at .01 for 1-tail tests

196

Table 33
Campaign Process Satisfaction Model
Psi Matrix Results

 

Unstd. Std. Std.
Estimate Error Estimate

Psi Matrix Elements (1)

(A) Clinton Voters

 

Disconfirmation: C. Ad. Effectiveness 3.358'” 0.558 0.782
Disconfirmation: C. Unfair Campaign 3.098' * ' 0.515 0.647
Disconfirmation: 8. Ad. Effectiveness 3.421‘ * ' 0.568 0.834
Disconfirmation: B. Unfair Campaign 4300' ' ' 0.716 0.699
Overall Disconfirmation 1.451 * " ' 0.241 0.764
Overall Satisfaction 1.123‘ * * 0.187 0.445
C. Ad. Effect. - C. Unfair Camp. -0.679* ' 0.387 -0.150
C. Ad. Effect. - 8. Ad. Effect. 0.852" * 0.411 0.203
C. Ad. Effect. - 8. Unfair Camp. 0794' * 0.456 0.154
C. Unfair Camp.- 8. Ad. Effect. 0.890* ' " 0.396 0.201
C. Unfair Camp.- 8. Unfair Camp -0.118 0.429 -0.022
8. Ad. Effect. - B. Unfair Camp. 0.319 0.452 0.064
(8) Bu_sh Voteg
Disconfirmation: C. Ad. Effectiveness 2.755* * ' 0.409 0.642
Disconfirmation: C. Unfair Campaign 4222" ' * 0.628 0.881
Disconfirmation: 8. Ad. Effectiveness 3.029’ " 0.450 0.738
Disconfirmation: 8. Unfair Campaign 3.653’ * * 0.543 0.593
Overall Disconfirmation 1.695‘ * * 0.252 0.892
Overall Satisfaction 1.193' ' * 0.177 0.473
C. Ad. Effect. - C. Unfair Camp. -0.270 0.360 -0.059
C. Ad. Effect. - 8. Ad. Effect. 0.057 0.304 0.013
C. Ad. Effect. - B. Unfair Camp. -0.128 0.334 -0.025
C. Unfair Camp.- 8. Ad. Effect. 0.448 0.379 0.101
C. Unfair Camp.- 8. Unfair Camp 0823* * 0.422 0.151
8. Ad. Effect. - 8. Unfair Camp. -0.199 0.350 -0.040

 

(1 ) All other psi matrix eiernents were set to zero in both groups.

* Signifcant at .10
* * Significant at .05
* * * Significant at .01

101'
the
C11]
all<

prc
Cli

V0

of

f0-

A.

CG

C211

197

Not only are more paths to overall satisfaction significant for Clinton
voters than for Bush voters, for those paths which are significant in each of
the two groups, the absolute value of each significant coefficient is larger for
Clinton versus Bush voters. The common metric standardized solution,
allowing for direct comparison of path magnitudes between the two groups, is
provided in Table 34, Figure 21, and Figure 22. The path from Expectations of
Clinton's Unfair Campaign to Overall Satisfaction has a common metric
solution value of -.323 among Clinton voters versus a value of -.108 for Bush
voters (significant difference at p < .01, 2-tail test). The path from Expectations
of Bush's Advertising to Overall Satisfaction is -.188 for Clinton voters and .126
for Bush voters (significant difference at p < .01, 2-tail test).

In addition, when the expectations - overall satisfaction link is
compared for each of the campaign elements in terms of choice versus non-
choice candidates, Clinton voters had larger path coefficients in all three of
the cases for which differences were significant. For Expectations of
Advertising Effectiveness of choice candidate, Clinton voters had a common
metric standardized path estimate of .399 with regard to Clinton while Bush
voters had a path of .126 with regard to Bush. (significant difference at p < .01,
2-tail test). For Expectations of Unfair Campaign by choice, Clinton voters had
a path coefficient of -.323 for Clinton while Bush voters had a path coefficient
of -.004 (significant difference at p < .01). With regard to Expectation of
Advertising Effectiveness for the non-choice, Clinton voters had a path
coefficient of -.188 while Bush voters had a path coefficient of -.088
(significant difference at p < .05, 2-tail test). Expectations regarding unfair

campaign are equal for the two groups.

I‘P

T.

To

198
Table 34
Campaign Process Satisfaction Model
Common Metric Standardized Solution

 

Clinton Voters Bush Voters

Path n.s.=nonsi nificant
To Overall Satisfaction from:

Overall Disconfirmation .406*** .735**"

Expectation: C. Ad. Effectiveness 399*” -.008

Performance: C. Ad. Effectiveness .042 .023

Expectation: C. Unfair Campaign -.323*** -.108*

Performance: C. Unfair Campaign -.008 -.077

Expectation: 8. Ad. Effectiveness -.188* * .126*

Performance: 8. Ad. Effectiveness -.003 .074

Expectation: 8. Unfair Campaign -.032 -.004

Performance: 8. Unfair Campaign -.116* .014
To Overall Disconfirmation from:

Disconfirmation: C. Ad. Effectiveness .306*** -.119

Disconfirmation: C. Unfair Campaign -.177* -.141*

Disconfirmation: 8. Ad. Effectiveness .149* .127

Disconfirmation: B. Unfair Campaign -.361*** -.032
To Disconfirmation: C. Ad. Effectiveness from:

Expectation: C. Ad. Effectiveness .131 -.004

Performance: C. Ad. Effectiveness .540*** .568***
To Disconfirmation: C. Unfair Campaign from:

Expectation: C. Unfair Campaign .160* .030

Performance: C. Unfair Campaign .S86*** 300*”
To Disconfirmation: 8. Ad. Effectiveness from:

Expectation: 8. Ad. Effectiveness .145* -.121*

Performance: 8. Ad. Effectiveness .403*** .474***
To Disconfirmation: B. Unfair Campaign from:

Expectation: B. Unfair Campaign -.101 .074

Performance: 8. Unfair Campaign .687*** .495”*

* Signifcmt at .10 for l-tall tests
* ' Significant at .05 for 1-tail tests
* * ' Significant at .01 for l-tail tests

199

.399
(.091)

 

 

 

 

 

 

 

 

 

Figure 21
Results: Campaign Process Satisfaction Model
Clinton Voters

The first number for each path is the common metric standardized solution parameter estimate, and the second
number is the associated standard error. Significant paths appear in bold.

200

 

 

 

 

 

 

..008
(.069)
.023
( 062)
< 668)
Expecteu'on: . .191
( 054)
..1 4 1
(.065)
-.077
(.055')\
Overall
Overall Satisfaction

Disconfirmation .oa‘ “them”

 

 

 

 

 

Figure 22
Results: Campaign Process Satisfaction Model
Bush Voters

The first number for each path is the common metric standardized solution parameter estimate, and the second
number is the associated standard error. Significant paths appear in bold.

201

Further, among Clinton voters, there are significant paths from Bush
Advertising Performance to Bush's Advertising Disconfirmation (p < .01), from
Bush's Unfair Campaign Performance to Bush's Unfair Campaign
Disconfirmation (p < .01) and from Bush's Expectation Advertising to Bush's
Advertising Disconfirmation (p < .10). Among Bush voters, there are
significant paths from Clinton Advertising Performance to Clinton's
Advertising Disconfirmation (p < .001) and Clinton's Unfair Campaign
Performance to Clinton's Unfair Campaign Disconfirmation (p < .001). Ihpg
paths provide additional support for research hypothesis #1 by illustrating

that each group processes antecedents related to their non-choice with respect

 

to specific campaign elements.

V-3.4. Additional Observations from the CPSM Two-Group Analysis
In addition to results provided to test research hypothesis #1, the CPSM
yielded two results not related to any research hypothesis but which are
interesting because they are incongruent with previous satisfaction research
findings:
1) The performance - satisfaction path is significant only for Clinton
voters and only with regard to Bush's performance with respect to running an
unfair campaign. The remaining seven possible performance - satisfaction
paths are not significant. This finding is especially surprising given the
strong support in the literature for the performance - satisfaction
relationship.
2) For Clinton voters, the expectation - disconfirmation path for candidate
Clinton is significant and positive with regard to unfair campaigns. The

expectation - disconfirmation path for Bush's advertising effectiveness is

202

positive for Clinton voters. These findings are surprising because positive
relationships between expectation and disconfirmation are uncommon
empirically and are counter-intuitive. The positive relationship indicates that
as expectations rise, the perception that performance exceeds expectations also
rises.

Both these findings will be discussed in detail in the next chapter.

V-4. Results of the EOSM LISREL Analysis of Satisfied Clinton

Voters and Dissatisfied Bush Voters

In order to test the research hypothesis that satisfied and dissatisfied
subjects process differently, the EOSM was reanalyzed using only satisfied
Clinton voters (n=70) and dissatisfied Bush voters (n=88). Under ideal
circumstances, Clinton voters would have been divided into satisfied versus
dissatisfied groups, and Bush voters would have been divided into satisfied and
dissatisfied groups, and two separate 2-group analysis would have been
performed. However, small sample sizes of dissatisfied Clinton voters (n =6)
and satisfied Bush voters (n = 8) precluded such analysis. Comparing satisfied
Clinton and dissatisfied Bush voter results is a relatively weak analysis because
differences between the two groups may be due to factors relating to voting
behavior rather than level of satisfaction. Differences between the entire
Clinton voter sample and satisfied Clinton voter sample is of more interest as
the effects of removing dissatisfied voters can be examined. Similarly,
differences between the entire Bush voter sample and dissatisfied Bush voter
sample is of interest as the effects of removing satisfied voters can be
examined. Results of such an analysis may provide insight into the question of

differential processing based on level of satisfaction by examining the

203

changes in structural paths which result when the respective samples are
made more homogeneous. Therefore, results will be presented in a manner
which will facilitate such comparisons.

In order to eliminate the possibility that paths would "disappear" simply
because of a reduction in sample size, the analysis was performed with the
same sample sizes specified as were used in the original EOSM analysis reported
in Section v.2.

Results of the analyses are presented in three tables. In the first, Table
35, unstandardized path estimates and standard errors of the satisfied Clinton
voters are presented next to the same information for all Clinton voters. In
the next table (Table 36) similar information is provided for dissatisfied Bush
voters and all Bush voters. Finally, in Table 37, summary information drawn
from the previous two tables is presented. Results of this analysis which relate
to research hypothesis #3 are presented in the next section. In addition, it is
possible to reexamine research hypothesis #1 and research hypothesis #2
using results from these more homogeneous samples; results of these analyses

appear in Sections V.4.2 and V.4.3, respectively.

V—4.1. Analysis of EOSM with Satisfied and Dissatisfied Voters -
Results Related to Research Hypothesis #3
If Research hypothesis #3 is true, that is, if satisfied people process less
extensively than dissatisfied people, one would expect fewer significant paths
from pre-election variables to post-election variables, hereafter referred to as
"pre - post paths," for the satisfied Clinton group relative to the total Clinton
Group. This is because the total Clinton group includes some dissatisfied

Clinton voters. In fact, the reverse is true: three such paths are significant

204
Table 35
Election Outcome Satisfaction Model
Unstandardized Estimates and Standard Errors of Structural Paths
Satisfied Clinton Voters Versus All Clinton Voters

 

Satisfied Clinton Voters Clinton Voters

Unstd. Std. Unstd. Std.
Estimates Errors Estimates Errors

Structural Path
To Overall Satisfaction from:

Clinton Satisfaction .21 9* * * .082 .269* * .1 1 6

Bush Satisfaction .109* * * .061 .141* .091
Clinton Disconfirmation —>

Clinton Satisfaction .368* * * .086 .376* * * .095

Bush Satisfaction .368" * * .127 .297* * .132
Bush Disconfirmation —>

Clinton Satisfaction -.074 .104 -.101 .112

Bush Satisfaction .044 .154 .054 .1 58
Clinton Desires —>

Clinton Disconfirmation .007 .010 -.001 .009

Clinton Satisfaction -.011* .007 .008 .007
Clinton Expectations —>

Clinton Disconfirmation .002 .025 -.005 .023

Clinton Satisfaction -.008 .017 .001 .018
Clinton Attitude —>

Clinton Satisfaction .005* * .003 .003 .003
Bush Desires —>

Bush Disconfirmation .01 1 .010 .016* .009

Bush Satisfaction -.022* .01 5 -.01 8 .007
Bush Expectations —>

Bush Disconfirmation -.016 .022 -.040* .021

Bush Satisfaction .031 .029 .025 .028
Bush Attitude —’

Bush Satisfaction .003 .003 .003 .004

 

* Signifcant at .10, 1-tail test
* * Significant at .05, 1-tail test
* * * Significant at .01, 1-taii test

205

Table 36
Election Outcome Satisfaction Model
Unstandardized Estimates and Standard Errors of Structural Paths
Dissatisfied Bush Voters Versus All Bush Voters

 

Dissatisfied Bush Voters Bush Voters

Unstd. Std. Unstd. Std.
Estimates Errors Estimates Errors

Structural Path
To Overall Satisfaction from:

Clinton Satisfaction .41S* * * .063 .554* * * .098

Bush Satisfaction 253* * * .075 .250* * * .090
Clinton Disconfirmation ->

Clinton Satisfaction .043 .058 .009 .066

Bush Satisfaction .067 .068 .066 .070
Bush Disconfirmation —>

Clinton Satisfaction .251 * * * .068 .288* * * .077

Bush Satisfaction .269* * * .083 .307* * * .085
Clinton Desires —>

Clinton Disconfirmation .01 3 .010 .006 .010

Clinton Satisfaction .01 1* * .006 .017* * * .007
Clinton Expectations —>

Clinton Disconfirmation -.032* * .019 -.028* .019

Clinton Satisfaction -.001 .010 -.007 .012
Clinton Attitude ——>

Clinton Satisfaction .003* * .001 .003” .001
Bush Desires ->

Bush Disconfirmation -.004 .006 -.004 .006

Bush Satisfaction -.021* * * .016 -.024* * * .006
Bush Expectations —>

Bush Disconfirmation -.03 3* * * .013 -.031* * .014

Bush Satisfaction .012 .01 1 .014* .01 1
Bush Attitude —>

Bush Satisfaction .001 .002 .001 .002

 

* Signifcant at .10, 1-tail test
* * Significant at .05, 1-tail test
* * * Significant at .01, 1—tail test

206

Table 37
Election Outcome Satisfaction Model
Summary of Differences in Model Results Using
Ali/Satisfied Clinton Voters and All/Dissatisfied Bush Voters

 

Clinton Voters Bush Voters

5.11m 811W

Total number of significant paths 6 7 10 9
Total number of paths significant at .01 1 3 6 6

Total number of significant paths from
pre-election to post-election variables 2 3 6 5

Total number of significant path from
pre-election to post-election variables
significant at .01 0 0 2 2

207

for satisfied Clinton voters versus two significant paths for all Clinton voters.
For satisfied Clinton voters, the following paths are significant: Clinton Desire
to Clinton Satisfaction, Clinton Attitude to Clinton Satisfaction, and Bush Desire
to Bush Satisfaction. For the total Clinton sample, two paths are significant:
Bush Desire to Bush Disconfirmation and Bush Expectation to Bush
Disconfirmation. Note that no pre-post path which is significant for the
entire Clinton group is also significant for the satisfied Clinton group. Thus,
removal of dissatisfied Clinton voters significantly altered all pre-post
relationships. The implication is that the relationship among pre-post
variables is significantly different for satisfied versus dissatisfied Clinton
voters. Therefore, although the research hypothesis is not supported in terms
of the number of significant paths in each group, the evidence supmrts the
idea that satisfied people and dissatisfied people process differently.

Using similar logic, if research hypothesis #3 is true one would expect
to observe more significant paths from pre-election to post-election variables
for dissatisfied Bush voters relative to all Bush voters. However, fewer were
observed (five versus six). For dissatisfied Bush voters, the following paths are
significant: Clinton Desire to Clinton Satisfaction, Clinton Expectation to
Clinton Satisfaction, Clinton Attitude to Clinton Satisfaction, Bush Desire to
Bush Satisfaction, and Bush Expectation to Bush Disconfirmation. For all Bush
voters, these same five paths are significant as is one additional path from
Bush Expectation to Bush Satisfaction. That is, all five of the paths significant
for the dissatisfied voter group were also significant for the entire voter
group. The implication is that satisfied Bush voters are not very different

from dissatisfied Bush voters in terms of pre - post variable relationships.

208

To summarize, we see that the dissatisfied Bush voter group had more
significant paths (nine versus seven) and more pre - post variable paths (five
versus three) than the satisfied Clinton voter group. Of the three significant
pre - post paths for the satisfied Clinton group, one was significant at the .05
level and two were significant at .10. Of the five significant pre - post paths
for dissatisfied Bush voters, three were significant at the .05 level and two

were significant at .01.

Thus, research hypothesis #3 receives mixed supmrt from this analysis.
Strong support is provided by the finding that dissatisfied Bush voters do
process more than satisfied Clinton voters while the shifts occurring within

the Clinton and Bush voter groups does not support the research hypothesis.

V-4.2. Analysis of EOSM with Satisfied and Dissatisfied Voters -
Results Related to Research Hypothesis #1

It is possible to once again to examine research hypothesis #1, that
multiple targets are relevant in satisfaction formation, using these more
homogeneous subsamples of Clinton and Bush voters. We saw in the previous
section that the significant paths connecting pre - post variables differed for
satisfied Clinton and all Clinton voters. If research hypothesis #1 is true, we
would expect to see paths to overall satisfaction from satisfaction with each
target in a_ll subgroups analyzed.

Among satisfied Clinton voters (Table 35), the paths from both
satisfaction with Clinton and satisfaction with Bush to Overall Satisfaction are
significant (p < .01, l-tail test). The major differences between the satisfied
Clinton voter and all Clinton voter groups with regard to these paths are 1)

that the path estimates are lower for the satisfied Clinton voters, and 2) that

209

the standard errors of each path estimate are lower for the satisfied Clinton
voters. The net result is that each path is significant at a lower alpha level in
the satisfied voter group. These relatively minor differences do not detract
from the overall support for research hypothesis #1 provided by these results.

Among Bush voters (Table 36), the same pattern of relationships is
evident between dissatisfied Bush and all Bush voters as was true among
satisfied Clinton voter versus all Clinton voters. That is, the paths from both
Clinton Satisfaction and Bush Satisfaction to Overall Satisfaction are
significant for all Bush voters and dissatisfied Bush voters. Once again, the
major differences between the two groups are 1) that the path estimates are
lower for the dissatisfied Bush voters, and 2) that the standard errors of each
path estimate are lower for the dissatisfied Bush voters.

In conclusionI analysis of both pairs of subsamples provides additional
evidence in support of research hypothesi_s #1.

V-4.3. Analysis of EOSM with Satisfied and Dissatisfied Voters -
Results Related to Research Hypothesis #2

It is possible to examine research hypothesis #2, that the processing
paths related to the two targets are related via cross-over paths, using these
relatively homogeneous subsamples of Clinton and Bush voters. As research
hypothesis #2 does not specify differences based on level of satisfaction, we
would expect to see all subgroups exhibit significant cross-over effects if
research hypothesis #2 is true.

For Clinton voters, the cross-over path from Clinton disconfirmation to
satisfaction with Bush remains significant (p < .01, l-tail test) in the satisfied

Clinton voter group, as it was among all Clinton voters. In addition, the cross-

210

over path from Bush disconfirmation to satisfaction with Clinton remains
nonsignificant for the satisfied Clinton voters as it was with the all Clinton
voter group. For Bush voters, too, the cross-over path from Bush
disconfirmation to Clinton satisfaction remains significant (p < .01, l-tail test)
while the cross-over path from Clinton disconfirmation to Bush satisfaction

remains nonsignificant for dissatisfied Bush voters.

In conclusion, analysis of both pairs of subsamples provides additional
mixed evidence in support of research hymthesis #2.

Parenthetically, these results also support the contention made by
Droge and Halstead (1991) that, unlike satisfied consumers, dissatisfied
consumers appear to be expectations-driven. In this analysis, expectations are
not related to other variables for the satisfied Clinton group, while
expectations are significantly related to consequent variables for both Clinton

and Bush among dissatisfied Bush voters.

V-S. Results of the CPSM Correlation Analysis of Satisfied versus

Dissatisfied Voters

To test research hypothesis #3, differential processing of satisfied
versus dissatisfied respondents, additional analysis was performed using CPSM
data. There was significant variance within each group of the CPSM analysis
with regard to overall satisfaction (see Figure 23). Therefore, correlation
analysis of satisfied versus dissatisfied respondents within each voter group
was performed (small sample sizes precluded the use of structural path
analysis, which would have been more appropriate for testing the research

hypothesis).

211

 
 

 

30 ‘r
27 --
24 --
Z'l --
18 ‘-
151—
1 2 ‘r
9 -i—
_ “9% .
1
extremely extremely
dissatisfied satisfied

Clinton Voters I Bush Voters

Figure 23
Overall Satisfaction with the Process (CPSM)

212

Dissatisfied voters were defined as those responding with a "1", "2" or "3"
to the overall process satisfaction question. Satisfied voters were defined as
those responding with a "S", "6" or "7" to the question. Voters who responded
with a "4", the midpoint of the scale, were omitted from further analysis. Thus,
four groups were created: Clinton voters satisfied with the process (n=40),
Clinton voters dissatisfied with the process (n=27), Bush voters satisfied with
the process (n=29) and Bush voters dissatisfied with the process (n=51).

Separate correlation analyses were performed on each of the four
groups. Results appear in Table 38. To test the research hypothesis that
dissatisfied people process more extensively than do satisfied people, several
simple counts of significant correlations (within each of the four groups)
were made:

1) Total number of correlations significant at the .10 level;

2) Total number of correlations between any variable and overall

satisfaction; and

3) Number of significant correlations between pre-election and post-

election variables.
Since simple correlations indicate two variables are linearly related, they are
interpreted here as evidence of systematic processing by the relevant
homogeneous subsample. A summary of the counts of significant correlations
and some proportions based on the counts is presented in Table 39.

Several observations from the table address research hypothesis #3.
First, given the hypothesis, we would anticipate that there would be more
significant correlations among dissatisfied voters than among satisfied voters.
While there are slightly more significant correlations for dissatisfied versus

satisfied Bush voters (23 versus 17), the reverse is true among Clinton voters

213

Table 38

Campaign Process Satisfaction Model
Subgroup Correlation Matrices

Satisfied Clinton Voters
1 1.000
2 0.112 1.000
(.490)
3 -0.171 0.396 1.000
(0.306) (0.014)
4 0.292 0.219 0.426 1.000
(0.071) (0.180) (0.008)
5 0.331 -0.089 -0.393 -0.152 1.000
(0.037) 0.585 (0.015) (0.354)
6 0.327 0.005 -0.266 -0.045 0.614 1.000
(0.039) (0.974) (0.106) (0.768) (0.000)
7 0.175 0.128 -0.017 0.262 0.389 0.157
(0.285) (0.437) (0.921) (0.112) (0.014) (0.339)
6 0.616 -0.044 -0.180 0.364 0.084 0.114
(0.000) (0.789) (0.260) (0.023) (0.607) (0.486)
9 —0.187 0.269 0.021 -0.217 0.082 -0.240
(0.255) (0.098) (0.903) (0.191) (0.709) (0.141)
10 -0.236 0.321 0.160 -0.095 -0.308 -0.253
(0.143) (0.043) (0.338) (0.567) (0.053) (0.115)
11 0.057 0.142 0.453 0.210 -0.054 -0.167
(0.730) (0.388) (0.005) (0.207) (0.744) (0.310)
12 -0.340 0.063 0.531 0.052 -0.261 -0.194
(0.034) (0.615) (0.001) (0.754) (0.109) (0.236)
13 0.412 -0.023 -0.020 0.066 0.110 0.051
(0.009) (0.890) (0.905) (0.609) (0.504) (0.757)
14 -0.042 0.115 0.496 0.661 -0.327 -0.215
(0.601) (0.485) (0.002) (0.000) (0.043) (0.188)
1 2 3 4 5 6
W
1 Disconfhmtion re: Chton Advertishg Effectiveness
2 Disconfirmtion re: Chton Unfa's Campsignhg
3 Disconfimation re: Bush Advertbhg Effectiveness
4 Disconfimation re: Bah Unfai Carrpaignhg
S Overal Disconfimation
6 Overal SatisfactionwithCarrpaign Process

1.000
0.165 1.000
(0.315)
0.400 -0.050 1.000
(0.012) (0.762)
'0.105 -0.0i O 0.290 1.000
(0.524) (0.953) (0.073)
0.274 -0.111 0.212 -0.181 1.000
(0.092) (0.501) (0.196) (0.271)
0.176 -0.500 0.069 -0.018 0.567 1.000
(0.290) (0.001) (0.681) (0.913) (0.000)
0.087 0.243 -0.318 -0.179 0.418 0.112 1.000
(0.598) (0.138) (0.048) (0.275) (0.008) (0.501)
0.089 0.137 '0.362 0.049 0.323 0.294 0.259 1.000
(0.595) (0.406) (0.023) (0.769) (0.047) (0.069) (0.116)
7 8 9 1 0 1 1 1 2 1 3 1 4
W

7 Expectation re: Chton Ad. Effectiveness

8 Performance re: Chton Ad. Effectiveness
Expectation re: Ciiton Unfa's Carrpsignhg
Performance re: Chton Unfak Cannabnhg
Expectation re: Bush Ad. Effectivenes
Performance re: Bush Ad. Effectiveness
Expectation rs: Bah Unfai Carmabnhg
Perforrrnnce rs: Bah Unfak Carmabnhg

214

Table 38, continued

Dissatisfied Clinton Voters

1 1.000
2 0.096 1.000
(0.633)
3 -0.039 0.175 1.000
(0.846) (0.382)
4 0.110 -0.206 -0.423 1.000
(0.591) (0.313) (0.031)
5 0.372 0.303 -0.019 0.015 1.000
(0.056) (0.125) (0.923) (0.941)
6 -0.220 0.141 0.246 -0.457 0.078 1.000
(0.270) (0.484) (0.215) (0.019) (0.698)
7 0.032 0.101 0.361 0.026 0.059 -0.120
(0.874) (0.616) (0.064) (0.899) (0.769) (0.550)
8 0.512 0.224 0.058 -0.309 0.013 0.089
(0.006) (0.261) (0.773) (0.125) (0.949) (0.658)
9 -0.336 0.309 -0.060 0.094 0.156 -0.205
(0.067) (0.117) (0.693) (0.683) (0.432) (0.305)
10 0.154 0.833 0.270 -0.353 0.371 -0.002
(0.442) (0.000) (0.173) (0.077) (0.056) (0.993)
1 1 -0.040 -0.117 0.307 -0.005 0.134 -0.392
(0.844) (0.562) (0.120) (0.981) (0.505) (0.043)
12 0.132 0.063 0.531 0.052 -0.261 -0.l94
(0.519) (0.572) (0.000) (0.172) (0.409) (0.180)
13 0.014 -0.064 0.180 0.149 0.077 ~0.194
(0.946) (0.752) (0.369) (0.469) (0.702) (0.332)
14 0.020 0.218 -0.041 0.602 -0.055 -0.266
(0.922) (0.274) (0.640) (0.001) (0.787) (0.147)
1 2 3 4 5 6
W
l Disconfhnation re: Chton Advertishg Effectiveness
2 Disconfimation re: Cl'mton Unfa's Carrpaignhg
3 Disconfimation re: Bush Adverthhg Effectiveness
4 Disconfirmation re: Bah Unfair Carrpsigning
5 Overal Dbconfi'mation
6 Overal Satisfaction with Canpaign Process

1.000
0.224 1.000
(0.262)
0.407 -0.235 1.000
(0.035) (0.239)
0.207 0.323 0.290 1.000
(0.993) (0.100) (0.143)
0.475 -0.061 0.141 0.100 1.000
(0.012) (0.666) (0.482) (0.621)
0.203 -0.007 -0.162 0.051 0.226 1.000
(0.321) (0.974) (0.375) (0.603) (0.266)
0.562 -0.005 0.202 0.066 0.735 0.195 1.000
(0.002) (0.960) (0.312) (0.745) (0.000) (0.340)
0.260 -0.143 0.366 0.073 0.042 -0.079 0.433 1.000
(0.190) (0.476) (0.06) (0.716) (0.636) (0.701) (0.024)
7 6 9 10 11 12 13 14
William

7 Expectation re: Chton Ad. Effectiveness

8 Perforrmnce re; Chton Ad. Effectiveness

9 Expectation re: Chton Unfa's Carnpsignhg
10 Performance re: Chton Unfair Carmaignhg
1 1 Expectation re: Bush Ad. Effectivenes
12 Performance re: Bush Ad. Effectiveness
13 Expectation re: Bush Unfal' Carmebnhg
14 Performance re: Bah Unfai Carmabnhg

215

Table 38, continued

Satisfied Bush Voters

1 1.000
2 -0.022 1.000
(0.909)
3 -0.052 -0.029 1.000
(0.789) (0.883)
4 0.311 0.316 0.295 1.000
(0.101) (0.095) (0.120)
5 -0.039 0.006 -0.023 -0.044 1.000
(0.642) (0.978) (0.904) (0.621)
6 0.030 -0.229 0.080 -0.181 0.350 1.000
(0.676) (0.231) (0.660) (0.349) (0.063)
7 0.443 -0.401 -0.050 -0.008 -0.002 0.157
(0.016) (0.031) (0.798) (0.968) (0.992) (0.366)
6 0.526 —0.150 0.191 -o.023 0.143 0.229
(0.003) (0.438) (0.320) (0.907) (0.460) (0.233)
9 -0.012 -0.025 0.092 -0.148 -0.116 -0.013
(0.951) (0.896) (0.635) (0.445) (0.551) (0.945)
10 0.096 0.675 0.145 0.156 0.015 -0.117
(0.62) (0.000) (0.453) (0.415) (0.938) (0.547)
11 -0.195 0.113 0.469 0.204 0.305 0.181
(0.312) (0.561) (0.007) (0.286) (0.108) (0.347)
12 -0.143 -0.152 0.362 0.132 0.064 -0.041
(0.458) (0.431) (0.041) (0.496) (0.664) (0.831)
13 0.289 -0.291 -0.072 0.200 0.169 0.228
(0.128) (0.125) (0.712) (0.297) (0.361) (0.234)
14 0.130 -0.066 0.086 0.335 0.276 0.067
(0.501) (0.733) (0.656) (0.076) (0.147) (0.728)
1 2 3 4 5 6
W
l Disconfimation re: Chton Advertishg Effectiveness
2 Disconfirmation re: Cinton Unfai' Carnpaignhg
3 Disconfirmation re: Bush Advertisiig Effectiveness
4 Disconfimation re: Bush Unfai Carnpaignhg
5 Overal Disconfimation
6 Overal Satisfaction with Canpaim Process

1.000
0.617 1.000
(0.000)
0.300 0.256 1.000
(0.114) (0.180)
-0.127 0.286 0.066 1.000
(0.513) (0.133) (0.732)
-0.131 0.078 0.119 0.107 1.000
(0.497) (0.689) (0.537) (0.582)
-0.021 0.053 0.283 -0.109 0.425 1.000
(0.915) (0.786) (0.137) (0.572) (0.021)
-0.077 ~0.024 -0.324 ~0.378 0.315 0.178 1.000
(0.691) (0.902) (0.086) (0.043) (0.010) (0.355)
-0.089 0.182 -0.123 -0.019 0.320 0.638 0.563 1.000
(0.648) (0.345) (0.526) (0.331) (0.090) (0.000) (0.002)
7 8 9 10 11 12 13 14
W

7 Expectation re: Chton Ad. Effectiveness

8 Performance re: Clinton Ad. Effectiveness

9 Expectation re: Ci'nton Unfair Carrpaignirg
10 Performance re: Chton Unfair Carmabnhg
1 1 Expectation re: Bush Ad. Effectivenee
12 Performance re: Bush Ad. Effectiveness
13 Expectation re: Bush Unfa's Carrpsignhg
14 Perforrrlnce re: Bah Unfai Canpsbnhg

216

Table 38, continued

Dissatisfied Bush Voters
1 1.000
2 -0.069 1.000
(0.627)
3 -0.080 0.211 1.000
(0.579) (0.137)
4 -0.282 -0.125 0.271 1.000
(0.045) (0.382) (0.052)
5 -0.229 0.155 -0.066 0.029 1.000
(0.103) (0.273) (0.546) (0.837)
6 —0.180 0.172 0.004 0.265 0.542 1.000
(0.202) (0.222) (0.975) (0.056) (0.000)
7 -0.112 —0.044 0.116 0.168 0.213 0.152
(0.430) (0.759) (0.413) (0.182) (0.126) (0.277)
8 0.505 -0.050 -0.011 0.079 -0.121 -0.108
(0.000) (0.725) (0.937) (0.560) (0.391) (0.447)
9 0.317 0.007 0.131 0.094 -0.217 -0.080
(0.022) (0.923) (0.354) (0.509) (0.119) (0.568)
10 0.303 0.203 0.175 0.194 -0.135 0.009
(0.029) (0.149) (0.219) (0.173) (0.341) (0.946)
11 0.062 -0.186 0.165 0.101 -o.106 -0.361
(0.667) (0.196) (0.251) (0.464) (0.461) (0.009)
12 0.226 -0.137 0.352 0.271 -0.225 -0.245
(0.106) (0.334) (0.010) (0.052) (0.106) (0.077)
13 -0.255 -0.004 0.072 0.446 0.237 0.164
(0.074) (0.977) (0.620) (0.001) (0.095) (0.197)
14 -0.221 0.027 0.227 0.571 0.004 0.219
(0.116) (0.648) (0.106) (0.000) (0.979) (0.115)
1 2 3 4 5 6
W
1 Disconfimation re: Chton Advertis'ng Effectiveness
2 Disconfirmation re: Chton Unfai Carmaignhg
3 Disconfirmation re: BushAdvertislng Effectiveness
4 Disconfirmation re: Bah Unfa's Carrpaign'ng
5 Overal Disconfirnation
6 Overal SatisfactionwithCanpaign Process

1.000
0.043 1.000
(0.761)
0.366 0.382 1.000
(0.007) (0.005)
0.210 0.078 0.385 1.000
(0.136) (0.581) (0.005)
-0.089 0.306 0.162 -0.130 1.000
(0.533) (0.031) (0.258) (0.367)
-0.052 0.396 0.176 0.121 0.558 1.000
(0.711) (0.004) (0.206) (0.395) (0.000)
0.073 0.144 -0.019 -0.179 0.294 0.113 1.000
(0.612) (0.308) (0.896) (0.213) (0.036) (0.430)
0.031 -0.078 0.082 0.072 -0.060 0.199 0.339 1.000
(0.828) (0.584) (0.557) (0.612) (0.675) (0.153) (0.015)
7 8 9 1O 1 1 12 13 14
W

7 Expectation re: Chton Ad. Effectiveness

6 Performance re: Chton Ad. Effectiveness

9 Expectation re: Chton Unfa's Carmaignhg
10 Performance re: Chton Unfair Carmaignhg
1 1 Expectation re: Bush Ad. Effectivenes
12 Performance re: Bush Ad. Effectivenea
13 Exactation re: Bah Unfai Carnpabnhg
14 Performsnce re: Bah Unfai Canpaignhg

217

Table 39
Campaign Process Satisfaction Model
Summary of Correlation Analysis

 

Clinton Voters Bush Voters
Satisfieg Dissatigfied Satisfied Dissatisfied

Total number of significant correlations 30 19 17 23

Total number of significant correlations
between any variable and overall satisfaction 2 2 1 4

Total number of significant correlations
between pre-election and post-election
variables 8 4 7 10

Proportion of significant correlations
accounted for by pre-election and
post-electionvariables .27 .21 .41 .43

218

(19 for dissatisfied versus 30 for satisfied voters). Second, if research
hypothesis #3 is true, we would expect to see more significant correlations of
any variable and overall satisfaction for dissatisfied versus satisfied voters.
Again, for Bush voters, this is observed (four significant correlations for
dissatisfied voters versus one for satisfied voters); also again, Clinton voters did
not exhibit the hypothesized pattern (two significant correlations for each
voter group). Third, one would expect to see more significant correlations
between variables measured pre-election and variables measured post-election
("pre-post" correlations) for dissatisfied voters. This would provide the best
indication of additional processing because the time difference between
measurements of pre-election and post-election variables reduces the
likelihood that factors other than processing are responsible for the
correlations. Once again, dissatisfied Bush voters have more significant
correlations than do satisfied Bush voters (10 versus 7), while satisfied Clinton
voters have more significant correlations than do dissatisfied Clinton voters (8
versus 4). Finally, one might expect that the proportion of significant
correlations accounted for by pre-post correlations would be higher for
dissatisfied voters. The difference in these proportions between dissatisfied
and satisfied Bush voters is very slight (.43 versus .41), while the difference
in proportions for satisfied versus dissatisfied Clinton voters is slightly greater
but in the opposite direction (.27 versus .21). Thus, these results provide, at
best, weak support for research hypothesis #3.

To determine whether the counts observed in Table 39 were due to the
differences in sample sizes among the four groups, the smallest significant
correlation (r = .24) which was significant in the dissatisfied Bush voter group

(the largest group) was used as a threshold on all four groups, and the analysis

219

was repeated. No differences in the pattern of counts or in the general
conclusions resulted.

One possible explanation of these results, other than that satisfied and
dissatisfied people do not process differently, is that overall campaign process
satisfaction is less influential on processing than is the election outcome.
Bush voters may be dissatisfied with the outcome of the election and this
dissatisfaction, rather than satisfaction or dissatisfaction with the campaign
process, may have dominated the determination of processing intensity. The
difference between Bush voters and Clinton voters with respect to proportion
of all correlations accounted for by pre-post correlations appears at first to
support this explanation. If this explanas'on and research hypothesis #3 were
both true, however, then Clinton voters satisfied with the process should have
processed the least, which clearly did not occur.

Further discussion of this issue, including another possible explanation

for the pattern of findings, appears in the next chapter.

220

V-6. Summary of Results Relating to the Research Hypotheses

In this section, selected results relating specifically to each research
hypothesis are summarized and cross-referenced to other sections of the
Results chapter. Note that a complete discussion of the results and their

implications appear in the next chapter.

0 Research hypothesis 1: Multiple targets of satisfaction may remain salient
post choice during satisfaction formation.

Support for this hypothesis is found in 1) the EOSM using data from all
Clinton and Bush voters (Section V-2.3), 2) the EOSM using data from satisfied
Clinton voters and dissatisfied Bush voters (Section V-4.2), and 3) the CPSM
using data from all Clinton and Bush voters (Section V-3.3). In both EOSM
analyses, paths from both satisfaction with Clinton and satisfaction with Bush
to Overall Satisfaction were significant for every group analyzed. Further,
antecedent variables significantly impact consequent variables within the

paths associated with each target, choice and non-choice in the EOSM and

 

CPSM analyses.

When one considers that the EOSM and CPSM use different variables and
different structural forms, yet yield similar results with regard to this
hypothesis, we can conclude thgt research hymthesis #1 is strongly
supported. This finding is probably the most important contribution of this
dissertation research as it clearly demonstrates that the traditional CS/ D model

with its focus on a single target is overly simplistic.

221

0 Research hypothesis 2: The processing paths associated with each of the
multiple targets of satisfaction are interrelated, i. e. there are cross-over
effects.

Support for this hypothesis is found in the EOSM using data for all
voters (Section V—2.4) and also using data from satisfied Clinton and dissatisfied
Bush voters (Section V-4.3). In all four groups analyzed (Clinton voters,
satisfied Clinton voters, Bush voters, dissatisfied Bush voters), the cross-over
path from disconfirmation with the choice to satisfaction with the non-choice
is significant while the cross-over path from disconfirmation with the non-
choice to satisfaction with the choice is not significant. Thus, there is strong

support for the existence of a cross-over effect but the effect is not as

 

extensive as hyppthesized (i.e., there is no support for the disconfirmation
with non-choice - satisfaction with the choice path).

In addition, there is evidence that cross-over effects not hypothesized
specifically may occur elsewhere along the satisfaction formation paths. For
example, in the EOSM, the single largest modification index (12.049) is for the
non-modeled path Bush Expectation to Clinton Satisfaction among Bush voters
(Table 24). In addition, several of the large standardized residuals are for
cross-over relationships such as Bush Desire - Clinton Satisfaction (-3.225) and
Clinton Desire - Bush Disconfirmation (2.489) among Bush voters, and Bush
Attitude - Clinton Satisfaction (2.431) and Clinton Attitude - Bush Satisfaction
(2.452) among Clinton voters.

Thus, while there is strong support for a cross-over effect, the exact
nature and extent of such an effect cannot be resolved from these results

alone.

222

0 Research hypothesis 3: Dissatisfied subjects process more extensively or
completely than do satisfied subjects.

Support for this hypothesis is found in the EOSM analysis of satisfied
Clinton and dissatisfied Bush voters (Section V-4.1) and the EOSM analysis of all
Clinton and all Bush voters (Section V-2.4). In both analyses, there are more
significant paths associated with Bush voters and dissatisfied Bush voters than
with Clinton voters and satisfied Clinton voters. In addition, in both analyses,
there are more significant paths connecting variables measured pre-election
and variables measured post-election for the dissatisfied groups. However,
contrary to what one would expect given research hypothesis #3, the number
of significant pre - post paths 1) increased when dissatisfied Clinton voters
were removed from all Clinton voters, and 2) decreased when satisfied Bush

voters were removed from all Bush voters. In short, the hymthesis is

supported in thgt dissatisfied voters do process more than do satisfied voters,

 

but it is not supmrted in that the minor changes which occur within each
voter group as that group becomes more homogeneous are oppogte to th_at

 

hypothesized.

The CPSM correlation analysis (Section v.5) does not support the
hypothesis in its current form, though it does support the idea that satisfied

and dissatisfied voters process differently.

0 Research hypothesis 4: Desires have a greater impact on satisfaction
formation than do expectations.
Support was found for research hypothesis #4. Results of the total
(direct and indirect) effects of both desires and expectations on the various

satisfaction measures of the EOSM were directly compared in Section V-2.6.

223

Where differences between the two exist, desires appear to have greater
effects than expectations on all three satisfaction measures. Further, we saw
that a total of three high modification indices indicated relationships between
desires and overall satisfaction while only one high modification index

indicated a similar expectations - overall satisfaction link (Table 24). Thus, the

hypothesis is supported.

Chapter VI
Discussion and Implications

The purpose of this research was to develop and test a model of
satisfaction formation which explicitly accounted for the effects of both
choice and nonchoice targets. Two models were specified and tested using data
from the 1992 U.S. presidential election. Though not all hypothesized path
relationships in each model are supported by the data, as a whole, the results
provide moderate-to—strong support for each of the four research hypotheses.
Detailed results, and brief interpretations, are presented in the previous
chapter. In this chapter, the theoretical and practical implications of the
findings are discussed, directions for future research are highlighted, and the
research limitations are presented.

This chapter is organized as follows. In the next section, VI—l, the
results and implications related to the first two research hypotheses are
discussed. Together, these two hypotheses provide the framework for the
competitive model of satisfaction formation. In Section VI-2, results and
implications related to research hypothesis #3, differential processing
associated with level of satisfaction, are discussed. In Section VI—3, results and
implications related to research hypothesis #4, the relative importance of
desires versus expectations, are discussed. In Section VI-4, other findings are
discussed in relation to past research including: the relationship of
expectations to disconfirmation and to satisfaction (VI-4.1), disconfirmation
(VI—4.2), attitudes (VI-4.3), and the context of this research (VI-4.5).
limitations of the research are presented in VI-S. Finally, in VI-6, brief

concluding remarks are presented.

224

225

VI-l. The Competitive Model of Satisfaction Formation

Because it is impossible to discuss separately the implications of multiple
targets (research hypothesis #1) and the nature of their effects (research
hypothesis #2), and because these two research hypotheses together provide
the competitive model's structure, the discussion in this section addresses both
hypotheses. The finding that multiple targets may be relevant to satisfaction
formation is probably the most significant result of this dissertation. It
implies that satisfaction formation is subject to competitive forces, at least in
an election context. Further, it implies that the traditional model of
satisfaction formation may be fundamentally flawed in the context of an
election and perhaps in other situations.

First, it should be noted that the satisfaction formation model developed
here does not reg uire that competitive forces be included. For situations in
which competition is not relevant, path values for all but the choice target
may be specified as zero. For many applications, therefore, the currently
accepted single target model may be utilized. However, for situations in which
nonchoices remain salient post-purchase, the proposed model should be used
with nonchoice paths estimated.

The multiple targets are relevant to the extent they affect the concept of
"overall satisfaction." Consumers are ultimately attempting to satisfy some
need or want, i.e., they are attempting to have a satisfying consumption
experience. This dissertation has demonstrated that the concept of overall
satisfaction is relevant and is the result of satisfaction with multiple
competitive targets, at least in the context of head-to-head, zero-sum

competition.

226

One powerful implication of an overall satisfaction construct is that it
can account for competition even among competitive choices which are
difficult to compare directly, such as "sequenced" choices. For example,
assume someone wants to have a nice-looking lawn. He recognizes that he can
cut his lawn himself or hire a lawn service to do so. He decides to cut the lawn
himself and considers buying one of two lawn mowers, Brand A or Brand B. He
selects Brand A. He is satisfied with the performance of his choice mower,
especially given that a neighbor bought a Brand B mower and has had several
problems with it. But, he is dissatisfied with his choice to select a mower at all,
especially given that his other neighbor has hired a lawn service and thus has
a better-looking lawn and more leisure time than does our subject.

The traditional satisfaction model would specify that satisfaction is the
result of antecedents related only to the Brand A mower. The competitive
satisfaction model as discussed thus far would add the antecedents of and
"satisfaction" with the Brand B mower as well. However, this same model could
be expanded further to include the antecedents of and "satisfaction" with the
lawn service, too. Although the subject may be perfectly satisfied with the
mower per se, overall he is less satisfied because the lawn service was not
chosen.

We see that the model can be specified in a manner which mirrors the
competitive reality in which the total consumption experience is occurring.
From a managerial perspective, this enhances the usefulness of the model. As
discussed earlier, traditional models attempt to incorporate competitive factors
through the disconfirmation standard. However, even the most realistic of
these, best brand norms and ideal brand norms, are product-related rather

than consumption experience related. The Brand A mower manufacturer

227

using the traditional model would have no way of examining postconsumption
processes in a meaningful way. The concept of overall satisfaction, coupled
with the recognition that multiple targets can remain salient postchoice,
provides a potentially powerful tool both theoretically and managerially. To
take full advantage of this potential, future research needs to examine the
degree to which non-comparable nonchoices can be included via the concept
of overall satisfaction.

Finding support for the hypothesized cross-over effect in the EOSM is
also extremely significant. Evidence of this effect suggests that the processes
associated with each of the relevant targets are themselves interrelated. The
traditional satisfaction literature treats competing products as being
irrelevant or as being relevant as a comparison standard at a single point only
(see Section II-3.5). For those models in which competing products are used as
a comparison standard, the comparison is operationalized in the
disconfirmation construct. The evidence here suggests that utilizing
competing products in only this manner is overly simplistic. A person's
perceived disconfirmation of a choice product may affect satisfaction with the
choice and also with the nonchoice. That this is true in a head-to-head
competitive situation is relatively easy to accept. For a Bush voter, whether
Bush did better or worse than expected is directly tied to whether Clinton did
better or worse than expected, so it is not surprising that disconfirmation of
Bush's performance affects satisfaction with each candidate.

Awaiting further study is the issue of whether similar cross-over
effects may also exist elsewhere along the processing path. For example, an
examination of the EOSM modification indices reveals that freeing the path

from Bush Expectation to Clinton Satisfaction would greatly improve model fit.

228

Also, it is likely that if perceived performance were included in a similar
model (i.e., one without clear, objective and widely understood performance
measures), there may be cross-over paths to or from it. For example, what one
expects or desires to occur relative to one's choice may reasonably influence
perceived performance of one's nonchoice. Likewise, perceived performance
of one's nonchoice may affect satisfaction with one's choice, both indirectly
through disconfirmation as well as directly. Obviously, both theoretical and
empirical attention is necessary with regard to this issue.

What is the practical meaning of such a cross-over effect in a non-zero-
sum competitive context? In Section [-3, an example was used to illustrate the
effect. A husband and wife each purchased a car at approximately the same
time. They discussed their selection process and each formed expectations and
desires regarding their own choice and the choice of the spouse (their
"nonchoice"). They each usually drive their own car but occasionally drive
(and often ride in) their spouse's car. If the husband is asked his level of
satisfaction with his car, it seems reasonable that, in addition to his perceived
performance and disconfirmation related to his own car, his disconfirmed
expectations of his wife's car would affect his satisfaction with his car.

One might argue that the husband has had extensive direct experience
with his wife's car and therefore the situation is not typical of consumer
products in general. Other forms of "experience" may provide enough
information for a similar process to occur in other situations. Two categories
of such experiences may be actual past experience or vicarious experience
(Houston, Sherman, and Baker 1991). Fi_rs_t_, actual past experience with a
competing brand may provide a salient nonchoice during satisfaction

formation, especially if the current choice is a trial purchase. Second,

229

vicarious experience may also support powerful nonchoice processing.
Seeing others drive a car you almost purchased, reading performance data
about the car in Consumer Reports. seeing advertisements about the car, etc.,
all can reasonably support parallel nonchoice, postpurchase satisfaction
processing. Of course, the situations under which interrelated processing
occurs can be identified and delineated through empirical testing; however, it
seems likely that such processes occur even when direct experience is absent.
Evidence of cross-over effects also has important managerial
implications. For firms associated with the customer's choice (choice firms),
opportunities for pre-emptive defensive marketing exist. Choice firms can
manage customer satisfaction with the choice through customer service and
quality control programs. They can also affect their customer's satisfaction
with nonchoices. Immediately after making a choice, customers begin
attempting to reduce dissonance caused by the act of choosing. Such
dissonance occurs because of 1) positive aspects of the nonchoices, and 2)
negative aspects of the choice (Festinger 1957). Some firms attempt to
capitalize on this consumer tendency, reinforcing the customer's decision by
highlighting positive aspects of the choice postpurchase. The opportunity
also exists for choice firms to take actions which reduce satisfaction with
nonchoices, i.e., a form of "blocking." This topic has not been addressed and is
in need of exploration. For example, for some products, providing customers
(or specific segments of customers) with comparative information
postpurchase may be more effective than simply saying "congratulations with
your new purchase" in 1) increasing satisfaction with the choice, and 2)
decreasing satisfaction with the nonchoice. Of course, such tactics have

significant inherent risks similar to those associated with comparative

230

advertising; however, the potential payoff as a defensive tactic is significant

enough to warrant further investigation.

Similarly, for firms associated with the customer's nonchoices
(nonchoice firms) opportunities for offensive marketing may exist. The
existence of cross-over paths imply that nonchoice firms should not focus
only on competition during the consumer's decision-making period. Instead,
competitive actions immediately following purchase of the choice should be
considered. The message from this research is "don't wait until the customer is
ready to purchase again." Nonchoice firms should attempt to counteract the
tendency of consumers to reduce postpurchase dissonance. Specifically, they
can try to increase the customer's attention towards negative aspects of the
choice and towards positive aspects of the nonchoice. There is evidence that
some firms have begun to do this. For example, food retailers have recently
begun to offer a service to nonchoice firms at the checkout stand. If a
consumer purchases brand A tomato soup, the consumer receives a cents-off
coupon for purchasing brand B tomato soup within the next month. Such a
strategy may be extended to nonprice elements of the marketing mix.
Similarly, activities directed towards reducing satisfaction with the choice are
beginning to appear. For example, an American automobile dealership near
Flint, Michigan, has begun to market auto maintenance services via direct
mail to purchasers of foreign-made automobiles. The featured message is:
"foreign car dealers often overcharge for parts and service. Though you
recently purchased a foreign car, consider having it serviced at an American
car dealership." As relationship marketing arenas become more widespread,

competition during postpurchase processes will be likely to increase; the

231

model presented here provides a framework for analyzing competitive forces

during these processes.

Another area for future research is the examination of which
nonchoices remain salient post-purchase. It may be possible to investigate
empirically the contents of consumer choice sets, or at least those elements of
the choice set which were most seriously considered, by identifying those
nonchoices which remain salient in post-purchase processes. Knowledge of
which nonchoices remain salient post-purchase would lead to questions of
why some nonchoices remain important while others do not. Answers to these
questions would be of critical interest to managers seeking to devise specific
strategies for gaining competitive advantages in this important but under-
researched competitive arena.

The extension of the model presented here to multiple nonchoices
is theoretically straightforward but raises significant measurement problems.
One potential problem with using this type of model is that data collection and
analysis may be cumbersome if large numbers of nonchoices remain salient.
To add one additional nonchoice increases the volume of basic data required by
50%. Adding two additional nonchoices would double the amount of data
required. If multiple measures and/ or multiple attributes are included,
questionnaire length quickly becomes excessive. Also, as the number of
salient nonchoices increases, the sample size required to estimate the model
also increases quickly because the number of paths to be estimated increases
more than arithmetically (if all theoretically relevant paths are included).
For example, adding candidate Perot to the model in this dissertation would

have resulted in the estimation of 27 paths for the EOSM (versus the 16 paths

232

currently estimated). Using the rule of thumb that a minimum sample size
should be five times the number of paths to be estimated, the minimum 11
required for each group would increase from about 80 to about 135. So adding
one candidate would have required an additional 1 10 subjects (assuming we
were not also interested in Perot voters) and a questionnaire about 50% longer.
Still, collecting sufficient data on one or two nonchoices is feasible and
productive, and may be all that is required to address the majority of situations

in which the traditional CS/ D model is inadequate.

VI-2. Differential Processing

The results presented in Chapter 5 provide support for the research
hypothesis that dissatisfied voters process more extensively than do satisfied
voters. Strongest support comes from results associated with the EOSM; these
results will be reviewed briefly here to introduce the subsequent discussion.
Arguably, the number of paths linking pre-election and post-election
variables is indicative of amount of processing (see Section III-2.3). Six of ten
such paths are significant for Bush voters while two of ten are significant for
Clinton voters (Table V-6). Somewhat surprising is the decrease in number of
significant paths for Bush voters and the increase in number of significant
paths for Clinton voters when each sample is made more homogeneous with
regard to overall satisfaction.

Looking first at the Clinton voters, the fact that the Bush desires -
disconfirmation and Bush expectation - disconfirmation paths both become
nonsignificant when neutral and dissatisfied Clinton voters are removed
indicates that these voters were disproportionately influenced by their focus

on Bush, and especially by their expectations regarding Bush (note the

233

change in size of the parameter estimate from -.040 to -.016 in Table V-16).
Note also that all pre-post paths shifted for the Clinton voter sample: neither

path significant before removal of the neutral/ dissatisfied Clinton voters

 

remained significant. All the paths added were related to desires and attitudes,
not expectations.

For Bush voters, the removal of satisfied and neutral voters had a
different effect. Though one path went from being just significant to being
just nonsignificant (the parameter estimate dropped from .014 to .012 while
the standard error remained the same), the other two paths involving
expectations experience slight increases in their unstandardized estimates and
increases in level of significance. All significant paths following removal of
satisfied/neutral Bush voters were also significant for the entire Bush sample.
The implication is that, if anything, expectations were disproportionately
more important to dissatisfied voters.

Collectively, these findings from the EOSM are consistent with the Droge
and Halstead (1991) contention that dissatisfied people are more expectations-
driven than are satisfied people. It is also consistent with cognitive
dissonance theory which would predict that dissatisfied voters probably
experience dissonance and thus are motivated to attempt to reduce the
dissonance by processing and reprocessing relevant cognitions. Satisfied
voters do not experience dissonance among these cognitions and thus are not
motivated to process.

It is tempting to speculate that an elaboration likelihood model (ELM)
explanation of this phenomenon is relevant (Petty and Cacioppo 1986). Petty
and Cacioppo state that people tend to process peripherally unless they have

the ability and motivation to process centrally. It is likely that in the context

234

of a presidential election and this particular study sample, most respondents
possess the ability to process the EOSM variables, so what separates those who
do and do not process centrally must be motivation. Thus, from the results
presented here, it appears that satisfaction/ dissatisfaction may be a
determinant of central versus peripheral processing.

The problem with this statement, of course, is that Clinton and Bush
voters differ in ways other than simply their level of satisfaction/
dissatisfaction, and it may be one or more of these other differences which
account for the differences in processing route. One such interesting
difference, the one with the greatest potential to provide a theoretical
explanation of differential processing route, is magnitude of disconfirmation.
With regard to both Bush's and Clinton's performance, Bush voters have a
greater absolute level of disconfirmation than do Clinton voters (where
"absolute level of disconfirmation" is defined as the absolute value of the
difference between mean disconfirmation and "4," the midpoint of the

disconfirmation scale). Results of the EOSM analysis are consistent with both

 

the disconfirmation and satisfaction/dissatisfaction explanation of processing
route. That is, Bush voters are more likely to process extensively ("centrally")
and both 1) are more dissatisfied, and 2) have greater absolute disconfirmation
than do Clinton voters. However, if absolute level of disconfirmation and n_ot
satisfaction/ dissatisfaction per se determine amount of processing, then a
group which is satisfied and has greater absolute disconfirmation will process
more than a group which is dissatisfied and has lo_wpr_' absolute levels of
disconfirmation. The CPSM data allows us to examine this situation.

In the CPSM, Clinton voters have higher overall satisfaction and also

slightly higher absolute value of disconfirmation. Thus, one might expect that

235

Clinton voters would exhibit greater processing than do Bush voters; in fact,
such appears to be the case, as 15 paths are significant for Clinton voters,
while 9 are significant for Bush voters (Table V-13). Also, there are more
significant correlations among Clinton voters than among Bush voters (Tables
V-19 and V-20).

Of course, the CPSM does not provide an ideal set of variables for testing
this idea: though "satisfaction with process" is the relevant satisfaction
variable for the model, it is likely that peoples' overall satisfaction is
dominated by the election outcome. Thus, even Clinton voters dissatisfied with
the process are likely to be satisfied overall while Bush voters satisfied with
the process are likely to be dissatisfied overall. (This factor may be driving
the high proportion of significant correlations attributable to variables paired
with "expectations," the one pre-election variable in this model).

Investigating this issue further is beyond the scope of this dissertation.
However, future research could address the issue by directly comparing high
versus low absolute value disconfirmation treatments and high versus low
satisfaction levels; such a comparison could best be accomplished using an
experimental or quasi-experimental design.

Disconfirmation possesses a few characteristics which make it
theoretically preferable to satisfaction as a determinant of processing route.
Because it occurs prior to satisfaction in any causal model, it is more likely to
be able to explain why processing is central versus peripheral. For a pre-post
relationship to be significant, only the "post" variable is subject to adjustment
post-performance (assuming the "pre" variable is actually measured pre-
performance). Disconfirmation, the judgment of the performance -

expectation relationship, is the most likely place such an adjustment could

236

occur. By the time satisfaction/dissatisfaction is formed, presumably all pre-
post relationships have been established, and thus satisfaction/dissatisfaction
does not determine processing path but can only function as a classification
variable after the fact; i.e., something else is responsible for the differential
processing and disconfirmation seems the most reasonable construct for
performing this function.

It may be difficult to test this idea in a typical marketing context
because of the difficulty of positively disconfirming expectations to the same
extent they can be negatively disconfirmed. Expectations are generally high
so there is a lot of "room" for performance to fall short of expectations, but
little room to exceed expectations in most commercial settings. Because of this
asymmetry, observations that dissatisfaction is associated with increased
processing are indistinguishable from observations that high absolute value
disconfirmation is associated with increased processing.

To address this issue well, the larger issue of what exactly
disconfirmation is must be resolved. The multiple serious problems with both
the conceptualization and operationalization of disconfirmation are discussed
in Section 11.3.5. There is strong evidence that disconfirmation of standards
other than expectations (e.g., desires) is more appropriate theoretically and
empirically (Cadotte, Woodruff, and Jenkins 1987; Spreng and Olshavsky 1993;
Westbrook and Reilly 1983). The existence and maybe even the source of
differential processing based on level of satisfaction/dissatisfaction can be
addressed, but completely understanding the determinants of central versus
peripheral processing route selection will probably have to wait until the

questions surrounding disconfirmation are resolved.

237

The heightened relevance of expectations among dissatisfied Bush
voters may also be due, in part, to an "inoculation" effect; that is, Bush voters
may have artificially lowered their expectations of Bush in an attempt to
moderate their ultimate level of dissatisfaction. Polls had consistently showed
Bush to have less support than Clinton. Though data which would have shed
light on this issue were not collected, such an effect would have been entirely

consistent with regret theory.

VI-3. Total Effects of Desires versus Expectations

Expectations has long been a key antecedent variable in models of
satisfaction formation (Czepiel and Rosenberg 1977; Miller 1977; Swan 1981).
While desires is a relatively recent addition to the literature, it has strong
theoretical and empirical support (see Section II-3.2). The purpose of research
hypothesis #4 was to investigate the relative importance of these two variables
on satisfaction formation.

From Chapter V, we see that, for both Clinton and Bush voters, desires
have greater direct and indirect effects on all three satisfaction measures than
do expectations. In addition, desires appears to provide unique explanatory
power. One might expect that, especially in a political context, desires and
attitudes would be highly correlated, but such is not the case in this research
(desires - attitude correlations range from -.09 to .52).

Logically, this is not surprising. The use of expectations in satisfaction
models, both independently and as a comparison standard for disconfirmation,
has multiple conceptual problems (e.g., Cadotte, Woodruff, and Jenkins 1987;
Spreng 1992; Tse and Wilton 1988). On the other hand, the case can be made

that satisfaction is dependent, at least in part, on the extent to which

238

performance is consistent with desires (Spreng 1992; Boulding et a1 1993). In
addition, the modification indices for the EOSM indicate that the addition of
paths from desires to other consequent variables may significantly improve
the fit of the model.

Clearly, additional examination of desires as an antecedent variable in
satisfaction formation is warranted. One important issue needing resolution is
how desires differ from similar constructs, most notably, from ideals. Others
have addressed the question (e.g., Spreng and Dixon 1992; Tse and Wilton
1988) but there has been no resolution which has been subjected to empirical
verification. The distinction is complicated by the reality that people are
usually faced with constrained choices. In this research, for example, gm
that Bush was the Republican nominee, is there any difference between
"desiring Bush to win the election" and "Bush is the ideal candidate on the
ballot to win the election?" On the other hand, under no constraints, wouldn't
one desire the ideal? Again, in this context, it seems that many Republicans
would "desire" a Reagan clone because that is their "ideal," while many
Democrats would desire a Kennedy clone. If, as some suggest, the difference is
the reality constraint, there should be no confusion as long as the constraints
(or lack of constraints) are clearly understood.

The relatively low correlations between desires and attitudes is
somewhat surprising. There are two possible explanations. First, attitude was
measured as the summation of beliefs times importance for four attributes.
Though there is strong support for selecting these four attributes, it is possible
that other variables were salient to respondents. Desires, being
operationalized generally, could be based on anything the respondents

considered important. Second, the issue of reality constraints may be relevant.

239

The implicit standard for the belief component of attitude measurement is an
ideal (i.e., the ideal candidate would receive a "10" on each belief measure). It
is very likely that some voters rated a candidate low on a belief item, but still
gave the candidate a high desires score because, given the limited choices, that
candidate was the most desired. For example, Clinton voters may have rated
Clinton low on a particular attribute (e.g., personal integrity) but still assigned
him all 100 points in the desires question because they did not want Bush or
Perot at all. Similarly, Bush voters could have rated Bush low on a particular
attribute (e.g., domestic economy) but still assigned him all 100 points in the

desires questions to indicate they did not want Clinton or Perot.

Finally, the issue of how best to measure desires needs to be resolved.
Desires may refer to a "more is better" versus an "ideal point" attribute, and
measurement scales may have to differ accordingly. Also, researchers should
address whether "desired outcome" or "strength of desires" is the most
appropriate construct for a given situation. While the desires scale used in
this research may have been flexible, interpretation of responses is difficult
in this regard. If a voter responded in terms of "desired outcome," a rating of
100 points implies he or she wanted the other candidates to receive 0% of the
vote. Is this really the outcome desired? On the other hand, a response of 100
points in a "strength of desires" context implies the respondent felt no
ambivalence towards the candidate. Both interpretations of a desires question
- desired outcome versus strength of desires -- are reasonable in different
contexts, depending on the exact research question of interest. When "desired
outcome" is the relevant construct and choice is binary, perhaps desires

should be operationalized as a categorical choice. When "strength of desires"

240

is relevant (for example, if one is trying to assess how difficult it would be to
change desires), a continuous scale seems to be the appropriate measurement
tool.

Desires may be more congruent with the affective nature of satisfaction
than are expectations. Some early researchers conceptualized satisfaction as a
"judgment" implying that cognitive factors alone could account for the
satisfaction formation process. Recently, however, satisfaction researchers
have been focusing more and more on the affective dimension of satisfaction
(e.g., Oliver and Westbrook 1991; Oliver and Mano 1993; Oliver 1993).
Affective elements of anticipation may be more accurately captured with a
"desire" construct than with an "expectation" construct. The findings
presented in this study are consistent with such an interpretation.

To make the case that desires should be included as a "standard"
antecedent variable in satisfaction formation models, the following should be
demonstrated: 1) there are significant theoretical reasons for including
desires as an antecedent variable, 2) "desires" are distinct from other
antecedent constructs currently included in such models, and 3) there is
empirical evidence that desires functions in this manner across multiple
contexts. The first requirement was addressed in Chapter II. The second and
third requirements are supported by the findings of this research. One
important requirement for future research is to determine whether "desires"
functions in its capacity as an important antecedent variable across a wide
variety of contexts using multiple methods.

One important implication of this finding is that managers should not
be urged to manage satisfaction by manipulating only expectations and

performance quality. In addition to the obvious risk of "under-selling" a

241

product's benefits in a competitive environment, the focus on expectations
(which are relatively easy to manipulate) versus desires (which are difficult

to manipulate) is on shaky theoretical and empirical ground.

VI-4. Other Findings in Relation to Past Research
This section contains a discussion of the results of the current research
which do not neatly fit into one of the research hypotheses but which are

interesting when considered in relation to previous research.

VI-4.1. The Relationship of Expectations to Disconfirmation and to
Satisfaction

The relationship between expectations and satisfaction in the
competitive satisfaction model presented here is similar to the relationship
described in the satisfaction literature. However, the signs of these paths
illustrates a point often omitted in discussions of this relationship: the path is
positive when the expectations refer to a desired trait and negative when the
expectations refer to an undesired trait. In most previous research,
expectations are usually expectations with regard to a desired trait, and so the
relationship is hypothesized to be positive. In cases of head-to-head
competition a trait may be desirable for both alternatives, undesirable for both
alternatives, desirable for one's choice and undesirable for one's nonchoice,
or undesirable for one's choice and desirable for one's nonchoice. In the
process model, for example, advertising effectiveness was an element which
was desirable for one's own candidate and undesirable for the opposing
candidate. Thus, for Clinton voters, the relationship was positive with respect

to Clinton's advertising effectiveness and negative with respect to Bush's

242

advertising effectiveness. The opposite is true for Bush voters. However, no
one wanted Clinton or Bush to run an unfair campaign. Thus, the relationship
between expectations and satisfaction was negative for both Bush and Clinton
voters. In other words, a competitive model of satisfaction introduces an
additional element of complexity even to relationships about which consensus
exists.

Similarly, the relationship between expectations and disconfirmation
appears to be more complex in this competitive model than in the traditional
model. Expectations have been thought to affect satisfaction directly as well as
indirectly through disconfirmation (Yi 1990) The expectations -
disconfirmation relationship is generally thought to be negative, i.e., as
expectations rise the judgment that performance meets or exceeds expectations
decreases. Of the eight expectations - disconfirmation paths in the CPSM, five
are significant; two of the five observed relationships are positive. The
positive paths appear for campaign elements which logically are undesirable
for a particular group. Specifically, Clinton voters probably would not want
Clinton to run an unfair campaign, nor would they want Bush's advertising to
be effective. In both cases, the positive expectations - disconfirmation path is
paired with a negative expectations - satisfaction path, consistent with the
current understanding of both direct and indirect effects of expectations on
satisfaction (albeit with both signs opposite to those generally reported). For
the two similar paths for Bush voters (expectation of Clinton's advertising
effectiveness and its related disconfirmation, and expectation of Bush's unfair
campaign and its disconfirmation), the signs are also positive, though the

paths are not significant.

243

More puzzling, however, is the pattern of relationships of expectations -
disconfirmation and performance - disconfirmation for these elements. In
both cases, the performance - disconfirmation relationship is positive, too.
These relationships seem to be illogical because they imply that people are
more likely to judge that performance exceeds expectations the higher
expectations are.

One possible explanation of this phenomenon is that the relevant
"expectation" used to make the disconfirmation judgment differs for desirable
versus undesirable elements. It is possible that retrieved expectations form
the basis for disconfirmation judgments when an element is undesirable while
actual expectations form the basis for the comparison for desirable elements.
Alternatively, it is possible that both comparisons rely on retrieved
expectations, but that retrieved expectations for desirable elements remain
similar to actual expectations while retrieved expectations for undesirable
elements decrease relative to actual expectations. Finally, it is possible that
expectations for undesirable elements have a greater impact on perceptions of
performance than do expectations regarding desirable elements. That is, high
expectations for an undesirable element bias individuals to perceive
performance as being "worse" (more undesirable) whereas the same does not
occur for desirable elements. It is well-accepted that consumers "over value"
negative information in decision-making processes (e.g., Mizerski 1982;
Richins 1983); perhaps different satisfaction formation processes operate
when negative elements are explicitly considered. Given recent fruitful work
on the distinction between retrieved versus prior expectations (Drtige and
Halstead 1991; Halstead 1993), future research should start with that

distinction as a possible explanation of the findings reported here.

244

VI-4.2. Disconfirmation

Although not a focus of the dissertation, results do shed light on the
utility of the disconfirmation model of satisfaction formation. Disconfirmation
with each candidate's performance is a significant antecedent of satisfaction
with the chosen candidate for both groups in the EOSM. Also, overall
disconfirmation is the dominant predictor of overall process satisfaction in the
CPSM. To this extent, the findings illustrate at least the predictive ability of
disconfirmation. However, two caveats are necessary. First, because no
subjective measures of performance are included, the strength of the
disconfirmation - satisfaction relationship may be overstated in the EOSM.
Second, while high in predictive ability, subjective disconfirmation lacks
strong theoretical meaning. This problem is illustrated by the absence of
links with variables which ought (theoretically) to be antecedent to
disconfirmation in both the CPSM and EOSM.

One problem with the traditional CS/ D conceptualization of
disconfirmation is that the standard is £11 internal (e.g., expectations,
desires) or pply external (e.g., best brand, experience based norms). In other
words, something important is excluded though there is strong theoretical and
empirical support that both internal and external standards are relevant.

With the proposed model, both internal and external factors are included in
the satisfaction formation process.

Finally, after data were collected for this research, the concept of
"desires congruency" was introduced to the literature (Spreng and Olshavsky
1993). This is basically a measure of the degree to which performance matches
desires. In the present research, such a measure of congruency was not

included, but a path from desires to disconfirmation of expectations was

245

estimated. This may have resulted in an inflation of the apparent effects of
disconfirmation on satisfaction; in the future, the more theoretically-
attractive construct of desires congruency should be included also, with it (and

not disconfirmation) mediating the effects of desires.

VI-4.3. Attitudes

The significance of some attitude - satisfaction links in the EOSM has
implications for the inclusion of attitude as an antecedent variable in future
research. As discussed in Section II-3.3, despite the reasonable theoretical
arguments for including attitude in satisfaction formation models, empirical
support for including this construct has been weak or non-existent. In the
EOSM, attitude towards Clinton was a significant predictor of satisfaction with
Clinton among _Bpph voters (Table V-6). In fact, among Bush voters, attitude
towards Clinton is a more important predictor of satisfaction with Clinton that
is expectation or even disconfirmation. In previous satisfaction research,
antecedents of satisfaction with nonchoices have not been included. Perhaps
it is attitude towards nonchoices which is more relevant and that is why
previous research has been unable to establish the relevance of attitude
empirically. In addition, attitude towards Clinton was a significant antecedent
of satisfaction with Clinton among satisfied Clinton voters only, indicating that
the relationship between the two constructs may be complex (e.g., non-
linear). For example, it is possible that voters with unfavorable attitudes
toward Clinton were dissatisfied because he won, those with moderately
favorable attitudes were satisfied because he won, and voters with extremely
favorable attitudes towards Clinton were less satisfied because Clinton did not

win by a very large margin, and in fact won with 4396 of the vote. (The press

246

and Clinton critics immediately pointed out that Clinton had not received a
"mandate" from the electorate.)

It is possible that the relationship between attitude and satisfaction was
confounded in this context because people were voting as much against a
candidate as they were for a candidate. Thus, it is possible that attitude towards
a nonchoice may affect satisfaction with one's choice, and vice versa. Also, it
is possible that attitudes are directly related to overall satisfaction, bypassing
satisfaction with the candidate. There is some evidence this occurred. Among
Bush voters, one modification index indicates model fit would be improved by
adding a path from attitude towards Bush to overall satisfaction, though the
path from attitude to satisfaction with Bush was not significant. The high
standardized residuals associated with attitude - satisfaction covariances
suggests that these relationships account for a large amount of the model's
lack of fit. Finally, the simple correlations for EOSM indicate that 1) among
Clinton voters, attitudes toward Bush are at least as strongly correlated with
satisfaction with Clinton as are attitudes toward Clinton, 2) among Bush voters,
attitudes towards Bush are nearly as highly correlated with satisfaction with
Clinton as are attitudes towards Clinton, and that attitude towards Clinton and
attitudes toward Bush are equally correlated with satisfaction with Bush, and 3)
attitudes toward Clinton and Bush are both highly associated with overall
satisfaction for Bush voters, but not for Clinton voters. Together, these results
indicate the importance of measuring attitudes toward nonchoices in
satisfaction research. From a managerial perspective, such a conclusion is not
surprising. In many situations consumers' negative attitude towards a product
or firm may play a significant, at times dominant, role in determining the

choice set and/ or specific nonchoices.

247

VI-4.4. Context of the Research

Though without precedent in the satisfaction literature, use of an
election context for studying satisfaction has proven useful. The expected
benefits of such a context (identified in Section I-S) did materialize.
Specifically, variance was found in the satisfaction measures, an objective
measure of performance did exist, expectations and desires were often
different and could be measured on scales directly comparable with each other
(and with the performance measure), "typical" antecedents of satisfaction
proved to be useful, competitive dimensions of the context were clear, etc.
While the study of satisfaction with the election results may or may not have
inherent value to political marketers, its value for theory-testing appears
clear. An important next step is to extend the analysis conducted here into a
more traditional marketing context to begin to establish the external validity
of key findings. First, situations which share certain competitive
characteristics with an election should be studied. For example, sports event
marketing or competitive bid situations in industrial marketing would be
logical next contexts for studying post-purchase satisfaction formation
processes. In both cases, competitive elements are obvious, choice sets are
clear and known in advance, there is an objective measure of performance,
pre-choice processing is likely to be widespread, and involvement is likely to
be high. Assuming it is demonstrated that multiple targets and interrelated
processes do operate in these more traditional contexts, the analysis could be
extended further. For example, typical consumer goods could be studied. As
long as nonchoices remain salient postchoice for any important consumer

segment, the competitive model of satisfaction formation could be utilized.

248

VI-S. Limitations

There are several limitations to the work presented here. While most
are the result of conscious trade-offs, and thus are balanced by associated
strengths, it is worthwhile to enumerate the most significant of these
limitations. limitations fall roughly into three categories: limitations due to
the research context, theoretical limitations, and methodological limitations.

Each will be briefly addressed.

Limitations Due to Context

First, the context of a political election may involve process and
dynamics very different from those found in more traditional marketing
contexts. While high external context validity is not critical to the basic
theoretical contributions of this dissertation, the non-traditional context does

limit direct extrapolation of research findings to other contexts.

Theoretical Limitations

First, this dissertation does not attempt to resolve, or even address, the
important conceptual or operationalization problems associated with any
individual construct. Thus, it is subject to all the criticisms regarding
construct validity found in the rest of the satisfaction literature. For example,
the EOSM and CPSM models utilized the most common conceptualization of
disconfirmation: subjective appraisal of the disconfirmation of predictive
expectations. In Section [1.3.5, the criticisms of this variable were discussed in
detail. Disconfirmation was operationalized as it was in this study to facilitate

comparison with previous research, though an alternative comparison

249

standard may have been more appropriate. Clearly this construct did not
effectively mediate the effects of desires, as well as a desires congruency
measure may have been able to.

Second, and more important, is the potential limitation resulting from
the omission of other possible antecedent variables. Omission of antecedent
variables from the EOSM, especially those which might logically be
uncorrelated with satisfaction with Bush or satisfaction with Clinton, could
result in model misspecification. For example, in an election context, the
quality of information about each candidate could affect expectations and
disconfirmation, and voters' satisfaction with information could also affect
overall satisfaction (as demonstrated by Spreng 1992). Similarly, the recent
intense interest in the role of affect on satisfaction formation (e.g., Westbrook
and Oliver 1991; Mano and Oliver 1993; Oliver 1993) indicates that affect can
have a powerful and unique effect on satisfaction formation; such potentially

important antecedent variables were not included in this analysis.

Methodological Limitations
First, the field setting of the research made it impossible to

tightly control for competing explanations of findings. This limitation affects
all field studies to varying degrees. Given that this dissertation represents a
first attempt at specifying the basic structure of a new model, it is impossible
to evaluate the potential consequences of this limitation.

Second, the sample was not randomly selected from the general
population. Again, this limitation is related to the external validity of research
results. While convenience sampling is often accepted as adequate for theory

development and theory testing (Calder, Phillips and Tybout 1981), non-

250

probability sampling schemes inhibit the generalizability of research
findings beyond a particular sample. In this research, even though attempts
were made to collect a sample which mirrored the electorate, inferences
beyond the sample should be made cautiously.

Third, use of single item constructs poses a limitation in that
measurement error cannot be estimated. One possible consequence is that
path estimates are less accurate than they would be had it been possible to

estimate measurement error.

The repeated calls for additional research and replicating key findings
in more traditional marketing settings were motivated, in part, by these

limitations.

VI-6. Concluding Remarks

The competitive model of satisfaction is based on the idea that
alternatives considered during the decision-making process of choosing
remain salient beyond the choice. Extending the explicit consideration of
competition from decision-making processes, with its extensive theoretical
and empirical literature to post-purchase processes, will help tie together the
entire consumption process. While the "emerging dialectic" of buying and
consuming (Oliver and Westbrook 1991) may yield useful micro-analytic
insights into bits of consumer behavior, it seems critical to remember that the
entire process of buying and consuming is ongoing and at least somewhat
repetitive with choice processes flowing into post-purchase processes flowing

into future choice processes, etc. This dynamic process perspective has begun

251

to receive attention in the literature (e.g., Boulding et a1 1993), but certainly
needs development.

Further, the potential contributions of this perspective to the emerging
focus on relationship marketing is clear. As firms seek to establish ongoing
relationships with their consumers, the firms are really trying to eliminate
other products/ services from consumer choice sets and to remain the only
salient target postchoice. Within the firm, the roles of "marketing manager"
and "customer service manager" are becoming blurred. For firms trying to
attract consumers from a competitor, knowledge of how customers form
satisfaction/ dissatisfaction becomes just as valuable as how consumers make
their choices in the first place, because in relationship marketing, the two
blend together. Previously, tools for investigating the competitive nature of
post-purchase processes did not exist in the satisfaction literature. Given the
trends towards forging long-term ties with customers, such tools are bound to

become important for both theoretical and managerial analyses.

LIST OF REFERENCES

LIST OF REFERENCES

Albrecht, Karl and Ron Zemke (1985), Service America! Doing Business in the
New Economy, Homewood, IL: Dow-Jones-Irwin.

Alderson, Wroe (1957), Marketing Behavior and Executive Action, Homewood,
11; Richard D. Irwin.

Anderson, Ralph E. (1973), "Consumer Dissatisfaction: The Effect of
Disconfirmed Expectancy on Perceived Product Performance," journal of
Marketing Research, 10:1 (February), 38—44.

Anderson, Eugene W. and Mary W. Sullivan (1993), "The Antecedents and
Consequences of Customer Satisfaction for Firms," Marketing Science, 12:2,
125-143.

Arnould, Eric J. and Linda 1.. Price (1993), "River Magic: Extraordinary
Experience and the Extended Service Encounter," Journal of Consumer
Research, 20: 1, 24—45.

Babin, Barry J., Mitch Griffin, and William R. Darden (in press), "An Empirical
Comparison of Alternative Conceptualizations of Postconsumption Reactions,"
Journal of Consumer Satisfaction, Dissatisfaction and Complaining Behavior.

Barbeau, J. Bradley (1984), "Predictive and Normative Expectations in
Consumer Satisfaction: A Utilization of Adaption and Comparison Levels in a
Unified Framework, " in Consumer Satisfaction, Dissatisfaction, and
Complaining Behavior Conference Proceedings, 27-32.

Bearden, William D. and Jesse E. Teel (1983), "Selected Determinants of
Consumer Satisfaction and Complaint Reports," Journal of Marketing Research,
20:4 (November), 21-28.

Bitner, Mary Jo (1990), "Evaluating Service Encounters: The Effect of
Physical Surroundings and Employee Responses," Journal of Marketing, 54
(April): 69-82.

Bitner, Mary Jo, and Amy R. Hubbert (1994), "Encounter Satisfaction Versus
Overall Satisfaction Versus Quality," in Service Quality: New Directions in
Theory and Practice, edited by Roland T. Rust and Richard L. Oliver, Thousand
Oaks, CA: SAGE Publications, Inc.

Bolfing, Claire P. and Robert B. Woodruff (1988), "Effects of Situational
Involvement on Consumers' Use of Standards in Satisfaction/ Dissatisfaction

Processes," Journal of Consumer Satisfaction, Dissatisfaction and Complaining
Behavior, 1, 16-24.

252

253

Boulding, William, Ajay Kalra, Richard Staelin, and Valerie A. Zeithaml (1993),
"A Dynamic Process Model of Service Quality: From Expectations to Behavioral
Intentions," Journal of Marketing Research, 30:1,7-27.

Brown, Stephen W. and Teresa A. Swartz (1989), "A Gap Analysis of Professional
Service Quality," Journal of Marketing, 53:2(April), 92-98.

Cadotte, Ernest R., Robert B. Woodruff, Roger L. Jenkins (1987), "Expectations
and Norms in Models of Consumer Satisfaction," Journal of Marketing
Research, 24:3, 305-3 14.

Calder, Bobby J., Lynn W. Phillips, and Alice M. Tybout (1981), "Designing
Research for Application," Journal of Consumer Research, 8:2,197-207.

Cardozo, Richard, N. (1965), "An Experimental Study of Customer Effort,
Expectation and Satisfaction," Journal of Marketing Research, 2:3 (August),
224-229.

Churchill, Gilbert A., Jr., and Carol Surprenant (1982), "An Investigation into
the Determinants of Consumer Satisfaction," Journal of Marketing Research,
19:4 (November), 491-504.

Churchill, Gilbert A., Jr. (1983), Marketing Research--Methodological
Foundations, Chicago, IL: The Dryden Press.

Cooper, Ann R., M. Bixby Cooper, and Dale F. Duhan (1989), "Measurement
Instrument Developed Using Two Competing Concepts of Customer
Satisfaction," Journal of Consumer Satisfaction, Dissatisfaction and
Complaining Behavior, Volume 2, 28-35.

Cronin, Joseph J. and Steven A. Taylor (1992), "Measuring Service Quality: A
Reexamination and Extension," Journal of Marketing, 56:3,55-68.

Czepiel, John A. and Larry J. Rosenberg (1977), "Customer Satisfaction:
Concept and Measurement," Journal of the Academy of Marketing Science, 5:4
(Fall), 403-41 1.

Day, Ralph L. (1984), "Modeling Choices Among Alternative Responses to
Dissatisfaction," in Advances in Consumer Research, Vol. 11, Thomas C.
Kinnear, ed. Ann Arbor, MI: Association of Consumer Research, 496-499.

Droge, Cornelia, and Diane Halstead (1991),"Postpurchase Hierarchies of
Effects: The Antecedents and Consequences of Satisfaction for Complainers
versus Non-complainers," International Journal of Research in Marketing,
8:4(November), 315-328.

Duhaime, Carole P. (1988), "Consumer Satisfaction with the Distribution System
for Durable Products," Journal of Consumer Satisfaction, Dissatisfaction, and
Complaining Behavior, 1, 53-59.

Engel, James R, David J. Kollet, and Roger D. Blackwell (1973), Consumer
Behavior, Hinsdale, IL: Dryden.

254

Festinger, Leon (1957), A Theory of Cognitive Dissonance, Stanford, CA:
Stanford University Press.

Fishbein, Martin (1967), "Attitude and the Prediction of Behavior," in Readings
in Attitude Theory and Measurement, M. Fishbein, ed., NY: John Wiley, 477-492

Fisk, Raymond P., and Clifford E. Young (1985), "Disconfirmation of Equity
Expectations: Effects on Consumer Satisfaction with Services," in Advances in
Consumer Research, E. Hirschman and H. Holbrook, eds., Provo, UT: Association
for Consumer Research.

Fornell, Claes (1992), "A National Customer Satisfaction Barometer: The
Swedish Experience," Journal of Marketing, 56:1,6-21.

Gardial, Sarah Fisher, D. Scott Clemons, Robert B. Woodruff, David W.
Schumann, and Mary Jane Burns (1994), "Comparing Consumers' Recall of
Prepurchase and Postpurchase Product Evaluation Experiences," Journal of
Consumer Research, 20 (March), 548-560.

Garland, Barbara C. and Robert A. Westbrook (1989) "An Exploration of Client
Satisfaction in a Nonprofit Context."Journal of the Academy of Marketing
Sciences, 17:4, 297-303.

Granbois, Donald, John O. Summers, and Gary l. Frazier (1977), "Correlates of
Consumer Expectations and Complaining Behavior," in Consumer Satisfaction,

Dissatisfaction, and Complaining Behavior, Proceedings of the Second Annual
CS/D&CB conference, ed. Ralph Day, 18-25.

Halstead, Diane (1992), "The Expectations-Satisfaction Relationship Revisited:
An Empirical Test and Directions for Future Research," in Proceedings of the
1 992 AMA Winter Educa tors' Conference.

Herzberg, Frederick, B. Mausner, and B. Snyderman (1959), The Motivation to
Work, New York, NY: John Wiley & Sons, Inc.

Holbrook, Morris B., and Elizabeth C. Hirshman (1982), "The Experiential
Aspects of Consumption: Consumer Fantasies, Feelings, and Fun," The Journal
of Consumer Research, 9:2,132-140.

Houston, David A., Steven J. Sherman, and Sara M. Baker (1991), "Feature
Matching, Unique Features, and the Dynamics of the Choice Process:

Predecision Conflict and Postdecision Satisfaction," Journal of Experimental
Social Psychology, 27:5, 41 1-430.

Howard, John A., and Jagdish N. Sheth (1969), The Theory of Buyer Behavior,
New York, NY: John Wiley & Sons.

Hunt, H. Keith (1977), Customer Satisfaction and Dissatisfaction: Perspectives
and Overview, Cambridge, MA: Marketing Science Institute (Report No. 77-
112).

255

Hunt, Shelby D. (1991), Modern Marketing Theory - Critical Issues in the
Philosophy of Marketing Science, Cincinnati, OH: South-Western Publishing
Co.

Izard, C. E. (1984), "Emotion-Cognition Relationships and Human
Development," in C. E. Izard, J. Kagan, and R. B. Zajonc, eds., Emotion,
Cognition, and Behavior, New York: Cambridge University Press.

Jereskog, Karl G. and Dag Sbrbom (1989), LISREL 7 - A Guide to the Program and
Applications, second edition, Chicago, Il: SPSS Inc.

Kahneman, D. and A. Tversky (1979), "Prospect Theory: An Analysis of
Decision Under Risk," Econometrica, 47, 952-963.

Katona, G. (1964), The Mass Consumption Society, NY: McGraw Hill.

Kennedy, John R., and Peter C. Thirkell (1988), "An Extended Perspective on
the Antecedents of Satisfaction, " Journal of Consumer Satisfaction,
Dissatisfaction and Complaining Behavior, 1, 2-10.

Kerlinger, Fred, N. (1986), Foundations of Behavioral Research, Third Edition,
Fort Worth, TX: Holt, Rinehart and Winston, Inc.

Kotler, Philip (1980), Marketing Managemen t--Analysis, Planning and
Control, Englewood Cliffs, NJ: Prentice-Hall, Inc.

LaBarberra, Priscilla, and David Mazursky (1983), "A Longitudinal Assessment
of Consumer Satisfaction/ Dissatisfaction: The Dynamic Aspect of the Cognitive
Process," Journal of Marketing Research, 20:9,393-404.

LaTour, Stephen A. and Nancy C. Peat (1979), "Conceptual and Methodological
Issues in Consumer Satisfaction Research," in Advances in Consumer
Research, Vol. 6, William L. Wilkie, ed. Ann Arbor, MI: Association for
Consumer Research, 43 1-43 7.

Lazarus, R. S. (1984), "On the Primacy of Cognition," American Psychologist, 39,
124-129.

Lele, Milind M., and Jagdish N. Sheth (1987), The Customer is Key, New York,
NY: John Wiley & Sons.

Loomes, Graham, and Robert Sugden (1982), "Regret Theory: An Alternative
Theory of Rational Choice Under Uncertainty," The Economic Journal, 92
(December), 805-824.

Lord, Frederick M. (1963), "Elementary Models for Measuring Change," in C. W.
Harris, ed., Problems in Measuring Change, Madison, WI: University of
Wisconsin Press.

Maddox, R. Neil (1981), "Two-Factor Theory and Consumer Satisfaction:
Replication and Extension," Journal of Consumer Research, 8:1,97-102.

256

Mano, Haim, and Richard L. Oliver (1993), "Assessing the Dimensionality and
Structure of the Consumption Experience: Evaluation, Feeling, and
Satisfaction," Journal of Consumer Research, 20 (December).

Miller, John A. (1977), "Studying Satisfaction, Modifying Models, Eliciting
Expectations, Posing Problems, and Making Meaningful Measurements," in
Conceptualiza tion and Measurement of Consumer Satisfaction and
Dissatisfaction, H. K

Hunt, ed.

Mizerski, Richard W. (1982), "An Attribution Explanation of Disproportionate
Influence of Unfavorable Information," Journal of Consumer Research, 9
(December), 301-308.

Nystrom, Paul H. (1929), The Economic Principles of Consumption, The Ronald
Press Co.

Oliver, Richard L (1980), "A Cognitive Model of the Antecedents and
Consequences of Satisfaction Decisions," Journal of Marketing Research, 17:4
(November), 460-469.

Oliver, Richard L. (1981), "Measurement and Evaluation of Satisfaction
Processes in Retail Settings," Journal of Retailing, 57:3 (Fall), 25-48.

Oliver, Richard L. (1993), "Cognitive, Affective, and Attribute Bases of the
Satisfaction Response," Journal of Consumer Research, 20 (December). -

Oliver, Richard L. and William O. Bearden (1983), "The Role of Involvement in
Satisfaction Processes," in Advances in Consumer Research, Volume 10, eds.
Richard P. Bagozzi and Alice M. Tybout, Ann Arbor, MI: Association for
Consumer Research, 250-255.

Oliver, Richard L. and William O. Bearden ( 1985), "Disconfirmation Processes
and Consumer Evaluations in Product Usage," Journal of Business Research,
13:2,235-246.

Oliver, Richard L. and Wayne S. DeSarbo (1988), "Response Determinants in
Satisfaction Judgments," Journal of Consumer Research, 14:4 (December), 372-
383.

Oliver, Richard L. and Gerald Linda (1981), "Effect of Satisfaction and Its
Antecedents on Consumer Preference and Intention," in Advances in
Consumer Research, Volume 8, Kent Monroe, ed., Ann Arbor, MI: Association
for Consumer Research.

Oliver, Richard L. and John E. Swan (1989), "Equity and Disconfirmation
Perceptions as Influences on Merchant and Product Satisfaction," Journal of
Consumer Research, 16:4 (December), 372-3 83.

Olshavsky, Richard W. and John A. Miller (1972), "Consumer Expectation,
Product Performance and Perceived Product Quality," Journal of Marketing
Research, 9:1 (February), 19-21.

257

Olson, Jerry C., and Philip A. Dover (1979), "Disconfirmation of Consumer
Expectations Through Product Trial," Journal of Applied Psychology, 64:2,179-
189.

Parasuraman, A., Valerie A. Zeithaml, and Leonard 1.. Berry (1988), "SERVQUAI:
A Multi-Item Scale for Measuring Consumer Perceptions of Service Quality,"
Journal of Retailing, 64:1, 12-40.

Patterson, Paul G. (1993), "Expectations and Product Performance as
Determinants of Satisfaction for a High-Involvement Purchase," Psychology
and Marketing, 10:5,449-465.

Peter, J. Paul, Gilbert A. Churchill, Jr., and Tom J. Brown (1993), "Caution in the
Use of Difference Scores in Consumer Research," Journal of Consumer
Research, 19 (March), 655-662.

Peterson, Robert A. and William R. Wilson (1992), "Measuring Customer
Satisfaction: Fact and Artifact," Journal of the Academy of Marketing Science,
20:1,61-71.

Petty, Richard E., and John T. Cacioppo (1986), "The Elaboration likelihood
Model of Persuasion," in Advances in Experimental Social Psychology, Vol. 19,
edited by L. Berkowitz, FL: Academic Press, 123-205.

Prakash, Ved and John W. Lounsbury (1984), "The Role of Expectations in the
Determination of Consumer Satisfaction," Journal of the Academy of Marketing
Science, 12:3 (Summer), 1-17.

Reichheld, Frederick F. and W. Earl Sasser, Jr. (1990) "Zero Defections: Quality
Comes to Services," Harvard Business Review, 68:5 (September-October): 105-
111.

Richins, Marsha L. (1983), "Negative Word-of-Mouth by Dissatisfied Consumers:
A Pilot Study," Journal of Marketing, 47 (Winter), 68-78.

Richins, Marsha L. and Peter H. Bloch (1991), "Post-Purchase Product
Satisfaction: Incorporating the Effects of Involvement and Time," Journal of
Business Research, 23:2, 145-158.

Sherif, M., and C. I. Hovland (1961), Social Judgment: Assimilation and
Contrast Effects in Communication and Attitude Change, New Haven, CT: Yale
University Press.

Sheth, Jagdish M., David M. Gardner, and Dennis E. Garrett (1988), Marketing
Theory: Evolution and Evaluation, New York: John Wiley & Sons.

Singh, Jagdip (1991) "Understanding the Structure of Consumers' Satisfaction
Evaluations of Service Delivery," Journal of the Academy of Marketing
Science, 19:3 (Summer): 223-244.

Sirgy, Joseph M. (1984), "A Social Cognition Model of Consumer Satisfaction/
Dissatisfaction," Psychology and Marketing, 1:2,27—43.

258

Spreng, Richard A (1992), A Comprehensive Model of the Consumer
Satisfaction Formation Process, unpublished doctoral dissertation, University
of Indiana.

Spreng, Richard A., Andrea L Dixon, and Richard W. Olshavsky (1993), "The
Impact of Value Judgments on Consumer Satisfaction," Journal of Consumer
Satisfaction, Dissatisfaction, and Complaining Behavior, in press.

Spreng, Richard A. and Andrea L. Dixon (1992), "Alternative Comparison
Standards in the Formation of Consumer Satisfaction/ Dissatisfaction,"
Enhancing Knowledge Development in Marketing, 1989 Educators'
Proceedings, Chicago, IL: American Marketing Association.

Spreng, Richard A., and Richard W. Olshavsky (1993), "A Desires Congruency
Model of Consumer Satisfaction," Journal of the Academy of Marketing
Science, 21:3, 169-178.

Swan, John E. and linda J. Combs (1976), "Product Performance and Consumer
Satisfaction: A New Concept," Journal of Marketing, 40:2, 25-33.

Swan, John E. and Warren S. Martin (1981), "Testing Comparison Level and
Predictive Expectations Models of Satisfaction, " in Advances in Consumer
Research, Volume 8, Kent Monroe, ed., Ann Arbor, MI: Association for
Consumer Research, 77-82.

Swan, John E. and 1. Frederick Trawick (1981), "Disconfirmation of
Expectations and Satisfaction with a Retail Service," Journal of Retailing, 57:3
(Fall), 49-67.

Tichy, Noel M. and Stratford Sherman (1993), "Jack Welch's Lessons for
Success," Fortune, 127:2, 86-93.

Tse, David K. and Peter C. Wilton (1988), "Models of Consumer Satisfaction
Formation: An Extension," Journal of Marketing Research, 25:2 (May), 204-212.

Westbrook, Robert A. (1980), "Intrapersonal Affective Influences Upon
Consumer Satisfaction with Products," Journal of Consumer Research, 7:4, 49-
S4.

Westbrook, Robert A. and Joseph W. Newman (1978), "An Analysis of Shopper
Dissatisfaction for Major Household Appliances," Journal of Marketing
Research, 15:3, 456-466.

Westbrook, Robert A. and Richard 1.. Oliver (1991), "The Dimensionality of
Consumption Emotion Patterns and Consumer Satisfaction," Journal of
Consumer Research, 18: 1, 84-9 1.

Westbrook, Robert A., and Michael K. Reilly (1983), "Value-Percept Disparity:
An Alternative to the Disconfirmation of Expectations Theory of Consumer
Satisfaction," in Advances in Consumer Research, Volume 10, eds. Richard P.
Bagozzi and Alice M. Tybout, Ann Arbor, MI: Association for Consumer
Research, 256-261.

259

Woodruff, Robert B., Ernest R. Cadotte, and Roger L. Jenkins (1983), "Modeling
consumer Satisfaction Processes Using Experience-Based Norms," Journal of
Marketing Research, 20:3 (August), 296-304.

Yi, Youjae (1990), "A Critical Review of Consumer Satisfaction," in Review of
Marketing 1990, edited by Valarie A. Zeithaml, Chicago, IL: American
Marketing Association.

Zajonc, R. B. (1980), "Feeling and Thinking: Preferences Need No Inferences,"
American Psychologist, 35, 15 1-175.

Zajonc, R. B. (1984), "On the Primacy of Affect," American Psychologist,39:2,
117-123.

Zaichowsky, Judith Lynn (1985), "Measuring the Involvement Construct,"
Journal of Consumer Research, 12 (December), 341-352.

Ill/ills

mil

61766

I

v ’1'!
I 1”

I;
U H
E II

0104

I!
l
I
l!
I]
(I
TI
I
I
1’
I
I
I
(I
I!
1’
I
i
I
I.
I

31293

ill!