The Antecedents and Consequences of Customer Satisfaction for Firms

Author(s): Eugene W. Anderson and Mary W. Sullivan

Source: Marketing Science, Vol. 12, No. 2 (Spring, 1993), pp. 125-143
Published by: INFORMS
Stable URL: https://www.jstor.org/stable/184036
Accessed: 27-04-2025 01:25 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Marketing
Science

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
MARKETING SCIENCE
Vol. 12, No. 2, Spring 1993
Printed in U.S.A.

THE ANTECEDENTS AND CONSEQUENCES OF

CUSTOMER SATISFACTION FOR FIRMS
EUGENE W. ANDERSON AND MARY W. SULLIVAN
University of Michigan
University of Chicago

This research investigates the antecedents and consequences of customer satisfaction. We develop
a model to link explicitly the antecedents and consequences of satisfaction in a utility-oriented
framework. We estimate and test the model against alternative hypotheses from the satisfaction
literature. In the process, a unique database is analyzed: a nationally representative survey of
22,300 customers of a variety of major products and services in Sweden in 1989-1990. Several
well-known experimental findings of satisfaction research are tested in a field setting of national
scope. For example, we find that satisfaction is best specified as a function of perceived quality
and "disconfirmation"-the extent to which perceived quality fails to match prepurchase expec-
tations. Surprisingly, expectations do not directly affect satisfaction, as is often suggested in the
satisfaction literature. In addition, we find quality which falls short of expectations has a greater
impact on satisfaction and repurchase intentions than quality which exceeds expectations. More-
over, we find that disconfirmation is more likely to occur when quality is easy to evaluate. Finally,
in terms of systematic variation across firms, we find the elasticity of repurchase intentions with
respect to satisfaction to be lower for firms that provide high satisfaction. This implies a long-run
reputation effect insulating firms which consistently provide high satisfaction.
(Choice Models; Defensive Strategy; Customer Satisfaction; Perceived Quality)

1. Introduction

There is growing managerial interest in customer satisfaction as a means of evaluating

quality. High customer satisfaction ratings are widely believed to be the best indicator
of a company's future profits (Kotler 1991, p. 19). Firms increasingly use customer
satisfaction as a criterion for diagnosing product or service performance and often tie
customer satisfaction ratings to both executive and employee compensation.' However,
providing incentives to maximize customer satisfaction may actually be detrimental to
the firm. To encourage actions which will lead to an optimal level of satisfaction, it is
necessary to understand the link between the antecedents of satisfaction and satisfaction's
behavioral and economic consequences.
The antecedents of satisfaction have long been a subject of study for consumer research
(e.g., Cardozo 1965, Churchill and Suprenant 1982; Oliver 1977, 1980; Oliver and
DeSarbo 1988; Oliver and Bearden 1985; Tse and Wilton 1988; Westbrook 1981; Yi
1991), but relatively few studies investigate the consequences of satisfaction (Bearden

'"King Customer," Business Week, March 12, 1990.

125

0732-2399/93/1202/01 25$0 1.25

Copyright ? 1993, The Institute of Management Sciences/Operations Research Society of America

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
126 EUGENE W. ANDERSON AND MARY W. SULLIVAN
and Teel 1983; Oliver and Swan 1989). In marketing science and managerial economics,
relatively few studies investigate the antecedents of satisfaction, but several aspects of
postsatisfaction behavior are examined. Theoretically, Hirschman (1970) identifies con-
ditions under which dissatisfied customers will complain or switch. Fornell and Wernerfelt
(1987, 1988) show conditions when firms should encourage dissatisfied customers to
complain. Empirically, Andreasen (1985) finds support for Hirschman's (1970) prediction
that dissatisfied customers will police product quality in absolute monopolies.
Hence, there is a need for developing a deeper understanding of the linkage between
the antecedents and consequences of satisfaction. The objective of this paper is to in-
vestigate-both analytically and empirically-the antecedents and behavioral conse-
quences of satisfaction. A model is developed in order to understand and predict relations
between the antecedents and consequences of customer satisfaction, as well as systematic
differences in these relations across firms. The micromodel specifies how customer ex-
pectations of product quality interact with the actual experience of product quality to
generate satisfaction, and how satisfaction influences the likelihood of subsequent pur-
chases. In developing such a model, the antecedents and consequences of satisfaction
are brought together in a single, utility-oriented framework.
The model is estimated and tested against alternative hypotheses using a unique data-
base obtained by sampling 22,300 customers of 114 companies in 16 major product and
service industries in Sweden. Several well-known experimental findings of satisfaction
research are tested in a field setting of national scope. For example, we find that satisfaction
is best specified as a function of perceived quality and "disconfirmation"-the extent to
which perceived quality fails to match prepurchase expectations. Surprisingly, expectations
do not directly affect satisfaction, as often suggested in the satisfaction literature. In
addition, we find quality which falls short of expectations has a greater impact on sat-
isfaction and repurchase intentions than quality which exceeds expectations. Moreover,
we find that disconfirmation is more likely to occur when quality is easy to evaluate.
Finally, in terms of systematic variation across firms, the data suggest support for the
model's prediction that the elasticity of repurchase intentions with respect to satisfaction
will be lower for firms that provide high satisfaction. This implies a long-run reputation
effect insulating firms which consistently provide high satisfaction.

2. Analytical Framework and Hypotheses

Satisfaction can be broadly characterized as a postpurchase evaluation of product

quality given prepurchase expectations (Kotler 1991). Yi (1991) provides an excellent
review of customer satisfaction and the main antecedents identified by consumer research:
expectations, perceived quality, and disconfirmation. Beginning with Oliver (1977, 1980),
research concerned with the antecedents of satisfaction focuses primarily on the expec-
tancy-disconfirmation paradigm illustrated in Figure 1 a. Drawing on adaptation level
theory (Helson 1964), Oliver (1980) posits expectations as an adapted standard which
provide a frame of reference for buyers' evaluative judgements. Accordingly, satisfaction
judgements are a function of the baseline effect of expectations plus any perceived dis-
confirmation of expectations.
Oliver (1980) describes the process by which satisfaction judgements are reached in
the expectancy-disconfirmation framework. First, buyers form expectations of the specific
product or service prior to purchase. Second, consumption reveals a perceived quality
level which is influenced by expectations if the difference between actual quality and
expectations is perceived as being small. Hence, perceived quality may increase or decrease
directly with expectations as indicated by the arrow drawn from expectations to perceived
quality in Figure 1a. Third, perceived quality may either confirm or disconfirm prepur-
chase expectations. The determination of the extent to which perceived quality expec-

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
ANTECEDENTS AND CONSEQUENCES OF CUSTOMER SATISFACTION 127

ationson + e

~~~~~- +

Perceived

FIGURE I a

tations are disconfirmed is depicted in Figure 1 a by arrows drawn from expectations and
perceived quality to disconfirmation. Fourth, as shown by the arrows in Figure la, sat-
isfaction is positively affected by expectations and the perceived level of disconfirmation.
Expectations provide a baseline or anchor for level of satisfaction. If disconfirmation is
perceived to have occurred, then customer satisfaction increases or decreases from this
baseline level.
However, while there is general consensus that disconfirmation is an important an-
tecedent of satisfaction, there is mixed empirical evidence as to whether expectations
directly affect satisfaction (Yi 1991 ). For example, Churchill and Suprenant ( 1982) find
that perceived quality rather than expectations directly affects satisfaction for durables.
Oliver and DeSarbo ( 1988) find disconfirmation and perceived quality to have a stronger
impact on satisfaction than expectations. Hence, there is reason to believe that perceived
quality may play a more direct role in determining satisfaction than accounted for by
expectancy-disconfirmation.
We introduce an analytical framework in which satisfaction is a function of perceived
quality and disconfirmation. Figure lb provides an overview of our model. As in the

EHpect- Evalua ting Ease of

a tions Quality
Neg tive

os G l
Perceived

FIGURE 1lb

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
128 EUGENE W. ANDERSON AND MARY W. SULLIVAN
expectancy-disconfirmation paradigm, expectations are expected to have a direct positive
effect on perceived quality. However, expectations affect satisfaction only via perceived
quality and disconfirmation. In addition, the model proposes an asymmetric gain-loss
framework for understanding the effect of disconfirmation on satisfaction. As shown in
Figure lb, the disconfirmation construct has been redefined in our framework as having
both a negative and positive component with separate effects on satisfaction. Moreover,
we propose the ease of evaluating quality as an important moderating influence on the
extent of disconfirmation. As discussed below, ease of evaluating quality is expected to
lead to greater disconfirmation in both directions. Finally, we expect satisfaction to have
a positive impact on repurchase intentions.
The utility-oriented model developed below offers an axiomatic basis for understanding:
(1 ) why satisfaction is appropriately modeled as a function of perceived quality and
disconfirmation; (2) why expectations are expected to influence perceived quality; (3)
how the ease or difficulty of evaluating quality can moderate the size of perceived dis-
confirmation; (4) how a rational consumer might form expectations; and (5) how past
satisfaction should influence future repurchase occasions. Moreover, the model formalizes
a number of diverse findings from consumer research into a single, coherent utility-
oriented framework for understanding the antecedents and consequences of satisfaction.
The model is developed in the remainder of this section. In the process, testable hypotheses
are derived and discussed.

The Satisfaction Function

To begin, we posit satisfaction as a function of perceived quality and disconfirmation.
Equation ( 1 ) presents a general expression for satisfaction as directly affected by perceived
quality and disconfirmation. In this formulation, perceived quality is represented by the
utility derived from consumption. Satisfaction is generated by the level of utility plus
any gain or loss experienced if perceived product utility is different from expected product
utility:

Sijt = fl ( Upt) + f2 ( Uipj-t ), (1)

f(?)=?, f'l>O, f'<09
> 0 if Up -, < 0,
f2(0) = 0, f2 > f? f '2 < if UP _ > where
< <0 ifUij,-/4,>0,
Sit= satisfaction from brand j for customer i at time t,
UPJt= perceived quality of brand j for customer i at time t,
,Ut= expectation of brand j's quality for customer i at time t,
fi ( ) = concave function for the impact of perceived quality on satisfaction, and
f2( ) = asymmetric loss function for the impact of confirmation on satisfaction.
The first term in Equation ( 1), ft ( Upt), represents the direct impact of perceived
product quality UpJit on satisfaction. Hence, Up? is analogous to Thaler's ( 1985) notion
of acquisition utility. Moreover, as shown in Figure 2a, the direct impact of perceived
quality on satisfaction, f, ( UPJt), is expected to increase at a decreasing rate.
If perceived quality exceeds (falls short of) expectations, then satisfaction is a function
of the direct impact of quality fi ( UsPJt) plus a gain (loss) in satisfaction from the unexpected
surprise of finding quality to be greater (less) than expectations, f2( UPjt - v,Ui). Whether
or not disconfirmation occurs is discussed in the next subsection. As shown in Figure
2b, perceived disconfirmation is characterized as an asymmetric loss function which is
concave for positive disconfirmation, and convex for negative disconfirmation. This por-
tion of the satisfaction function is similar in form to the value function of Prospect
Theory (Kahneman and Tversky 1979). As indicated by the illustration, losses are ex-

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
ANTECEDENTS AND CONSEQUENCES OF CUSTOMER SATISFACTION 129

12

/ ~~~~~~~~~~~~~UP7t -,sj
u~~~~~~~~~.~~~P

UP.7

FIGURE 2a FIGURE 2b
FIGURE 2. Illustration of the Satisfaction Function.

pected to loom larger than gains (Einhorn and Hogarth 1981, Hogarth 1987). Hence,
the perceived gap between perceived quality and expectations, U?, - y4,, is analogous
to Thaler's (1985) notion of transaction utility. The proposed asymmetric loss function
has the straightforward implication that satisfaction should be more sensitive to negative
disconfirmation than positive disconfirmation.
The major hypotheses associated with Equation (1) are summarized in Table 1. As
shown, the hypothesis that satisfaction is an increasing function of perceived quality and
disconfirmation is denoted as the Quality-Disconfirmation hypothesis, X. The hypothesis
that satisfaction is an increasing function of expectations and disconfirmation as posited
in the satisfaction literature is summarized in Table 1 as the Expectancy-Disconfirmation,
12. Hence, XI 2 represents the key alternative hypothesis to the Quality-Disconfirmation
hypothesis derived from the model. Finally, the hypothesis of an asymmetric loss function
for the effect of disconfirmation on satisfaction is summarized in Table 1 as the Asym-
metric-Disconfirmation hypothesis, .21

Confirmation of Expectations as a Function of Assimilation

Extensive research finds that perceived product quality is positively affected by expec-
tations of product quality (e.g., Anderson 1973, Deighton 1984, Hoch and Ha 1986,
Oliver 1977, Olshavsky and Miller 1972, Olson and Dover 1979). For example, Anderson
(1973) manipulates expectations for ballpoint pens and finds subsequent judgements of
product quality move in the direction of the manipulated expectations. This movement
of perceived quality in the.direction of expectations is called "assimilation". In general,
assimilation theory implies that if the difference between expectations and product quality
is small enough to fall within a customer's "latitude of acceptance," then perceived
quality will tend to move closer to expectations (Sherif and Hovland 1961). Hence,
Figure lb indicates a direct relation between expectations and quality. Higher (lower)
expectations should lead to higher (lower) quality.
In related research, Deighton (1984) and Hoch and Ha (1986) draw an analogy between
the comparison of expectations with product quality and hypothesis testing. For example,
if advertising creates an expectation of a certain type of feeling, than this expectation
acts as a hypothesis which consumers are likely to confirm when exposed to product
information. Hoch and Ha (1986) not only find evidence of the assimilation effect, but
also find that the effect is moderated by ambiguity. If product quality is difficult to judge,
then assimilation is more likely to occur. Hence, increased ambiguity increases the latitude
of acceptance associated with the null hypothesis that perceived quality equals expecta-
tions. Conversely, if product quality is easy to judge (lower ambiguity), then the latitude

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
130 EUGENE W. ANDERSON AND MARY W. SULLIVAN
TABLE I
Major Null and Alternative Hypotheses

Hypothesis Description
#,Y ,: Quality-Disconfirmation Satisfaction is an increasing function of perceived quality and
disconfirmation.

lh12: Expectancy-Disconfirmation Satisfaction is an increasing function of expectations and

disconfirmation.

#2:1 Asymmetric-Disconfirmation Satisfaction is more sensitive to negative disconfirmation than

positive disconfirmation.

,W31. Assimilation Perceived quality is directly increasing with expectations.

*4.: Evaluation As the ease of evaluating quality increases, negative and
positive disconfirmation will increase.

h42 Information As the ease of evaluating quality increases, negative and

positive disconfirmation will decrease.

.)ri: Insulation Firms consistently providing high satisfaction will develop a

positive reputation. Hence, these firms should be insulated
from deviations in satisfaction, perceived quality and
disconfirmation.

1(52: Immediate Response Firms providing high satisfaction do so in response to demand

being more sensitive to deviations in satisfaction, perceived
quality and disconfirmation.

of acceptance will decrease and it is more likely that the null hypothesis will be rejected
(disconfirmation will occur).
To summarize, assimilation effects occur when the difference between expectations
and quality is too small to be perceived. Moreover, the ease of evaluating quality may
influence the likelihood that assimilation will occur. The analogy to hypothesis testing
used by Deighton (1984) and Hoch and Ha (1986) appears strong enough to warrant
its use in modeling the effects of assimilation and ease of evaluating quality on perceived

=,~ i l (2)
quality and, subsequently, on satisfaction. Accordingly, Equation (2) captures how per-
ceived quality Upt is influenced by whether or not a null hypothesis comparing expec-
tations yq4t and perceived quality Up, can be rejected:

4,j, if Ho: Up? E [u/4, ? boj),] cannot be rejected,

UIP if Ho: Up, E [Ai'4j, ? bW'] is rejected, where

Ue. Fe. (It.,uq 3
v q.,= standard deviation of expectations for brand j's quality for customer i at time t,
b = constant representing a noticeable departure from expectations (number of
ff Ij' S ),

Uq.t = individual i's expectation for brand j's quality at time t, and
Fq. = individual i's distribution for expected product utility of brand j at time t.
According to Equation (2), if the null hypothesis of equality Ho cannot be rejected,
then perceived product utility at time t, Up,, is equal to expectations ,u.(,. In such a case,
expectations are "confirmed," and the buyer perceives no difference between quality and
expectations (Up, - Au' ) = 0. Alternatively, if Ho is rejected, then perceived product
quality at time t, UP$,, remains equal to product quality Up,. In this case, the perceived
difference is large enough to become noticeable, and positive or negative disconfirmation
occurs. As long as disconfirmation does not occur, then we expect product quality to

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
ANTECEDENTS AND CONSEQUENCES OF CUSTOMER SATISFACTION 131

increase or decrease with expectations. We summarize this implication in Table 1 as the

Assimilation Hypothesis, F31, which states that product quality should be directly in-
creasing with expectations.
Whether or not confirmation occurs depends on the difficulty of discriminating between
quality and expectations. In terms of the model, Equation (3) presents the expected
utility of a brand as an individual buyer's subjective belief expressed as a distribution
F',, with mean '4, and standard deviation crc,,. A tight distribution of expectations may
reflect customer expertise, advertising exposure and personal experience, product com-
plexity, reliability, or product differentiation (Alba and Hutchinson 1987, Hoch and Ha
1986, Klayman and Ha 1986). Hence, the standard deviation of expectations, {',, reflects
the ease with which buyers discriminate between fine differences in quality. Therefore,
Equations (2) and (3) imply that both negative and positive disconfirmation will be
greater as the ease of evaluating quality increases, cri, -* 0. This implication of the model
is summarized as the Evaluation Hypothesis, 41.
An important alternative hypothesis to the Evaluation Hypothesis comes from research
on the economics of information (Stigler 1961, Nelsen 1970) and is summarized in Table
1 as the Information Hypothesis, 42. Specifically, Nelsen ( 1970) posits two broad cat-
egories of products: search and experience. Search goods are products and services which
can be evaluated prior to purchase. Experience goods are products and services which
must be experienced before their quality can be evaluated. Although Nelsen ( 1970) does
not directly address the issues of assimilation or satisfaction, his classification implies
that we should expect more disconfirmation to occur for experience goods than for search
goods. Experience goods must be sampled to determine their quality and the likelihood
of both "good" and "bad" surprises (positive and negative disconfirmation) should be
correspondingly higher. If experience goods are equivalent to products which are difficult
to evaluate, than this prediction is the opposite of the implications derived from our
model. Similarly, if quality is easier to evaluate for search goods customers should be
better able to avoid poor quality products. Hence, in a mature category where quality is
easy to evaluate, we would expect a larger proportion of customers to have identified
products which consistently provide satisfaction. If this is the case, then we would expect
to observe less negative disconfirmation when product quality is easy to judge.

A Bayesian Framework for Relating Satisfaction to Choice

We can integrate each of the aforementioned components of the model and formally
relate satisfaction to choice behavior by means of Bayesian decision theory (Berger 1985).
Although a Bayesian framework may not provide a good normative model of consumer
psychology in terms of the optimality of behavior (Kahneman and Tversky 1972). Mas-
saro and Friedman ( 1990) demonstrate that a Bayesian approach to updating and choice
behavior provides a reasonable approximation of observed behavior when all sources of
information are viewed as subjective. In the present study, product experience is inherently
subjective. Hence, it seems reasonable to model how expectations adapt to product ex-
perience using Bayesian updating and decision theory (Berger 1985). Hence, a consumer
at time t will choose a brand, j, from the set of available brands in the product market,
QB, which will maximize expected satisfaction (minimize expected loss) as given by
Equation ( 1). Formally, the choice problem can be written as

jEQ= a j
max fJ' SijtdFi,t( 'tk,u,jt

=C max | fi UPjt) +t(jt ,aet ]Fljt iejt I u W), where, (4)

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
132 EUGENE W. ANDERSON AND MARY W. SULLIVAN
Sijt = satisfaction from brand j for consumer i at time t, and
QB= set of available brands { bI, . . ., bj, . .. , bj} .
According to Equation (4), satisfaction is related to repurchase intentions using the
satisfaction function (Equation ( 1 )) as a loss function and the distribution of expectations
(Equation (3)) as the prior distribution. Buyers are expected to purchase brands which
they anticipate will maximize their satisfaction. Hence, if a consumer expects the satis-
faction from a brand to be high, then the consumer is more likely to choose that brand.
Following a negative product experience with a certain brand, a consumer should be
more likely to switch to another supplier. The distribution of expectations will be revised
to have a lower mean and a greater probability that the utility experienced will be low.
Alternatively, more satisfied customers are more likely to remain customers. In this case,
the distribution of expectations should be revised upwards. The mean expected utility
should be higher. Hence, increased satisfaction should lead to increased repurchase in-
tentions. Increased repurchase intentions increase the probability of repurchase and,
consequently, the expected future revenue from current customers. The choice rule can
be easily modified to allow for switching costs to affect choice. Additionally, expectations
might later be broadened to explicitly capture the impact of controllable marketing mix
variables which influence expectations, such as advertising.

Repurchase Intentions, Expectations and Reputations

Expectations have long been viewed as playing an important role in customer satis-
faction, as well as other aspects of consumer behavior. For example, Johnson and Plott
(1989) and Oliver and Winer (1987) show that individuals adapt their expectations to
account for past prices. In terms of product quality, Meyer ( 1981 ) and Cadotte et al.
( 1987) show that expectations change with product experience. In order to account for
the effect of past experience on expectations as well as future repurchase behavior, Bayesian
decision theory also provides a unified approach to modeling expectations and under-
standing their influence on choice.
The Bayesian approach describes how a firm can develop a reputation for providing
satisfaction. Firms that consistently provide high satisfaction have customers with low
variance of expectations about the firm's quality. The variance of expectations can be
interpreted as the firm's reputation. What does this reputation do for the firm? It deter-
mines how customers' sensitivity to short-run deviations in product quality and satis-
faction. Our model makes predictions about the effect of a firm's consistently high sat-
isfaction on its ability to retain customers. Accordingly, observed utility in each time
period is expected to revise expectations upwards or downwards. If perceived quality falls
short of ( exceeds) expectations, then expectations for future quality decrease (increase).
In addition, each observation will affect the variance of expectations. As experience with
a brand grows, the distribution for the individual brand should tighten about the actual
product location. Hence, past satisfaction with a brand should lead not only to increased
expectations of future quality, but increased confidence in the brand as well. This implies
that there will be smaller adjustments in expectations as experience with a brand increases.
The Bayesian approach is useful in predicting the effect of consistently providing high
satisfaction on perceived quality, satisfaction and retention. Without the analytical
framework, a plausible prediction might be that firms with a higher elasticity of retention
with respect to quality and satisfaction should provide greater satisfaction, just as firms
with high price elasticity of demand offer lower prices. However, within an industry, the
analytical framework predicts that the elasticity of retention with respect to quality and
satisfaction will be lower for firms which consistently provide high quality. As a customer
samples a product over consecutive periods, Bayesian updating suggests the distribution
of expectations should tighten about the objective locations of the chosen products. Con-
sequently, as buyers become more familiar or gain greater expertise within the category,

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
ANTECEDENTS AND CONSEQUENCES OF CUSTOMER SATISFACTION 133

this should be reflected in stronger expectations and reduced uncertainty about product
locations. The more precise expectations become, the less sensitive consumers will be to
defects. A buyer with a history of positive experiences with a product will not adjust
expectations as rapidly as buyers that are relatively new to the brand. These experienced
customers will be more forgiving. Hence, the customer base of firms which consistently
provide high quality should be relatively insulated in the short-term. The argument for
the existence of a reputation insulating firms providing higher satisfaction is summarized
in Table 1 as the Insulation hypothesis, X51, and the alternative hypothesis as the Im-
mediate Response hypothesis, ?52-

3. Data and Methodology

Description of the Database

The data are provided by the Customer Satisfaction Project at the University of Mich-
igan. The database was constructed from a computer-aided telephone survey administered
to 22,300 customers of 57 companies in Sweden in 1989 and 1990. The survey was
designed to obtain a nationally representative sample of customers of major companies
in a wide variety of industries: airlines, banks, cars, charter travel, clothing retail, de-
partment stores, furniture stores, gas stations, insurance, mainframe computers, PCs,
railroads and supermarkets. The companies represent the largest firms such that cu-
mulative share is 70%.
Respondents were selected on the basis of having recently bought or used a company's
product. To participate, each respondent was required to pass a battery of screening
questions. The sample was designed to have a maximum sampling error of 2.5% and
consists of over 200 interviews per company (except in special cases such as mainframe
computers). The questionnaire employed 10-point scales to assess each respondent's
satisfaction, repurchase intentions, expectations, perceived quality, degree of C/D, and
ease of evaluating quality. Respondents indicated their Satisfaction, SAT, on a 10-point
scale from low = 1 to high = 10. Retention is reflected in Repurchase Intentions, REPINT,
a 10-point scale with '1' indicating low repurchase intentions and '10' indicating high
repurchase intentions. Respondents indicated their level of expectations, EXP, on a scale
from low = 1 to high = 10. Two questions were asked to obtain a measure the value of
perceived quality, QUAL. In one, respondents were asked to rate quality, given price,
from low = 1 to high = 10. In the second, they were asked to assess price, given quality,
from unfavorable = 1 to favorable = 10. Perceived quality, QUAL, was determined by
taking the square-root of the product of these two variables. Hence, this variable captures
the value provided by a product or service. The ease of evaluating quality, QE, reflects
the degree of perceived difficulty in evaluating quality, with '1' equal to the highest
difficulty and '10' the lowest difficulty. Finally, respondents were asked to what extent
their expectations had been realized. A response of '1' indicated that quality was much
worse than expected and a response of ' 10' indicated that quality was much better than
expected. The responses were subsequently transformed by centering the original 10-
point scale at 0 and multiplying by a constant to produce a new scale ranging from -10
to 10. The benefit of subtracting a constant to center the scale is to allow increasingly
negative (positive) values of this disconfirmation variable, CD, to correspond to greater
negative (positive) disconfirmation. Multiplying by a constant allows for direct compar-
isons in the size of our estimated coefficients. Neither transformation alters the information
provided by the scale or the significance of the estimates.

Approach to Testing the Hypotheses

In this subsection, we develop the specification of Equations (5.1 ) to (5.4), on which
our analytical framework is based, and discuss how hypotheses associated with the

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
134 EUGENE W. ANDERSON AND MARY W. SULLIVAN
equations can be tested. These equations are presented in Table 2. The use of this system
of equations in testing our hypotheses is summarized in Table 3.
The first equation in Table 2, Equation (5.1 ), captures the assimilation effect described
earlier and depicted by the arrow from expectations to perceived quality in Figure lb.
According to the Assimilation hypothesis, X3 1, we expect the coefficient f1 1 , representing
the impact of expectations EXP, on perceived quality QUAL to be positive. As sum-
marized in Table 3, we test this hypothesis by constraining the coefficient for each firm
to be equal to zero, f1 l = 0, determining whether this constraint cannot be rejected, and
reporting the number of firms for which the coefficient is greater than zero when uncon-
strained. If the constraint is rejected and the unconstrained coefficient is greater than
zero, then we would not be able to reject the Assimilation hypothesis.
Equation (5.2) estimates the effects of QUAL, EXP, and ease of evaluating quality,
QEVAL, on confirmation/disconfirmation CD. If disconfirmation is a function of the
gap between perceived quality, and expectations, then we expect the coefficients f21 and
022 to be positive and negative, respectively. The Evaluation hypothesis, X41, states that
as the ease of evaluating quality increases, then negative disconfirmation and positive
disconfirmation also increase. Accordingly, two new variables are created to capture the
effect of ease of evaluating quality when disconfirmation is positive, QEPD, or negative,
QEND. If disconfirmation is positive then QEPD equals QE and zero otherwise. If discon-
firmation is negative then QEND equals QE and zero otherwise. Hence, Equation (5.2)
is analogous to a switching regression in which the coefficient for the ease of evaluating
quality has been allowed to vary depending on whether disconfirmation CD is greater
or less than zero. As there is no theoretical basis for expecting otherwise, the remaining
coefficients are constrained to be equal. According to this specification, if the Evaluation
hypothesis is supported, then we should observe 023 > 0 and 024 < 0, respectively. If the
alternative Information hypothesis is supported, then we should observe 023 < 0, 024
> 0, and 023 * 024. As summarized in Table 3, we can test these hypotheses by constraining
the coefficients, 023 and 024, to be equal to zero, determining whether the constraints can
or cannot be rejected, and reporting the number of firms for which the unconstrained
coefficients are greater than or less than zero. If the constraints are rejected and the
unconstrained coefficients are such that 023 < 0, f24 > 0 then the Evaluation hypothesis
would be supported. If the constraints are rejected and the unconstrained coefficients are
such that 023 > 0, 024 < 0 then the Information hypothesis would be supported.

TABLE 2
Model Specification
System of Equiations Underlying The Analytical Framework

QUAL = a, + /1IEXP + El, (5.1)

CD = a2 + /21QUAL + 322EXP + 023QEPD + 024QEND + E2, (5.2)

SAT = a3 + fl31LQUAL + fl32LND + fl33LPD + E3, (5.3)

REPINT = a4 + /41SAT + E4, where, (5.4)

QUAL = Perceived Quality,
EXP = Expectations,
CD= Disconfirmation,
QEPD = Ease of Evaluating Quality (Positive Disconfirmation),
QEND = Ease of Evaluating Quality (Negative Disconfirmation),
SAT = Satisfaction,
LQUAL = Perceived Quality (natural log).
LND Negative Disconfirmation (natural log).
LPD Positive Disconfirmation (natural log), and
REPINT = Repurchase Intentions.

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
ANTECEDENTS AND CONSEQUENCES OF CUSTOMER SATISFACTION 135

TABLE 3
Approach to testing the Hypotheses

Hypothesis Conditions Suggesting Support

#1(: Quality-Disconfirmation Respecify (5.3) to include#34 In (EXP).
The constraint 3#34 = 0 cannot be rejected and f331 = 0 is rejected.

)'12 Expectancy-Disconfirmation Respecify (5.3) to include#34 In (EXP).

The constraint 31 = 0 cannot be rejected and #34= 0 is rejected.

Yf21 Asymmetric-Disconfirmation The constraint f332 = 0833 is rejected.

Unconstrained estimation finds 332 > 033-

.4g31 Assimilation The constraint #II = 0 is rejected.

Unconstrained estimation finds $,, > 0.

eW41 Evaluation The constraints f323 = 0 and f324 = 0 are rejected.

Unconstrained estimation finds f323> 0 and 324 < 0-

if*42 Information The constraints f323 = 0 and O324 = 0 are rejected.

Unconstrained estimation finds 323 < 0 and /O24 > 0-

0Y5,: Insulation Correlation between satisfaction and elasticities of retention with

respect to satisfaction, quality and negative and positive
disconfirmation is negative and significant.

r* 52: Immediate Response Correlation between satisfaction and elasticities of retention with
respect to satisfaction, quality and negative and positive
disconfirmation is positive and significant.

The satisfaction function as specified by Equation ( 1) of the model is estimated by

Equation (5.3) in the system of equations shown in Table 2. As indicated, satisfaction,
SAT, is specified as a function of perceived quality and disconfirmation. The natural
logarithm of perceived quality, LQUAL = ln (QUAL), is used to capture the diminishing
returns effects discussed in developing the model and illustrated in Figure 2a. To test the
Asymmetric-Disconfirmation hypothesis, X21, of asymmetric effects of negative and pos-
itive disconfirmation, the disconfirmation variable CD is decomposed into negative dis-
confirmation LND, and positive disconfirmation LPD. Accordingly, if CD is negative,
then the measurement variable for negative disconfirmation LND, is equal to -ln (-CD),
and positive disconfirmation LPD is equal to zero. If CD is positive, then positive dis-
confirmation LPD is set equal to ln (CD), and negative disconfirmation LND is set
equal to zero. Hence, given our measures, we expect each coefficient in Equation (5.3)
to be positive, 031 > 0, 032l> 0, and 033 > 0.
As indicated in Table 3, the Asymmetric-Disconfirmation hypothesis, X21 , is tested by
constraining the coefficients for negative and positive disconfirmation to be equal, /32
= 033, determining whether the constraint can or cannot be rejected, and reporting whether
032 > 033. Hypothesis 21 is supported if this constraint is rejected and the absolute size
of the coefficient for negative disconfirmation is greater than the coefficient of positive
disconfirmation, 032 > 033.
Equation (5.3) is also central to testing whether the Quality-Disconfirmation hypothesis,
X1,, or the Expectancy-Disconfirmation hypothesis, XI 2, is supported by the data. As
summarized in Table 3, the procedure for discriminating between these hypotheses in-
volves first specifying a new unconstrained specification of (5.3) to include the direct
effect of expectations on satisfaction:
SAT = a3 + 03,LQUAL + 032LND + 033LPD + 034 ln (EXP) + IE.
The hypotheses are tested by alternatively constraining the perceived quality and ex-
pectations coefficients to be equal to zero, 031 = 0 and f34= 0. If the "no effect" constraint

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
136 EUGENE W. ANDERSON AND MARY W. SULLIVAN
for expectations, 031 = 0, cannot be rejected, but the same constraint for perceived quality,
034= 0, is rejected, then the Quality-Disconfirmation hypothesis is supported. Conversely,
the constraint for perceived quality, f34 = 0, cannot be rejected, but if the constraint for
expectations, /31 = 0, is rejected, then the Expectancy-Disconfirmation hypothesis is
supported.
Finally, Equation (5.4) captures the effect of satisfaction, SAT, on repurchase intentions,
REPINT. Hence, the coefficient for satisfaction, f41 is expected to be positive. As indicated
in Table 3, the Insulation and Immediate Response hypotheses, ?t51 and 52, are tested
by tracing the effects of the antecedents of satisfaction onto repurchase intentions through
the model in the form of elasticities. If the Insulation hypothesis that satisfaction should
lead to higher repurchase intentions is supported, then we should observe a negative
correlation between the retention elasticities for each firm and the average level of sat-
isfaction for that firm's customers. If the Immediate Response is supported, then we
should observe a negative correlation between the retention elasticities for each firm and
the average level of satisfaction for that firm's customers.

Estimation Method
The system of equations is recursive and identified, so we can estimate each equation
separately and trace its effects through the system. However, the errors of these equations
are expected to be correlated since a customer reporting high or low perceived quality is
likely to report correspondingly high or low disconfirmation, satisfaction and retention.
To overcome nonindependence of the error terms, we use Zellner's ( 1963) seemingly-
unrelated regression (SUR) method.2 An additional concern is the possibility of estimation
bias from skewed distributions that can result from the use of a 10-point scale. However,
examination of frequency distributions, normal probability plots, the position of the
mean with respect to the median, skewness (average of -0.63 for satisfaction), and
unimodality of the data suggest this bias is not a severe problem in this dataset.3

4. Empirical Findings

The system of equations presented in Table 1 is estimated 114 times, once for each
firm for each year of the sample. Table 4 reports the mean values of the coefficients,
whether the coefficient has a significant effect for all firms, the mean, minimum and
maximum t-values for each coefficient, and the average R-squared for each equation. To
determine whether each coefficient has a significant effect for all firms, each coefficient
is constrained to be equal to zero and the constrained system is estimated for each of the
1 14 firms. The residuals from the constrained estimation and the residuals from the
unconstrained estimation are used to conduct a restricted-unrestricted Chow test with
114 restrictions. In addition, a likelihood-ratio test using these residuals is conducted.
As indicated in Table 4, each coefficient was found to have a significant effect across all
firms. The average, minimum, and maximum of the t-values for each coefficient is pro-
vided to indicate the stability of the model and aid the reader in evaluating the strength
of support for our analytical framework.
The estimates presented in Table 4 generally support the analytical framework. The
estimates for Equation (5.1 ) indicate that expectations are positively related to quality

2 As a precaution, the system was also estimated via three-stage least squares. Numeric differences in the
estimates from SUR were found to be slight. The outcomes of the hypothesis tests do not change if 3SLS is
used instead of SUR.
3 The implications of the shape of the distribution of satisfaction for the firm is an important topic not fully
addressed by traditional econometric methods which rely on mean tendencies. Future research may wish to
employ techniques such as flexible regression (Rust 1988) to account for the effect of the distribution of satisfaction
on retention.

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
ANTECEDENTS AND CONSEQUENCES OF CUSTOMER SATISFACTION 137

TABLE 4
Firm-by-Firm Estimation Results
Average R2 and Average Coefficient Values
Total Sample Size: 22,300

QUAL = 4.88* + 0.26* EXP, (5.1)

R 2 =0.08 3.7 3a
(-I.1,+7.5)

A2=0.65 3.85 2.37 5.22 -5.69

CD =-2.88* + 0.43* QUAL + 0.25* EXP + 0.43* QEPD - 0.51* QEND, (5.2)
(-0.7,8.0) (-1.7,7.7) (-0.2,9.3) (-9.7,-1.5)

A2=0.41 6.07 3.11 4.07

SAT = 3.12* + 2.14* LQUAL + 0.80* LND + 0.56* LPD, (5.3)
(1.6,12.0) (-0.3,6.5) (0.8,8.5)
REPINT= 3.16* + 0.58* SAT. (5.4)
fi2=0.19 6.78
(1.1,13.1)

a Top row: Mean t-value; bottom row: (Minimum, Maximum) t-values.

* Indicates rejection of the hypothesis that the coefficient equals zero for all firms at the 0.0001 level for both
Chow and likelihood-ratio tests.

as predicted by the model. The estimates for Equation (5.2) indicate that disconfirmation
is positively influenced by perceived quality. Interestingly, we find expectations to be
positively related to CD, although it was expected to be negatively related to CD. In fact,
both QUAL and EXP are positively correlated with CD, while the difference between
quality and expectations, (QUAL - EXP), is also positively correlated with CD. Hence,
the observed positive coefficient is likely to be due to the effects of aggregating responses
to the firm level. In aggregating, deviations in performance from expectations experienced
by different individuals may cancel each other out. Hence, at an aggregate level, expec-
tations and quality may appear closely related. In other words, aggregate customer ex-
pectations of quality may appear to be "rational expectations." This might be expected
in the types of mature product and service categories involved in this study. Further
investigation of this finding is a good topic for future research.
Satisfaction, SAT, is positively influenced by perceived quality as indicated by the
positive coefficient, 031 > 0. The positive coefficient for negative disconfirmation, f32
> 0, indicates that larger negative values of disconfirmation lower satisfaction. Positive
disconfirmation has a positive impact on satisfaction, f33 > 0, as expected. It is worth
noting that the average estimated coefficient for perceived quality on satisfaction is greater
than those for either positive or negative disconfirmation. This holds for 105 of the 1 14
firms, suggesting that perceived product quality is more important for satisfaction and
retention than disconfirmation.
Finally, the average estimates in Table 4 show that repurchase intentions, REPINT,
is positively influenced by the level of satisfaction, as expected. Hence, buyers are more
likely to be retained as satisfaction increases. The large or small intercept might imply
other factors such as switching costs or switching benefits affect repurchase intentions.
Tracing effects through the system, the estimated coefficients imply that increasing quality
or decreasing negative disconfirmation can have important benefits for the firm.
The outcome of the hypothesis testing procedures described in Table 3 is summarized
in Table 5. The first column indicates the hypothesis to be tested. The second column
identifies the constraint (s) imposed on the system of equations. The third column shows
the outcome necessary if the hypothesis is to be supported. Each constraint is imposed
for all firms, and the estimates from the constrained model are compared to those from
the unconstrained model. The fourth column presents the calculated F-statistic for a
restricted-unrestricted Chow test with 1 14 restrictions. In each case, a likelihood-ratio
test was also conducted with identical outcomes as those reported for the Chow test. The

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
 138 EUGENE W. ANDERSON AND MARY W. SULLIVAN
TABLE 5
Summary of Hypothesis Tests
Total Sample Size: 22,300

Constraints Observed
Imposed Result if fij Observed Coefficient
Hypothesis in Question on System Is Supported F Relations Action
Yf I,: Quality-Disconfirmation 03L = 0 Reject 35.18* /31 > 0 Fail to Reject
034 = 0 Fail to Reject 0.97
f12: Expectancy-Disconfirmation 031 = 0 Fail to Reject 35.18* 334 > 0 Reject
034 = O Reject 0.97
#f21: Asymmetnic-Disconfirmation /32 = 033 Reject 2.19* 332 > /33 Fail to Reject
-#f31 Assimilation Oil = 0 Reject 15.44* 3Bl > 0 Fail to Reject
#yi41. Evaluation 023 = 0 Reject 29.88* 123 > 0 Fail to Reject
024 = 0 Reject 34.16* /24 < 0
**y42 Information /23 = 0 Reject 29.88* /23 > 0 Reject
/24 = 0 Reject 34.16* /24 < 0
* Indicates rejection of the constraint for all firms at the 0.0001 level using both Chow and likelihood-ratio tests.

fifth column provides the observed relations for the average estimates associated with
each hypothesis. Finally, the sixth column indicates whether or not the hypothesis in
question is supported. As indicated, the data support the Quality-Disconfirmation, Asym-
metric Disconjirmation, Assimilation, and Evaluation hypotheses. Each of these findings
is discussed in detail below.

Quality-Disconfirmation and Expectancy-Disconfirmation Hypotheses

The Quality-Disconfirmation and Expectancy-Disconfirmation hypotheses are tested
by respecifying Equation (5.3) to include 334 ln (EXP) as summarized in Table 3. Next,
the constraints /31 = 0 and 334 = 0 are imposed. As indicated in Table 5, we find the
first constraint is rejected (F = 35.18) and the second constraint cannot be rejected (F
= 0.97). Hence, the data suggest support for the Quality-Disconfirmation hypothesis and
reject the Expectancy-Disconfirmation hypothesis. Satisfaction is most appropriately
modeled as a function of perceived quality and disconfirmation at the firm level.

Asymmetric-Disconfirmation Hypothesis
To determine whether or not the Asymmetric-Disconfirmation hypothesis is supported,
the constraint 32 = /33 is imposed on the system of equations presented in Tables 2 and
4. As indicated in Table 5, the calculated F for this constraint is 2.19 which is significant
at the 0.0001 level given 114 restrictions. Moreover, as shown in Table 4, the average
coefficient for negative disconfirmation, /32, is greater than the average coefficient for
positive disconfirmation, 033. Hence, satisfaction is significantly more sensitive to negative
disconfirmation than positive disconfirmation. On a case-by-case basis, the effect of neg-
ative disconfirmation on satisfaction was larger in magnitude than that of positive dis-
confirmation for nearly 80% of the cases (83 out of 1 14). Hence, the data suggest support
for the hypothesis that losses should loom larger than gains in determining satisfaction.
This finding implies that it is better for firms to err on the high side in determining an
"optimal" level of satisfaction. A product which falls short of expectations is less likely
to retain present customers, and it is typically more costly to obtain new customers than
to retain present ones.

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
ANTECEDENTS AND CONSEQUENCES OF CUSTOMER SATISFACTION 139

Assimilation Hypothesis

As indicated in Tables 4 and 5, the constraint, ,I3 1 = 0, of no effect for the impact of
expectations, on perceived quality is rejected (F = 15.44). However, it is worth noting
that the R 2 of 0.08 reported for Equation (5.1 ) indicates that the assimilation effect does
not explain a very large proportion of the variance in perceived quality. Further inves-
tigation of the assimilation effect and the conditions under which this effect is strong or
weak is a good topic for future research.

Evaluation and Information Hypotheses

As indicated in Table 5, the constraints, 023 = 0 and /24 = 0, of no effect for the ease
of evaluating quality on disconfirmation are rejected. In addition, the unconstrained
estimation finds /23 > 0 and f24 < 0, as indicated in Table 4. Hence, the data appear to
support the Evaluation hypothesis and not the Information hypothesis. The effect of ease
of evaluating quality on disconfirmation suggests that when buyers find it easier to dis-
tinguish between high and low quality, they are more likely to experience disconfirmation.
As a result, it may be more important to control quality and manage dissatisfaction in
situations where buyers are familiar with the product or quality is difficult to standardize,
such as services. On the other hand, disconfirmation may be less of a concern in categories
where quality is difficult to evaluate. This finding implies that firms should consider
raising expectations on attributes which are difficult for customers to disconfirm. This
would allow the firm to enjoy the attractive power of high expected quality, while avoiding
lost customers and negative word-of-mouth which can result from disconfirmed expec-
tations.

Insulation and Immediate Response Hypotheses

To determine whether firms develop brand loyalty for consistently providing satisfac-
tion, we calculate the retention elasticities with respect to satisfaction, quality assuming
no disconfirmation, and disconfirmation. Table 6 presents these elasticities for a few
select industries. The derivation of the elasticities is presented in Appendix A. The table
shows that the high satisfaction firms in each industry tend to have low retention elas-
ticities. For example, Firm A in airline travel, Firm B in consumer banking, and Firm
B in supermarkets have relatively high satisfaction and relatively low elasticities in their
respective industries.
To quantify the relation between customer retention and each elasticity, we calculate
correlations between mean satisfaction and each retention elasticity. To adjust for industry

TABLE 6
Average Satisfaction and Retention Elasticities for Select Firms in 1989

Quality Negative C/D Positive C/D

Firm Average Elasticity Elasticity Elasticity
Industry Name Name Satisfaction 'JR. Q 'JR- ND 'JR- PD
Airlines Firm
Firm AB 7.57
7.50 0.04
0.17 0.02
0.05 0.01
0.01
Banks (Consumer) Firm B 8.45 0.13 0.08 0.02
Firm
Firm A
D 8.37 0.16
8.02 0.15 0.14
0.11 0.04
0.07
Firm C 7.16 0.19 0.19 0.07
Supermarkets Firm B 7.76 0.11 0.03 0.01
Firm A
Firm C 7.37
6.75 0.33
0.20 0.07
0.04 0.06
0.06
Averages 7.45 0.17 0.06 0.04
This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
140 EUGENE W. ANDERSON AND MARY W. SULLIVAN
differences in the level of satisfaction and size of elasticities, differences from industry
means were used to calculate the correlations.
As shown in Table 7, the correlation between firm satisfaction and the elasticity of
retention with respect to satisfaction is -0.19. Hence, high satisfaction firms are less
sensitive to changes in satisfaction relative to other firms in the same industry. This is a
conservative test, since mean satisfaction is part of the numerator, the elasticity is biased
upward. Tracing effects further back into the system, the correlation between firm sat-
isfaction and the elasticity of retention with respect to quality is found to be -0.29. As
before, this indicates that high satisfaction firms in an industry are less sensitive to changes
in satisfaction and quality. Finally, the correlations between the firm satisfaction and the
elasticities of negative and positive disconfirmation are found to be -0.20 and -0.30,
respectively. These correlations indicate that high satisfaction firms are less vulnerable
to the impact of negative disconfirmation on retaining current customers. The positive
correlations in column 3 are expected since the quality and disconfirmation elasticities
are partially a function of the satisfaction elasticity (e.g., 77R. Q = 77R S77S Q). The column
3 and 4 correlations between the quality and disconfirmation elasticities indicate that
while the quality elasticity and the positive disconfirmation elasticity move together,
neither is significantly correlated with the negative disconfirmation elasticity. Further
investigation of the conditions under which firm sensitivity to quality or disconfirmation
increases or decreases is a good topic for future research.
Overall, the findings presented in Table 7 indicate current satisfaction, current perceived
quality, and disconfirmation have less of an impact for high satisfaction firms. This strong
relation between satisfaction level and the elasticities indicates support for the hypothesis
that high satisfaction firms are insulated. If customers are scarce and relatively costly to
attract, achieving satisfaction would appear to have important financial benefits for
the firm.

However, there are limitations to the preceding analysis. Adjusting for industry char-
acteristics by subtracting out the industry means is practical but imposes a large number
of restrictions given the small number of observations per industry. Future research may
wish to directly account for category and competitive characteristics of each industry-
such as switching costs and concentration-before examining firm differences.
This finding implies that a reputation for satisfying customers makes the firm less
vulnerable to temporary quality declines. Investing in customer satisfaction is like taking
out an insurance policy. If some hardship temporarily befalls the firm, customers will be
more likely to remain loyal.

6. Summary and Concluding Comments

This paper has investigated the antecedents and behavioral consequences of satisfaction
both analytically and empirically. A model was developed to predict and explain relations
between the antecedents and consequences of customer satisfaction, as well as systematic

TABLE 7
Correlation Matrix: Average Satisfaction and Elasticities

Avg. SAT ?JREPINT * SAT nREPINT * QUAL nREPINT * ND

7JREPINT SAT _0. 19a

nREPINT QUAL -0.27a 0.66a
nREPINT ND -0.20a 0.36a 0.06
?JREPINT PD -0.30a 0.65a 0.30a 0.14
a p-value less than 0.05.

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
ANTECEDENTS AND CONSEQUENCES OF CUSTOMER SATISFACTION 141

differences in these relations across firms. Satisfaction was linked explicitly to retention
in a utility-oriented framework. By integrating the rich substantive findings of consumer
research with the utility-oriented perspective of marketing science, the micromodel pro-
vides an axiomatic basis for future research on customer satisfaction and retention.
The model was estimated and tested against alternative hypotheses using a unique
database: 22,300 customers in Sweden in 1989-1990. In the process, several well-known
experimental findings of satisfaction research were tested in a field setting of national
scope. For example, satisfaction was found to increase with both perceived quality and
disconfirmation. This is an important finding given the strong focus of satisfaction research
on expectations and disconfirmation as the antecedents of satisfaction. In addition, sat-
isfaction was found to have a positive impact on repurchase intentions. Moreover, both
positive and negative disconfirmation were found to increase with the ease of evaluating
quality. Hence, it may be more important to manage customer satisfaction when cus-
tomers are very familiar with a product or the product is not complex. When quality is
ambiguous or difficult to evaluate, then expectations will play a greater role in determining
satisfaction. In addition, quality which falls short of expectations was found to have a
greater impact on satisfaction and retention than quality which exceeds expectations.
Finally, in terms of systematic variation across firms, the elasticity of repurchase intentions
with respect to satisfaction is found to be lower for firms that provide high satisfaction.
This implies a long-run reputation effect insulating firms consistently providing high
satisfaction.
We believe this paper has important implications for both academics and practitioners.
First, firms which consistently provide high-quality products should have a more satisfied
customer base and one that is more likely to be retained. If it is more costly to add new
customers than to maintain relationships with current customers, then such firms should
enjoy greater profitability in the long run. Second, an important component of managing
satisfaction is the ability to control the impact of negative disconfirmation through com-
plaint handling and effective customer service. Third, since negative disconfirmation
hurts the firm more than positive disconfirmation helps it, firms should take measures
to maintain reliability when improving quality. It would be helpful to understand the
conditions under which firms should allocate resources to maintaining quality before
improving quality. This is a good topic for future research. Fourth, our findings indicate
that the firm's future profitability depends on satisfying customers in the present. If
consistently providing high satisfaction leads to higher repurchase intentions, then the
expected number of times a buyer will repurchase should rise accordingly. Moreover, if
acquisition costs for new customers are high, then such firms should be more efficient
in generating revenue. In light of this, it is easier to understand why many firms' com-
pensation plans now include satisfaction measures, as well as quality measures based on
accounting numbers. However, future research needs to examine this issue more closely.
Future research could address broad issues such as how marketing mix tactics on
satisfaction and retention, the relative importance of switching costs and satisfaction in
creating retention, the relationship between satisfaction and long-term profitability, and
the allocation of resources to achieve optimal levels of customer satisfaction and retention.
Immediate directions for future research include investigating: ( 1 ) the relationship be-
tween perceived quality, satisfaction, retention and business performance; (2) systematic
differences in satisfaction and retention across product categories; and (3) resource al-
location and compensation plans for achieving optimal levels of satisfaction.4
- Acknowledgements. This research has benefited from the comments of participants in the Customer Sat-
isfaction Workshop at the University of Michigan. In addition, the authors are grateful for the helpful suggestions

4 This paper was received October 26, 1990, and has been with the authors 8 months for 3 revisions. Processed
by Brian Ratchford.

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
142 EUGENE W. ANDERSON AND MARY W. SULLIVAN
of Claes Fornell, Michael Johnson, Mike Ryan, Steven Shugan, and Youjae Yi, as well as the data provided
through the funding of the Swedish Post Office.

Appendix A

To calculate the elasticities of retention with respect to satisfaction (17R. SAT), quality without disconfirmation
(71R-QUAL), and negative and positive disconfirmation (1R ND) and fR PD), the relevant coefficients from the
model estimation are used in place of the partial derivatives, and firm averages are used for REPINT, SAT,
QUAL, ND, and PD.

OR S
77R-S= --- R
O9S R

ORQ ORQ [S 1
1R-Q -- aQ R aS R aQQ

OR ND OR ND [ S 1
N ND R OS R dND ND,
t1R -ND

OR PD OR PD [ S 1
7R PD = OPD R OS R d9PD PD]

References

Alba, Joseph W. and J. Wesley Hutchinson (1987), "Dimensions of Consumer Expertise," Journal of Consumer
Research, 13 (March), 411-453.
Amemiya, Takeshi (1985), Advanced Econometrics, Cambridge, MA: Harvard University Press.
Anderson, R. E. (1973), "Consumer Dissatisfaction: The Effects of Disconfirmed Expectancy on Perceived
Product Performance," Journal of Marketing Research, 10, 38-44.
Andreasen, Alan R. (1985), "Consumer Responses to Dissatisfaction in Loose Monopolies," Journal of Consumer
Research, 12 (September), 135-141.
Bearden, William 0. and Jesse E. Teel (1983), "Selected Determinants of Consumer Satisfaction and Complaint
Reports," Journal of Marketing Research, 20 (February), 21-28.
Berger, James 0. (1985), Statistical Decision Theory and Bayesian Analysis (2nd Ed.), New York, NY: Springer-
Verlag.

Cadotte, Ernest R., Robert B. Woodruff and Roger L. Jenkins (1987), "Expectations and Norms in Models of
Consumer Satisfaction," Journal of Marketing Research, 24 (August), 305-314.
Cardozo, Richard N. (1965), "An Experimental Study of Customer Effort, Expectation, and Satisfaction,"
Journal of Marketing Research, 2 (August), 244-249.
Churchill, Gilbert A. and Carol Suprenant (1982), "An Investigation into the Determinants of Customer Sat-
isfaction," Journal of Marketing Research, 19 (November), 491-504.
Deighton, John ( 1984), "The Interaction of Advertising and Evidence," Journal of Consumer Research, 11
(December), 763-770.
Dhrymes, Phoebus J. (1971), Introductory Econometrics, New York, NY: Springer-Verlag.
Einhorn, Hillel J. and Robin M. Hogarth ( 198 1 ), "Behavioral Decision Theory: Processes of Judgement and
Choice," Annual Review of Psychology, 32, 53-88.
Fornell, Claes and William Robinson (1983), "Industrial Organization and Consumer Satisfaction/Dissatis-
faction," Journal of Consumer Research, 9 (March), 403-412.
and Birger Wernerfelt (1987), "Defensive Marketing Strategy by Customer Complaint Management:
A Theoretical Analysis," Journal of Marketing Research, 24, 337-346.
and (1988), "Model for Customer Complaint Management," Marketing Science, 7 (Summer),
271-286.
Helsen, Harry (1964), Adaptation-Level Theory, NY: Harper & Row.
Hirschman, Albert 0. (1970), Evii, Voice, and Lovalty - Responses to Declines in Firms, Organizations and
States, Cambridge, MA: Harvard University Press.
Hoch, Stephen J. and Young-Won Ha (1986), "Consumer Learning: Advertising and the Ambiguity of Product
Experience," Journal of Consumer Research, 13 (September), 221-233.
Hogarth, Robin M. (1987), Judgement and Choice, (2nd Ed.), New York NY: John Wiley & Sons.
Intriligator, Michael D. (1978), Econometric Models, Techniques, & Applications, Englewood Cliffs, N.J.:
Prentice-Hall, Inc.
Johnson, Michael D. and Charles R. Plott (1989), "The Effect of Two Trading Institutions on Price Expectations
and the Stability of Supply-Response Lag Markets," Journal of Economic Psychology, 10, 189-216.

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms
ANTECEDENTS AND CONSEQUENCES OF CUSTOMER SATISFACTION 143

Kahneman, Daniel and Amos Tversky (1972), "Subjective Probability: A Judgement of Representativeness,"
Cognitive Psychology, 47, 263-291.
and (1979), "Prospect Theory: An Analysis of Decisions under Risk," Econometrica, 47, 263-
291.
Klayman, Joshua and Young-Won Ha (1986), "Confirmation, Disconfirmation and Information in Hypothesis-
Testing," Psychological Review, 94, 211-228.
Kotler, Philip (1991), Marketing Management-Analysis, Planning, Implementation and Control, 7th Ed.,
Englewood Cliffs, NJ: Prentice-Hall, Inc.
Maddala, G. S. (1977), Econometrics, New York, NY: McGraw-Hill Book Company.
Massaro, Dominic W. and Daniel Friedman (1990), "Models of Integration Given Multiple Sources of Infor-
mation," Psychological Review, 97, 225-252.
Meyer, Robert J. (1981), "A Model of Multiattribute Judgements under Attribute Uncertainty and Informational
Constraint," Journal of Marketing Research, 18 (November), 428-441.
Nelsen, Phillip (1970), "Information and Consumer Behavior," Journal of Political Economy, 82, 311-329.
Oliver, Richard L. (1977), "Effects of Expectations and Disconfirmation on Postexposure Product Evaluations,"
Jolurnal of Applied Psychology, 62 (April), 246-250.
( 1980), "A Cognitive Model of the Antecedents and Consequences of Satisfaction Decisions," Journal
of Marketing Research, 17 (November), 460-469.
and William 0. Beardon (1985), "Disconfirmation Processes and Consumer Evaluations in Product
Usage," Jolurnal of Business Research, 13 (June), 235-246.
and Wayne DeSarbo (1988), "Response Determinants in Satisfaction Judgments," Journal of Consumer
Research, 14 (March), 495-507.
and John E. Swan (1989), "Consumer Perceptions of Interpersonal Equity and Satisfaction in Trans-
actions: A Field Survey Approach," Journal of Marketing, 53 (April), 21-35.
and Russell S. Winer (1987), "A Framework for the Formation and Structure of Consumer Expectations:
Review and Propositions," Journal of Economic Psychology, 8, 469-499.
Olshavsky, Richard W. and John A. Miller (1972), "Consumer Expectations, Product Performance and Perceived
Product Quality," Journal of Marketing Research, 9 (February), 19-21.
Olson, Jerry C. and Philip Dover ( 1979),"Disconfirmation of Consumer Expectation Through Product Trial,"
Journal oJ Applied Psychology, 64 (April), 179-189.
Rust, Roland T. (1988), "Flexible Regression," Journal of Marketing Research, 25 (February), 10-24.
Sherif, M. and C. I. Hovland (1961), Social Judgement: Assimilation and Contrast Effects in Communication
and Attitude Change, New Haven, CT: Yale University Press.
Stigler, George (1961), "The Economics of Information," Journal of Political Economy, 69, 213-225.
Thaler, Richard (1985), "Mental Accounting and Consumer Choice," Marketing Science, 4 (Summer), 199-
214.
Theil, Henri ( 1 97 1 ), Principle.s of Econometrics, New York, NY: John Wiley & Sons.
Tse, David K. and Peter C. Wilton (1988), "Models of Consumer Satisfaction Formation: An Extension,"
Jouirnal ofMarketing Research, 25 (May), 204-212.
Westbrook, Robert A. (1981), "Sources of Consumer Satisfaction with Retail Outlets," Journal of Retailing,
57 (Fall), 68-85.
Wilton, Peter C. and David K. Tse (1983), "A Models of Consumer Response to Communication and Product
Experiences," in Larry Percy and Arch G. Woodside (Eds.), Advertising And Consumer Psychology,
Lexington, MA: Lexington, 315-332.
Woodruff, Robert B., Ernest R. Cadotte and Roger L. Jenkins (1983), "Modeling Consumer Satisfaction Processes
Using Experience-based Norms," Journal of Marketing Research, 20 (August), 296-304.
Yi, Youjae (1991), "A Critical Review of Consumer Satisfaction," in Valarie A. Zeithmal (Ed.), Review of
Marketing 1989, Chicago, IL: American Marketing Association.

This content downloaded from 170.239.168.67 on Sun, 27 Apr 2025 01:25:04 UTC
All use subject to https://about.jstor.org/terms