An Empirical Investigation of Customer Satisfaction after Service Failure and Recovery

Abstract

Relatively little research has addressed the nature and determinants of customer satisfaction following service failure and recovery. Two studies using scenario-based experiments reveal the impact of failure expectations, recovery expectations, recovery performance, and justice on customers’ postrecovery satisfaction. Customer satisfaction was found to be lower after service failure and recovery (even given high-recovery performance) than in the case of error-free service. The research shows that, in general, companies fare better in the eyes of consumers by avoiding service failure than by responding to failure with superior recovery.

Full text

http://jsr.sagepub.com
Journal of Service Research
DOI: 10.1177/109467050032002
2000; 3; 121 Journal of Service Research
Michael A. McCollough, Leonard L. Berry and Manjit S. Yadav
An Empirical Investigation of Customer Satisfaction after Service Failure and Recovery
http://jsr.sagepub.com/cgi/content/abstract/3/2/121
The online version of this article can be found at:
Published by:
http://www.sagepublications.com
On behalf of:
Center for Excellence in Service, University of Maryland
can be found at:Journal of Service Research Additional services and information for
http://jsr.sagepub.com/cgi/alerts Email Alerts:
http://jsr.sagepub.com/subscriptions Subscriptions:
http://www.sagepub.com/journalsReprints.navReprints:
http://www.sagepub.com/journalsPermissions.navPermissions:
http://jsr.sagepub.com/cgi/content/refs/3/2/121 Citations
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

JOURNAL OF SERVICE RESEARCH / November 2000McCollough et al. / CUSTOMER SATISFACTION
An Empirical Investigation
of Customer Satisfaction After
Service Failure and Recovery
Michael A. McCollough
University of Idaho
Leonard L. Berry
Manjit S. Yadav
Texas A&M University
Relatively little research has addressed the nature and de-
terminants of customer satisfaction following service fail-
ure and recovery. Two studies using scenario-based
experiments reveal the impact of failure expectations, re-
covery expectations, recovery performance, and justice on
customers’ postrecovery satisfaction. Customer satisfac-
tion was found to be lower after service failure and recov-
ery (even given high-recovery performance) than in the
case of error-free service. The research shows that, in gen-
eral, companies fare better in the eyes of consumers by
avoiding service failure than by responding to failure with
superior recovery.
Service failure and recovery is a critical issue for both
service managers and researchers. However, until recently,
research on the nature and determinants of customer satis-
faction following service recovery (i.e., the actions a ser-
vice provider takes in response to service failure
[Gronroos 1988]) has been limited. Therefore, recovery
has been identified as a neglected area requiring addi-
tional research (Andreassen 1999; Fisk, Brown, and
Bitner 1993; Singh and Widing 1991; Tax, Brown, and
Chandrashekaran 1998). As a result of the limited atten-
tion given to recovery, little is known about how customers
evaluate recovery efforts, what constitutes successful re-
covery, and the potential (and limits) of recovery to con-
vert customer dissatisfaction into satisfaction.
Understanding recovery is important for managers.
Service failure is one “pushing determinate” that drives
customer switching behavior (Roos 1999), and successful
recovery can mean the difference between customer reten-
tion and defection. In turn, customer retention is critical to
profitability (Stauss and Friege 1999). Reicheld and
Sasser (1990) maintain that, in certain circumstances, a
service company can boost profits almost 100% by in-
creasing customer retention just 5%. For service provid-
ers, recovery has special significance. Fisk, Brown, and
Bitner (1993) argue that due to the unique nature of ser-
vices (specifically, coproduction and the inseparability of
production and consumption) it is impossible to ensure
100% error-free service.
This research addresses two broad research questions.
First, the antecedents of postrecovery satisfaction are ex-
amined within the context of the disconfirmation and jus-
tice literature. Second, and of equal importance, this
research explores the recovery paradox (McCollough and
The authors wish to thank the Center of Retailing Studies and the Graduate School of Business at Texas A&M University and the Col
-
lege of Business and Economics at the University of Idaho for providing partial support for this research.
Journal of Service Research, Volume 3, No. 2, November 2000 121-137
© 2000 Sage Publications, Inc.
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

Bharadwaj 1992) or the question of whether customers
who experience a failure followed by superior recovery
might rate their satisfaction as high as or even higher than
they would have had no failure occurred. Marketing re-
searchers have long wondered if truly superior recovery
efforts actually can create greater satisfaction than if noth-
ing had gone wrong (Etzel and Silverman 1981). Some re-
searchers report results consistent with a recovery paradox
effect (Bitner, Booms, and Tetreault 1990; Goodwin and
Ross 1992; Hansen and Danaher 1999; Hart 1993; Kelly,
Hoffman, and Davis 1993; Morris 1988; Smith and Bolton
1998; TARP [Technical Assistance Research Programs,
Inc.] 1979, 1986; Triplett 1994). For instance, Hart,
Heskett, and Sasser (1990) state, “A good recovery can
turn angry, frustrated customers into loyal ones. It can, in
fact, create more goodwill than if things had gone
smoothly in the first place” (p. 148). However, other re-
searchers have reported results inconsistent with a recov-
ery paradox effect (Berry, Zeithaml, and Parasuraman
1990; Bolton and Drew 1991; Fornell 1992; Halstead and
Page 1992; Smart and Martin 1993; Zeithaml, Berry, and
Parasuraman 1996). In general, no theoretical explanation
of why a recovery paradox effect is possible has been of-
fered. In addition, very little research has directly com-
pared postrecovery satisfaction with consumers who
experienced error-free service while controlling for con-
founds that might affect satisfaction evaluations. Such a
comparison is necessary if the conditions of the recovery
paradox as laid out by McCollough and Bharadwaj (1992)
for a recovery paradox are to be met. Therefore, our inves-
tigation of postrecovery satisfaction and the recovery para-
dox addresses a subject with contradictory findings.
CONCEPTUAL FRAMEWORK
AND HYPOTHESES
Although, in general, the literature on postrecovery sat-
isfaction is limited, some significant headway has been
made recently (Smith and Bolton 1998; Tax, Brown, and
Chandrashekaran 1998). This research adds to the grow-
ing literature on recovery by evaluating recovery from the
perspective of disconfirmation (with emphasis on the role
of failure and recovery expectations) and justice.
Disconfirmation
The most widely used model within the consumer
satisfaction/dissatisfaction (CS/D) literature is the discon-
firmation paradigm (Bearden and Teel 1983; Oliver 1980,
1981, 1989, 1993; Oliver and Bearden 1985; Oliver and
Burke 1999; Swan and Trawick 1981). Disconfirmation
also has been advanced as a model for understanding cus
-
tomers’ reactions to recovery (Oliver 1981; Singh and
Widing 1991). The disconfirmation paradigm holds that
customers compare perceived product performance to ex
-
pectations. Performance that exceeds expectations is posi-
tively disconfirmed, performance that meets expectations
is confirmed, and performance that falls short of expecta-
tions is negatively disconfirmed. In general, the more neg-
ative the disconfirmation, the greater the dissatisfaction,
whereas the more positive the disconfirmation, the greater
the satisfaction.
In the left-hand portion of Figure 1, a disconfirmation
model of recovery (which is empirically evaluated in
Study 1) is presented. In this section, the disconfirmation
portion of the model is explained, and concurrent with this
discussion, a disconfirmation-based explanation of the re-
covery paradox will be advanced.
The disconfirmation portion of Figure 1 holds that sat-
isfaction (modeled as a postrecovery transaction-specific
judgment and not as a general attitude regarding the pro-
vider’s overall service quality) is a function of initial
disconfirmation and recovery disconfirmation. Initial
disconfirmation is defined as the discrepancy between
failure expectations (expectations that the service might
fail) and service performance (initial service performance
perceptions, broadly conceptualized in this context as ei-
ther successful service performance or failure). Recovery
disconfirmation is defined as the discrepancy between re-
covery expectations (expectations by the consumer re-
garding what the service provider will do given failure)
and recovery performance (perceptions regarding steps
taken by the service provider in response to failure).
Our model includes expectations given their impor-
tance in the disconfirmation, recovery, and complaining
literature. In terms of understanding postrecovery satisfac-
tion, recovery expectations have been held to be the stan-
dard against which recovery performance is judged (Kelly
and Davis 1994; Oliver 1981; Singh and Widing 1991).
From a motivational perspective, recovery expectations
are also important. For instance, researchers investigating
complaining behavior have found that many customers do
not complain to the provider about a dissatisfying experi-
ence. However, those who do seek redress often are moti-
vated by recovery expectations (referred to in the
complaining literature as probability or likelihood of suc-
cess) (see Blodgett, Granbois, and Walters 1993;
Hirschman 1970; Richins 1983, 1987; Singh 1990).
An evaluation of the recovery paradox from a
disconfirmation perspective would seem to indicate, all
other things being equal, that because customers do not de-
liberately seek dissatisfaction, any transaction resulting in
a problem (negative disconfirmation), even if successfully
addressed, would lead to satisfaction lower than that re
-
sulting from error-free service. It appears reasonable that,
122 JOURNAL OF SERVICE RESEARCH / November 2000
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

if a customer must first experience a problem and seek re-
dress to achieve the level of satisfaction originally ex-
pected, this experience would be evaluated as inferior to
one that had produced the desired result without any inter-
vening problems. In essence, this perspective holds that
the initial disconfirmation drives the final CS/D evaluation
with recovery mitigating the dissatisfaction arising from
the failure. To illustrate this situation, consider the follow-
ing example:
Customers Jones and Smith both have been satisfied
customers of the same bank for many years. Cus-
tomer Jones always has found the service of the bank
to meet or exceed expectations and has never en-
countered a problem. Customer Smith also found
the bank’s service met or exceeded expectations un-
til the bank made an error and returned a check writ-
ten to a local grocery store marked “insufficient
funds.” When Customer Smith complained to the
bank, a very courteous service representative
promptly determined that the bank had made an er-
ror. The service representative apologized and out-
lined the steps the bank would take to correct the
problem, which included sending a letter to the gro-
cery store acknowledging that the bank was at fault
and offering to cover any charges the store might as-
sess. In addition, the service representative arranged
to have Customer Smith’s checking account cred-
ited for an additional $25.00 as compensation for
any embarrassment and inconvenience. “After all,”
said the service representative, “if we expect our
customers to pay for their errors, we should be pre
-
pared to pay for ours.”
In this situation, why should Customer Smith’s
postrecovery satisfaction be greater than Customer
Jones’s if Customer Smith had to contact the bank to have
the problem corrected and possibly experienced embar-
rassment and anxiety because of the returned check? Cus-
tomer Smith might be pleased with the prompt and
efficient recovery effort, but Smith would probably have
preferred error-free service.
One possible answer to the previous question regarding
Smith’s satisfaction versus Jones’s is that customers have
expectations regarding appropriate recovery efforts in the
event of failure. Extending this argument, some customers
may not only have recovery expectations, but some may
also have failure expectations. For instance, many custom-
ers recognize that consumption entails some potential for
dissatisfaction (Murray and Schlacter 1990). Therefore, to
determine what will be done in the event of a failure, they
inquire about warranties, exchange, and refund policies.
If service failure is not totally unexpected, satisfaction
judgments may be suspended until the recovery efforts can
be evaluated. If recovery efforts meet expectations, then
the evaluation of the overall transaction would be one of
confirmation. (The customer’s expectation of a possible
problem has been confirmed as has the customer’s expec-
tations of appropriate recovery efforts.) However, this
evaluation should be lower than that of a customer who
recognized the potential for a problem but did not encoun-
ter one, resulting in a state of positive disconfirmation.
Let us imagine another situation in which a customer
felt that sooner or later all banks would make an error and
that it would be very difficult to have the problem cor-
rected. The check returned in error would, therefore, re-
sult in a state of confirmation, but having the problem
quickly and agreeably resolved would be a positively
disconfirming experience. In this case, the prompt cor-
rection of the problem and the monetary credit might
so exceed Customer Smith’s original expectations of
problem resolution that high postrecovery satisfaction
would result. Therefore, within the framework of
disconfirmation, the recovery paradox would best be ex-
plained by recovery performance that exceeds expecta-
tions given failure expectations.
Our conceptualization of recovery modeled in Figure 1
can be viewed as a second disconfirmation path (for recov-
ery judgments) that can be incorporated within the stan-
dard disconfirmation model. This conceptualization is
potentially consistent with Halstead and Page (1992), who
state that the customer’s ultimate, postrecovery satisfac-
tion is driven by the initial satisfaction judgment with re-
covery mediating the final, postrecovery satisfaction
judgment. The treatment of recovery performance is also
harmonious with the recovery model proposed and evalu
-
ated by Smith and Bolton (1998), who found an increase in
McCollough et al. / CUSTOMER SATISFACTION 123
FIGURE 1
Organization Framework for Investigating
Customer Satisfaction After Service
Failure and Recovery
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

cumulative satisfaction when the customer was very satis
-
fied with an organization’s recovery effort. In many as-
pects, this perspective also is consistent with the
disconfirmation-based recovery models proposed (but
which were not empirically evaluated) by Oliver (1981)
and Singh and Widing (1991). On the basis of the previ-
ous discussion, the following hypotheses are offered (see
Figure 1):
Initial Disconfirmation Hypotheses
Hypothesis 1: The greater (lower) failure expectations,
the less (more) negative initial disconfirmation.
Hypothesis 2: The higher (lower) perceived service per-
formance, the more (less) positive initial
disconfirmation.
Hypothesis 3: The more negative (positive) initial
disconfirmation, the greater dissatisfaction
(satisfaction).
Recovery Disconfirmation Hypotheses
Hypothesis 4: The greater (lower) recovery expectations,
the less (more) positive recovery disconfirmation.
Hypothesis 5: The higher (lower) recovery performance,
the more positive (negative) recovery
disconfirmation.
Hypothesis 6: The more positive (negative) recov-
ery disconfirmation, the greater satisfaction
(dissatisfaction).
Recovery Paradox Hypothesis
Hypothesis 7: Postrecovery satisfaction will be equal to
or greater than the satisfaction when no-service fail-
ure occurs given (a) high failure expectations, (b)
low recovery expectations, and (c) high recovery
performance.
Distributive and
Interactional Justice
An emerging literature has examined recovery evalua-
tions from the perspective of justice (Goodwin and Ross
1992; Hocutt, Chakraborty, and Mowen 1997b; Tax,
Brown, and Chandrashekaran 1998). This research adds to
this work by exploring the role played by distributive and
interactional justice in recovery situations (see Figure 1).
Distributive justice (DJ) specifies that individuals eval-
uate the fairness of an exchange by comparing costs with
the gains received (Greenberg 1987, 1990b). Although DJ
can be conceptualized as customers’ evaluations of
whether they got “their money’s worth,” it also can include
nonmonetary inputs and outputs involving such intangi-
bles as emotions (anger and embarrassment), complaining
costs (time and effort), and ego benefits. From a DJ per
-
spective, classic inequity arises when an individual be
-
lieves the outcome is inadequate given inputs. The most
common recoveries are those in which the customer
responds to “negative” inequity by seeking redress and in-
volve some combination of additional rewards or lowered
costs, for example, a refund, adjustment of costs, or an
exchange.
Feelings of preference or advantageous inequity can
occur when an individual believes the outcome is greater
than deserved given the inputs (Oliver and Swan 1989a,
1989b). Given such positive inequity, it is possible for feel-
ings of guilt or indebtedness to arise. However, justice re-
search suggests that individuals have an egocentric bias
(Greenberg 1987) wherein they are more tolerant of ad-
vantageous than disadvantageous inequity (Oliver and
Swan 1989b).
Because neither inputs nor outputs need to be economic
and are viewed from the subjective viewpoints of the ex-
change principles, it is possible for differences in perspec-
tive to lead to differences in perceived DJ. In a redress
situation, a problem of inequity could result in dissatisfac-
tion if a service provider does not recognize the same costs
as the customer (the customer’s time and effort to seek re-
dress). Alternatively, if the customer’s norm is that only di-
rect monetary costs will be reimbursed, then
compensating nonmonetary costs would result in feelings
of advantageous inequity. Although feelings of guilt are
possible, the most likely outcome of reimbursing both
monetary and nonmonetary costs (given the customer’s
egocentric bias) should be postrecovery satisfaction supe-
rior to that experienced when only monetary costs are re-
imbursed. Therefore,
Hypothesis 8: The greater (lower) perceptions of distrib-
utive justice, the greater postrecovery satisfaction
(dissatisfaction).
Interactional justice (IJ) concerns the fairness of the re-
covery process itself or the interactional aspects of the en-
counter (Bies and Moag 1986; Goodwin and Ross 1992).
Interactional justice is generally considered to be a
subelement of the more global justice construct of proce-
dural justice (PJ), which concerns the fairness or the pro-
cesses used to determine the distribution of outputs
(Greenberg 1990b). In a recovery situation, PJ would con-
cern the perceived fairness of the organization’s recovery
policies, whereas IJ would be concerned with how the or-
ganization’s policies would be carried out. An example of
IJ would be an apology to the customer for a service fail-
ure. This research will focus specifically on the IJ aspects
of PJ.
Customers may judge the recovery effort on the basis of
both the recovery process (IJ) and outcome (DJ). For ex
-
ample, consider two different recovery situations in which
the same financial outcome is achieved by the customer
124 JOURNAL OF SERVICE RESEARCH / November 2000
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

(identical DJ). If the service provider is perceived as apolo
-
getic, empathetic, and responsive in one situation and is
curt and indifferent in the other, the customer’s
postrecovery satisfaction almost certainly would be higher
in the first situation even though the same tangible out-
comes are achieved in both cases.
Hypothesis 9: The greater (lower) perceptions of IJ, the
greater postrecovery satisfaction (dissatisfaction).
DJ and IJ are not independent constructs but aspects of
the same overriding construct, justice. To some degree, DJ
is a necessary, but not sufficient, condition for IJ and PJ;
without a truly fair outcome, the procedures and interac-
tion very well may be judged as flawed (McFarlin and
Sweeney 1992). Likewise, DJ does not ensure favorable
perceptions of IJ. (Distribution might be fair even if the in-
teraction is not.) Therefore, organizations seeking to pro-
vide high satisfaction must be perceived as offering both
DJ and IJ. Indeed, Greenberg (1990a) labeled situations in
which individuals attempt to be recognized as being fair
without actually behaving fairly as “hollow justice” and
noted it may backfire if individuals suspect insincerity.
Likewise, it has been argued that in some situations, PJ ef-
forts that do not lead to appropriate changes in DJ may cre-
ate a frustration effect (Folger et al. 1979). Such situations
have been labeled as “sham” participation or as
“pseudoparticipation” (Cohen 1985; Goodwin and Ross
1992) and may result in greater dissatisfaction and exit be-
havior by employees and customers. Likewise, Tax,
Brown, and Chadrashekaran (1998) found evidence of an
interactive effect between DJ and IJ.
Hypothesis 10: DJ and IJ will interact with each other as
they affect postrecovery satisfaction.
RESEARCH METHOD
Research Plan
Scenario-based experiments were conducted to investi-
gate service recovery. Bitner (1990) notes that role-play-
ing experiments (scenarios) allow expensive or difficult
manipulations to be more easily operationalized, provide
the researcher with control over otherwise unmanageable
variables, and facilitate the compression of time by sum-
marizing events that might otherwise unfold over days or
weeks. Furthermore, the use of scenarios avoids the ex-
pense and ethical consideration associated with observing
or enacting actual service failure while avoiding the re
-
sponse bias due to memory lapses and rationalization
likely to be present in surveys that rely on recall (Smith and
Bolton 1998). The key drawbacks of role-playing are a
greater likelihood of demand effects and the possible in-
ability of participants to project their behavior and to re
-
spond as they actually would in a real situation. To
minimize these problems, we recruited individuals who
were in the midst of the actual service encounter being
studied. This procedure would ensure that participants
were familiar with the service offering and minimize con-
cerns that participants in a laboratory setting might re-
spond differently from those actually involved in the
service setting.
Research Setting
Airline travel was chosen as the research setting. Air-
line travel represents a service for which failure is com-
mon (Andreeva 1998). As a result, it was anticipated that
most airline travelers would find manipulations regarding
recovery expectations, recovery performance, and justice
realistic and believable. Using an airport served by numer-
ous airlines also eliminated any possible brand bias. In ad-
dition, most passengers, when facing a flight delay or
cancellation, will have no choice but to seek redress as
canceling the trip is not an option for most passengers.
Therefore, in this research setting, no bias is created by im-
posing redress-seeking behavior on respondents who
might not ordinarily voice complaints. Finally, the admin-
istration of scenario-based experimental manipulations by
intercepting airline passengers was established by Bitner
(1990) as a reliable and valid research methodology.
Research Design
To test the research hypotheses, two separate studies
were developed. In Study 1, the disconfirmation-based hy-
potheses (Hypotheses 1-7) were investigated by the ma-
nipulation of recovery performance and recovery
expectations. Study 2 evaluated the justice hypotheses
(Hypotheses 8-10) by the manipulation of DJ and IJ. Two
separate studies were conducted because the excessive
length and complexity of a single research instrument to
investigate both disconfirmation and justice explanations
of recovery were not deemed practical.
Data Collection
Pretests. Three pretests were conducted. The first two
employed student participants and assisted in determining
the realism of the scenarios and in the development and re-
finement of the measures. After modifying the research in-
strument, concurrent verbal protocols were conducted to
verify that respondents were interpreting the scenarios as
intended.
McCollough et al. / CUSTOMER SATISFACTION 125
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

The final pretest (N = 240) was intended to mirror the fi
-
nal data collection plan as closely as possible. A small air-
port used exclusively by commuter airlines was the
location for the final pretest. Previous research has shown
that the consumers’ attributions
1
regarding a failure can in-
fluence consumer evaluations (Folkes 1984; Folkes and
Kotsos 1986). To establish firm responsibility for the fail-
ure (and avoid possible attribution confounds), the cause
of the service failure was initially attributed to a mechani-
cal problem. However, this seemed to create confound by
triggering negative beliefs regarding the airline’s safety,
possibly due to the smaller aircraft used by the airlines at
the pretest airport and a well-publicized crash of a commu-
ter aircraft just before the sampling period. Therefore, for
the final data collection, the cause of the service problem
was changed from a mechanical problem to crew unavail-
ability.
2
This change maintained the airline’s responsibil-
ity for the service failure but eliminated the safety
confound.
Main study. The data were collected at a medium-sized
regional airport in the southwestern United States. This
airport was viewed as ideal for the final data collection be-
cause it was served by the top nine domestic carriers, was
geographically accessible, and was served primarily by jet
service. After securing the cooperation of the airport man-
ager, the principal researcher met with the managers of the
carriers to review the research objectives and methodology
and to address any concerns. The airport’s one commuter
airline was not included in the study, and all passengers
surveyed were preparing to board jet aircraft. All partici-
pants were randomly assigned to treatment conditions.
Passengers were approached in the waiting areas (most
were seated) and asked to participate. In total, passengers
from 50 departures were included in the sample, and all
major carriers were represented by several departures. Ob-
servations were obtained for all hours of operation and all
days of the week during a 2-week period. To minimize dis-
ruption to the operation of the airport, and in consultations
with airport management, data for both studies were col-
lected concurrently. Therefore, the population of both
studies is the same.
For passengers unable to complete and return the sur-
vey before boarding, a postage-paid, preaddressed enve-
lope was attached to the research instrument. The primary
purpose of this option was to overcome travelers’ concerns
that they would not have time to complete the survey be-
fore boarding. This approach also allowed the inclusion of
late arrivals in the research, precluding any systematic
bias. The vast majority of participants (75.7%) returned
the survey in the terminal. The method of survey return (in
terminal or by mail) was entered into all subsequent data
analyses, and in no case did it have a material effect on the
research conclusions.
A total of 1,005 airline passengers were approached
while waiting to board flights and asked to participate in
the study. Of those, 727 (72.3%) agreed to cooperate. A to-
tal of 550 passengers returned the surveys before boarding,
whereas 177 took surveys with them. Of these 177 sample
members, 65 mailed back their surveys within 3 weeks of
the completion of sampling. Therefore, the final sample
was 615 (61.2% of those approached and 84.6% of those
who agreed to participate).
Study 1: Failure Expectations,
Recovery Expectations, and
Recovery Performance
Study 1, a 2 × 3 between-subjects design, evaluated the
recovery disconfirmation-based hypotheses (Hypotheses
1-7) by manipulating recovery expectations (high/low)
and service performance (service failure with either high
or low recovery or the control condition of no service fail-
ure).
3
Recovery expectations were manipulated by includ-
ing in the scenarios a sign posted at the airline ticket
counter. In the high-recovery expectation condition, the
sign promised a $150 ticket voucher for a delay of 2 hours
or more that was the airline’s responsibility. In the
low-recovery expectation condition, the sign consisted of
a disclaimer in which the airline refused to accept respon-
sibility for delays and cancellations beyond that required
by law. Failure expectations were not manipulated but
were measured before the participants read the experimen-
tal scenario. This approach simplified the research while
allowing the impact of failure expectations to be assessed.
The high- and low-recovery performance conditions
were based on a review of the recovery literature and feed-
back from consumers. Many of the aspects of recovery that
were manipulated were consistent with those identified
subsequent to the data collection by Boshoff (1999) as im-
portant determinants of recovery satisfaction. High recov-
ery was designed to be a prototypical, superb recovery that
would significantly exceed the typical airline response to a
cancellation. In the high-recovery situation the airline
treated the passenger courteously, made several apologies,
and anticipated the traveler’s needs. In addition, the pas-
senger was given a $150 ticket voucher, meal vouchers,
and offered use of the phone. The passenger was rebooked
on the next available flight.
126 JOURNAL OF SERVICE RESEARCH / November 2000
1. Attributions regarding both the service failure and the recovery are
potentially important factors for understanding postrecovery satisfaction
judgments. However, due to the limitations inherent in any one study,
they were not the focus of this research.
2. Examples of crew unavailability might be when crew members are
delayed in arriving at the airport due to a previous flight that is late.
3. Asummary of all manipulations for both Studies 1 and 2 is included
in the appendix.
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

In the low-recovery condition the agent was polite but
not as courteous as under the high-recovery condition and
responded to, but did not anticipate, the traveler’s needs.
The passenger was rebooked on the next available flight.
The passenger was given a meal voucher and offered the
use of a phone but was not offered the $150 ticket voucher.
When, in the high-recovery expectation condition, the pas-
sengers inquired as to whether the $150 ticket voucher
would be issued, the airline agent stated, “No, I am sorry,
but that policy applies to our full-fare passengers, and you
have a discounted ticket.” In the case of both high- and
low-recovery performance, the total length of the delay (3
hours) was the same.
Method of data analysis. The Study 1 disconfirmation
model for the case of service failure was examined using
LISREL-VIII (Jöreskog and Sörbom 1993). The recovery
paradox was evaluated by using analysis of covariance
(ANCOVA).
Measures. The scales employed in Study 1 are given in
Table 1. All questions used a 7-point (strongly dis-
agree-strongly agree) Likert-type scale, except for two of
the Satisfaction scale questions, which were measured us-
ing a 9-point scale to limit skewness (Fornell 1992) and
which employed different anchors (see Table 1 and the fol-
lowing discussion of the specific satisfaction measures).
During the final data collection, respondents reported very
little confusion regarding the questionnaire items. Be-
cause of the uniqueness of the research setting (airline ser-
vice failure and recovery) and the presence of many related
and distinct constructs such as failure and recovery expec-
tations, initial and recovery performance, and initial and
recovery disconfirmation, many measures are substan-
tially original to this research.
Satisfaction. The Satisfaction scale employed three
items. One question, adapted from Westbrook (1980), as-
sessed how well the service experience met the consum-
McCollough et al. / CUSTOMER SATISFACTION 127
TABLE 1
Scale Items and Measurement Properties
Study 1
a
Measurement Item SL IR CR
Failure Expectations .79
In general, I am not surprised if I encounter some kind of problem when I fly. .55 .32
I would consider myself lucky if I did not experience some kind of problem with my flight today. .85 .69
I consider the odds of running into a problem when I fly as being pretty high. .81 .67
Service Performance .55
The airline’s on-time performance was very poor.
b
.48 .22
The airline’s reliability was very high. .74 .51
Recovery Expectations .73
My expectations were high that I would receive compensation if I encountered a long delay. .63 .39
For the situation described, I had very high expectations concerning actions the airline would take to
deal with a lengthy delay. .65 .44
After reading the posted sign, I expected the airline to do whatever it took to guarantee my satisfaction. .76 .58
I didn’t expect this airline to do much for me if I encountered a long delay.
b
–.48 .24
Recovery Performance .83
I would rate the performance of the airline in dealing with the cancellation as exceptional. .81 .64
For the situation described, I would rate the efforts of the airline to deal with my problem as superior. .88 .78
Initial Disconfirmation .56
The airline’s reliability was about what I expected. .47 .12
This airline’s on-time performance was much better than I expected. .77 .28
Recovery Disconfirmation .75
I expected the airline would do more in response to the canceled flight.
b
.72 .52
The compensation for my problem was much better than I expected the airline would provide. –.72 .53
After reading the airline’s posted policy, I expected the airline would do more for me.
b
.68 .48
Final Satisfaction/Postrecovery Satisfaction .90
Overall, how satisfied or dissatisfied did this experience leave you feeling?
c
.88 .77
How well did this service experience meet your needs?
d
.84 .72
Overall, I am very satisfied with this experience. .87 .76
NOTE: Unless noted, all items used a 7-point, strongly disagree/strongly agree, Likert-type format. See Notes c and d below.
a. SL = standardized loadings; IR = item reliability; CR = composite reliability; VE = variance extracted. VE and CR calculated consistent with Fornell and
Larcker (1981).
b. Reverse coded.
c. Used a 9-point, extremely dissatisfied/extremely satisfied, Likert-type format.
d. Used a 9-point, extremely poor/extremely well, Likert-type format.
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

ers’ needs and was anchored by extremely poor/extremely
well. The other satisfaction items were consistent with
common satisfaction measures reported in the literature.
One item treated satisfaction/dissatisfaction as a bipolar
construct, anchored by extremely dissatisfied/extremely
satisfied (with the midpoint labeled neither). This measure
was similar to that employed by Oliver and Bearden
(1985). Consistent with Fornell (1992), the two preceding
satisfaction measures used a 9-point scale to limit skew-
ness. Another satisfaction measure (embedded in a differ-
ent battery of questions) treated satisfaction as a
unidimensional construct (similar to Oliver and Bearden
1985; Westbrook 1980, 1981) and employed a 7-point
(strongly agree/strongly disagree) Likert-type scale. For
all Satisfaction scales, care was taken to ensure that the
measures tapped satisfaction with the service encounter
portrayed in the scenarios and not a generalized attitude re-
garding the service provider by explicitly asking partici-
pants to rate their satisfaction with the service experience
given in the scenarios (see Table 1).
Performance, expectations, and disconfirmation mea-
sures. The research setting and the need to carefully distin-
guish between two types of expectations (failure and
recovery), two sets of performance (initial service perfor-
mance and recovery performance), and two types of
disconfirmation (initial and recovery) necessitated the de-
velopment of original scales. The disconfirmation litera-
ture was examined carefully in constructing these scales.
For instance, both Disconfirmation scales included a ques-
tion related to “much better,” which was adapted from
Swan and Trawick (1981). Consistent with the needs of
this study and the advice of Yi (1990), perceived
disconfirmation measures were employed.
Manipulation checks. Manipulation checks indicated
that the two recovery performance conditions (high/low
[HI/LO]) differed significantly in perceived recovery per-
formance, X
HI/LO
= 5.44/3.54, F(1, 113) = 55.4, p < .0001,
η
2
= .33. The manipulation of recovery performance had a
small effect (based on effect size) on recovery expecta-
tions, X
HI/LO/NO-FAIL
= 4.28/4.85/4.48, F(2, 184) = 3.5, p =
.03, η
2
= .04. Manipulation checks also indicated that the
high- and low-expectation treatments differed signifi-
cantly in perceived recovery expectations, X
HI/LO
=
5.04/4.04, F(1, 184) = 31.6, p < .0001, η
2
= .15. The ma-
nipulation recovery expectations had a slight effect on
perceived recovery performance, X
HI/LO
= 4.27/4.72,
F(1, 113) = 3.2, p = .08, η
2
= .03. Using the guidelines pro-
posed by Perdue and Summers (1986), these results indi-
cate that the manipulations worked as intended.
Measurement model. The analysis of the disconfirmation
model of recovery given in Figure 1 followed the two-step
approach recommended by Anderson and Gerbing (1988)
with the measurement model examined followed by the
structural equations model. The standardized loadings, item
reliabilities, and composite reliability for the items and scales
used to measure the latent constructs in the disconfirmation
model of recovery are given in Table 1. The total sample size
available for the structural model was N = 117. Standardized
loadings (all significant at p < .001) and the item reliabilities
suggest that the constructs’ items cohere reasonably well.
Bagozzi and Yi (1988) suggest as a guideline that composite
reliability should be equal to or greater than about .60. All
scales with the exception of Service Performance (.55) and
Initial Disconfirmation (.56) meet this criterion.
Discriminant validity was assessed using Anderson’s
(1987) criterion that the correlation between two latent
constructs plus or minus two standard errors does not in-
clude one. This criterion is satisfied for all construct pairs
except for between Recovery Disconfirmation and Recov-
ery Performance [0.93 ± 2 (.05) = 1.03, .83]. Thus, even
here the rule is violated only marginally. The discriminant
validity of the scales was also assessed using confirmatory
factor analysis (Anderson and Gerbing 1988), focusing
specifically on the conceptually similar constructs that
could potentially overlap: initial disconfirmation, recov-
ery disconfirmation, and satisfaction. The results of this
analysis suggest that the three-factor solution is better than
a one-factor solution (three-construct solution: χ
2
[17] =
26.16; one-construct solution: χ
2
[20] = 65.76; χ
2
[3] differ-
ence = 39.59, p < .001). Thus, the measures of these con-
ceptually similar constructs are empirically distinct.
The overall fit of the structural model (i.e., the Study 1
disconfirmation portion of the model portrayed in Figure 1)
was first assessed by examining the chi-square statistic,
which is significant, χ
2
(140) = 262.3, p < .001. However,
as this statistic is sensitive to sample size (Bagozzi and Yi
1988), additional measures of fit were examined. Bentler
and Bonett’s (1980) Normed Fit Index (NFI) (D) is .79,
whereas Tucker and Lewis’s (1973) Non-Normed Fit In-
dex (NNFI) is .86. The Goodness-of-Fit Index (GFI) is .82,
whereas the Adjusted Goodness-of-Fit Index (AGFI) is
.75. The Comparative Fit Index (CFI) is .89, and the Incre-
mental Fit Index (IFI) is .89. The standardized root mean
square residual (RMR) is .077. The root mean square error
of approximation (RMSEA) was .08. The modification in-
dexes did not suggest any significant modifications of ei-
ther the measurement or structural model.
Thus, as was the case for the measurement model, the
structural model demonstrates modest fit. When assessing
the overall fit of the model, it is important to take into con-
sideration the number of related constructs that were mod-
eled, the use of new scales for many of the constructs, and
the nature of the research setting. For instance, two differ-
ent types of expectations (failure and recovery), two differ-
128 JOURNAL OF SERVICE RESEARCH / November 2000
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

ent types of performance (initial service performance and
recovery), and two different types of disconfirmation (ini-
tial and recovery) were included in the model. Given the
overall adequate fit of the model, an examination of the hy-
potheses tests was deemed appropriate.
Hypotheses tests. Table 2 provides the standardized es-
timates and t-values for the hypothesized relationships in-
volving service failure and recovery. The direction of all
paths is consistent with that hypothesized and is signifi-
cant at p < .001, with the exception of the effect of failure
expectations on initial disconfirmation (p < .1).
Failure expectations exert a relatively weak, positive ef-
fect on initial disconfirmation. Those with low (high) fail-
ure expectations felt the airlines’ reliability was worse
(better) than expected. However, Hypothesis 1 is not sup-
ported at p < .05. Service performance exerts a significant
positive influence on initial disconfirmation, supporting
Hypothesis 2. In turn, initial disconfirmation has a signifi-
cant positive effect on satisfaction. Thus, consistent with
Hypothesis 3, those who reported the airlines’ reliability
was worse (better) than expected reported greater dissatis-
faction (satisfaction).
The recovery disconfirmation relationships mirrored
the initial disconfirmation findings. Those with high (low)
recovery expectations perceived the recovery performance
as worse (better) than expected, supporting Hypothesis 4.
Also, consistent with Hypothesis 5, perceived recovery
performance had a significant effect on recovery
disconfirmation. Finally, recovery disconfirmation ex-
erted a positive effect on satisfaction (Hypothesis 6).
Those who perceived that the recovery was better (worse)
than expected were more satisfied (dissatisfied).
Examining the relative magnitudes of the standardized
estimates sheds some light on the strength of the relation
-
ships in the model. In general, the impact of performance
(initial or recovery) on disconfirmation is greater than the
effect of expectations (either failure or recovery) on
disconfirmation. This finding mirrors the consistent find-
ings from the disconfirmation literature that performance
exerts a greater impact on disconfirmation than expecta-
tions. Interestingly, initial disconfirmation exerts a much
stronger impact on satisfaction than does recovery
disconfirmation. This implies that recovery can only miti-
gate the impact of failure on satisfaction and that initial ser-
vice performance is the primary driver of final satisfaction.
As stated in Hypothesis 7, the recovery paradox was
proposed to be most likely given (a) high failure expecta-
tions, (b) low recovery expectations, and (c) high recovery
performance. Under these circumstances, failure should
be the least disconfirming and recovery should have the
greatest positive disconfirmation effect, potentially result-
ing in satisfaction equal to or greater than the control case
of no-service failure. To evaluate this hypothesis, an
ANCOVA model (using the general linear model) was run
with satisfaction as the dependent variable and the treat-
ment conditions entered as the main effects and failure
expectations
4
treated as a covariate. Figure 2 graphically
portrays the mean satisfaction ratings in the failure and
no-failure conditions.
To test the recovery paradox, the mean of the low-
recovery expectations and high-recovery performance
group was compared with the means of the no-failure con-
ditions. The mean of the low-recovery expectations and
high-recovery performance group is 5.69, significantly
lower than the mean of either of the no-failure groups (the
mean of the high-recovery expectations and no-failure
condition was 7.15, p < .001; the mean of the low-recovery
expectations and no-failure condition was 6.96, p < .001).
These results do not support a recovery paradox effect, and
Hypothesis 7 is rejected.
Study 2: Distributive
and Interactional Justice
Study 2, a 3 × 3 between-subjects design, evaluated Hy-
potheses 8 through 10 by manipulating three levels of dis-
tributive justice (high, moderate, and low) and three levels
of interactional justice (high, moderate, and low). In addi-
tion, a control condition of no-service failure was in-
cluded. The failure was, as in the case of Study 1, depicted
as “crew unavailability,” resulting in a 3-hour delay to the
passenger. Under high DJ, the passenger received a $150
McCollough et al. / CUSTOMER SATISFACTION 129
TABLE 2
Standardized Estimates of
Hypothesized Relationships
Standardized t-
Relationship (Hypothesis) Estimate Value
Effect of failure expectation on
Initial Disconfirmation (Hypothesis 1: +) .17 1.34**
Effect of service performance on
Initial Disconfirmation (Hypothesis 2: +) .96 3.60*
Effect of initial disconfirmation on
Satisfaction (Hypothesis 3: +) .70 4.56*
Effect of recovery expectation on
Recovery Disconfirmation (Hypothesis 4: –) –.51 –4.82*
Effect of recovery performance on
Recovery Disconfirmation (Hypothesis 5: +) .71 7.02*
Effect of recovery disconfirmation on
Satisfaction (Hypothesis 6: +) .43 4.68*
*p < .001 (one-tailed). **p < .10 (one-tailed).
4. The recovery paradox was evaluated on a sample composed of
those that responded to scenarios involving failure and recovery (the
same as the path analysis data analysis) as well as the control condition of
no-service failure.
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

ticket voucher, a meal voucher, and the use of a phone. Un-
der moderate DJ, the passenger did not receive a ticket
voucher but did receive meal vouchers and use of a phone.
Under low DJ, the passengers did not receive a ticket
voucher, meal voucher, or use of the phone (see the appen-
dix for an overview of the manipulations).
Under high IJ, the agent was portrayed as anticipating
the passengers’ needs and as being very professional,
courteous, and apologetic. Under moderate IJ, the agent
was less apologetic and responded to, but did not antici-
pate, passengers’ needs. Under low IJ, the agent was rude
and indifferent to passengers’ plights and did not offer an
apology. In addition, under low IJ, and unlike the high and
moderate conditions, the passenger was placed on standby
and not immediately rebooked on another flight. However,
to keep the objective level of harm caused by the failure the
same, the total length of the delay (3 hours) was the same
in all conditions.
Method of data analysis. For the Study 2 justice model,
analysis of variance (ANOVA) using the general linear
model was employed.
Measures. The items composing the interactional and
distributive justice scales are given in Table 3. Some items
were adapted from Oliver and Swan (1989a). Others were
developed to assess aspects of justice that have been cited as
important in the recovery literature such as compensation
for out-of-pocket expenses and frustration and the presence
of an apology (Goodwin and Ross 1992). The items com-
posing the Satisfaction scale were identical to Study 1. The
Study 2 coefficient alpha for satisfaction is .92 (N = 376).
Thus, all scales demonstrate adequate reliability.
Manipulation checks. Manipulation checks using the
perceived-IJ scale indicated that the IJ manipulation was
successful, X
HI/MOD/LO
= 5.56/5.19/2.37, F(2, 314) = 223.1,
p < .0001, η
2
= .59. The difference between the high- and
moderate-IJ treatment is significant at p = .03, whereas the
difference between all other mean comparisons is signifi-
cant at p < .0001. The manipulation of IJ had a slight but
significant effect on DJ, X
HI/MOD/LO
= 3.26/3.28/2.40, F(2,
302) = 18.9, p < .0001, η
2
= .11. The check of the manipu-
lation of DJ indicates that this treatment also was success-
ful. The respective perceived-DJ means for the DJ
treatment groups are X
HI/MOD/LO
= 4.19/2.87/1.87, F(2,
302) = 92.4, p < .0001, η
2
= .39. The difference between all
means is significant at p < .0001. Manipulating DJ had a
slight, significant effect on IJ, X
HI/MOD/LO
= 4.92/4.55/3.64,
F(2, 314) = 31.7, p < .0001, η
2
= .17. A comparison of the
η
2
indicates that the manipulations were successful (see
Perdue and Summers 1986).
Hypotheses tests. To evaluate the justice hypotheses, an
ANOVA model was run with satisfaction as the dependent
variable and the main effect treatments and interaction as
the dependent variables. Figure 3 graphically presents the
effect of interactional and distributive justice on
postrecovery satisfaction. In general, the higher the dis-
tributive justice, the higher the satisfaction. The impact of
interactional justice is most pronounced when comparing
either high or moderate IJ with low IJ. DJ had a significant
effect on postrecovery satisfaction, X
HI/MOD/LO
=
4.63/3.44/2.20, F(2, 319) = 50.7, p < .0001, η
2
= .24. The
difference between all means is significant at p ≤ .0001.
Therefore, Hypothesis 8 is supported. IJ also had a signifi
-
cant effect on postrecovery satisfaction, X
HI/MOD/LO
=
130 JOURNAL OF SERVICE RESEARCH / November 2000
FIGURE 2
Satisfaction in the Failure
and No-Failure Conditions
TABLE 3
Scale Items and Coefficient Alpha for Study 2
Scale Items Cronbach’s Alpha
Interactional Justice .79
The agent demonstrated a poor understanding
of my needs.
a
In dealing with me, the agent treated me in a
courteous manner.
I was offered an apology.
Distributive Justice .83
This trip resulted in a very positive outcome for me.
I was more than compensated for any
out-of-pocket expenses I might have incurred.
I was more than compensated for any frustration.
I got more out of this transaction than the airline.
NOTE: All scales used a 7-point strongly disagree/strongly agree
Likert-type format.
a. Reverse coded.
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

4.10/4.01/2.16, F(2, 319) = 44.5, p < .0001, η
2
= .22. The
difference between high and moderate IJ is not significant
(p = .7005), whereas all other mean differences are signifi-
cantly different at p ≤ .0001. Therefore, Hypothesis 9 is
supported. The interaction between IJ and DJ is also sig-
nificant, supporting Hypothesis 10, F(4, 319) = 2.8, p =
.0265, η
2
= .03.
To shed further light on the effect of IJ and DJ on
postrecovery satisfaction, the interaction effect was iso-
lated from the main effects for each cell (interaction effect =
cell mean – grand mean – row effect – column effect)
where the grand mean is the overall mean, the row mean ef-
fect = row mean – grand mean, and the column mean = col-
umn mean – grand mean (see Ross and Creyer 1993). The
interaction effects, presented in Table 4, are illuminating.
When the relative effects of IJ and DJ are similar (the diag-
onal elements), the interaction is “positive.” In these cases,
IJ and DJ act together to produce greater postrecovery sat-
isfaction than average. However, when the main effects are
mismatched (the off-diagonal elements), the interaction
approaches zero and becomes negative the more “mis-
matched” the respective justice elements are. For instance,
the most negative values occur in the case of high DJ and
low IJ or high IJ and low DJ. These negative interactions
for cases of high or moderate IJ and low DJ are consistent
with a “sham” effect (Cohen 1985; Goodwin and Ross
1992) or “hollow” justice (Greenberg 1990a). These
findings also mirror those of Tax, Brown, and
Chandrashekaran (1998), who found significant interac-
tions among DJ and PJ as well as DJ and IJ. Likewise,
Hirschman (1970) notes that voice that leads to frustration
may actually increase exiting behavior. In the case of high
IJ and low DJ, it is possible the respondents doubted the
sincerity of the gate agents’ apologies when even the most
basic of compensation was not provided. Likewise, when
gate agents provided high levels of compensation without
corresponding levels of empathy and understanding (high
DJ and low IJ), passengers may have felt they were being
“bought off.”
No specific recovery paradox hypothesis was proposed
concerning Study 2. However, in an attempt to gain a
greater understanding of the recovery paradox, the high-IJ
and high-DJ cell was compared to the no-failure control
group for satisfaction. The mean satisfaction of the
no-failure condition (7.43) is significantly higher (p <
.0001) than the mean satisfaction for the high-IJ and
high-DJ failure group (5.52). Therefore, replicating the
findings of Study 1, no recovery paradox emerges on the
strength of recovery performance alone.
DISCUSSION AND IMPLICATIONS
In general, the research hypotheses were well sup-
ported, with evidence confirming both the disconfirmation
model as well as the important role of justice. Among the
most intriguing findings of this research are (a) a lack of
support for a recovery paradox effect, (b) the impact of
both initial and recovery disconfirmation on final
postrecovery satisfaction, (c) the relatively greater impact
of initial disconfirmation versus recovery disconfirmation
on satisfaction, and (d) the finding of an interaction be-
tween DJ and IJ, which underscores the importance of pro-
viding consistent levels of these two types of justice in
recovery efforts. Following is a review and discussion of
the implications of these findings.
Disconfirmation
Ultimately, final satisfaction provides the litmus test
against which the impact of service failure and recovery
should be judged. In Study 1, satisfaction is greatest in the
case of no failure. Given failure, the path analysis shows
that initial disconfirmation has a greater impact on satis-
faction than recovery disconfirmation. In turn, initial ser-
vice performance is the primary predictor of initial
disconfirmation, whereas recovery performance is the pri-
mary predictor of recovery disconfirmation. Therefore,
consistent with Halstead and Page (1992), it appears that
satisfaction is primarily driven by the initial service fail-
ure with recovery performance acting to mitigate the
damage to satisfaction caused by the failure. However,
given the impact of recovery performance, recovery ex-
pectations, and recovery disconfirmation on postrecov-
ery satisfaction, failure service providers should strive to
offer high-recovery performance that exceeds customer
expectations.
McCollough et al. / CUSTOMER SATISFACTION 131
FIGURE 3
The Effect of Interactional and Distributive
Justice on Postrecovery Satisfaction
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

Interactional and
Distributive Justice
Both DJ and IJ are important predictors of postrecovery
satisfaction. Research reported in the organizational be-
havior literature has found that when the initially desired
outcome is blocked, evaluations can still be relatively
high given high IJ or PJ (Folger and Konovsky 1989;
Greenberg 1990b; McFarlin and Sweeney 1992). In this
case, the originally desired outcome (arriving at their
destination on time) was unobtainable; therefore, IJ (the
how) may have added importance in determining
postrecovery satisfaction.
To reap the maximum impact on postrecovery satisfac-
tion, the relative levels of IJ and DJ must be consistent. A
synergistic effect seems to emerge when the levels of IJ and
DJ are similar. In this case, the sum of the effect of IJ and DJ
on postrecovery satisfaction is greater than the individual
effects. Conversely, situations in which IJ and DJ are mis-
matched appear to create a situation where DJ and IJ work
against one another. For instance, high IJ and low DJ may be
perceived by consumers as indicating that the service pro-
vider’s apologies are insincere. In the extreme, a “sham ef-
fect” appears to emerge. Conversely, high DJ and low IJ
may leave consumers feeling “bought off” by a service pro-
vider who is not sincerely interested in their welfare.
For airlines (and service providers, in general), the find-
ing that IJ and DJ are both necessary is noteworthy. Al-
though expensive, giving monetary compensation is a
straightforward response. Delivering IJ may be more of a
challenge. Even the best gate agent might find it a chal-
lenge to be truly empathetic and responsive when facing a
long line of upset customers following a flight cancella-
tion. Service managers should consider special training
and support services that help frontline service providers
respond effectively to service failures.
The Recovery Paradox
By using a control condition of no-service failure, this
research was able to directly evaluate the recovery para-
dox, whereas most previous studies that lacked such a con-
dition have only been able to report results that indirectly
addressed the recovery paradox. In this study, no support
was found for a recovery paradox. In neither Study 1 nor
Study 2 were the mean satisfaction ratings of participants
who experienced service failure as great or greater than
those who did not experience service failure. One possible
explanation for the lack of a recovery paradox in this re-
search is that for the service failure portrayed, a 3-hour de-
lay, no recovery effort can completely mitigate the harm
caused by the failure. Although travelers might find the
$150 ticket voucher and the smooth and professional re
-
covery portrayed as a pleasant surprise, the recovery effort
cannot completely erase the harm caused by the failure.
For instance, if the delay causes a business traveler to miss
an important meeting, no realistic compensation is apt to
totally erase the harm caused by the failure.
Data were collected in both studies on the perceived
harm caused by the service failure. To explore further the
role of harm in determining satisfaction, an ANCOVA
model was run for both studies with satisfaction as the de-
pendent variable and the experimental main effects and the
perceived harm covariate
5
as the independent variables. In
Study 1, the harm variable is significant. The mean of the
high-recovery performance/low-recovery expectation
condition (adjusted for harm) is 5.85. This is not statisti-
cally greater than the mean of either the no-failure/high-
recovery expectation condition of 6.26 (p = .28) or the
mean of the no-failure/low-recovery expectation condi-
tion of 6.34 (p = .14). This supports the possibility of a re-
covery paradox effect when the harm caused by the failure
is taken into account. The results are similar for Study 2.
The mean for the control condition of no failure is 5.97,
whereas the mean of the high-IJ and high-DJ condition is
5.24 (p = .1006). Therefore, when full or near-full recovery
is possible, superior recovery efforts might be able to pro-
duce a recovery paradox effect. This finding is consistent
in some respect with the finding of Smith and Bolton
(1998) that postrecovery cumulative satisfaction could in-
crease if the customer was very satisfied with the recovery
effort and the magnitude of the failure was considered.
Likewise, Webster and Sundaram (1998) and Sundaram,
Jurowski, and Webster (1997) report that service con-
sumption criticality, similar to the construct of harm dis-
cussed here, interacts with recovery effort to influence
customer satisfaction.
132 JOURNAL OF SERVICE RESEARCH / November 2000
TABLE 4
Interactional and Distributive
Justice Interaction Effects on
Postrecovery Satisfaction (Study 2)
Interactional Justice
High Moderate Low
Distributive justice
High .1463 .0535 –.3966
Moderate .0629 .1827 –.3423
Low –.4058 –.3329 .5422
NOTE: Interaction effect = cell mean – grand mean – row effect – column
effect. Grand mean = overall mean; row effect = row mean – grand mean;
column mean = column mean – grand mean. See Ross and Creyer (1993).
5. The following three items composed the perceived harm scale: (a)
This service encounter created a major problem for me; (b) This service
experience caused me a great deal of inconvenience; and (c) Overall, this
service performance did not result in any real harm (reverse coded). Co
-
efficient alpha Study 1 = .84, Study 2 = .74.
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

The finding that a recovery paradox effect may be pos
-
sible when the recovery effort can completely mitigate the
harm caused by failure may explain why some research
has indicated that a recovery paradox effect is possible,
whereas others have not. Hocutt, Chakraborty, and Mowen
(1997a) found that under conditions of high redress, re-
sponsiveness, empathy, and courtesy, postrecovery satis-
faction can be higher than the control condition of
no-service failure. However, this recovery paradox effect
may be explained by the failure, in this case an improperly
prepared steak in a restaurant. Under the high-recovery
performance condition, the steak was replaced, the cus-
tomer was not charged, and the server was portrayed as re-
sponsive and empathetic. This would appear to be a case of
relatively low harm in which the recovery effort could
completely mitigate the harm caused by the failure. Would
the results have been different had preparing another steak
required 20 additional minutes and made the diners late for
a show?
Conversely, Kelly, Hoffman, and Davis (1993), in a
study of retail failure and recovery, reported that “correc-
tion plus recovery” (which involved additional compensa-
tion beyond the correction of the failure) was rated slightly
less favorably than recovery strategies that simply cor-
rected the problem. In explaining this counterintuitive
finding, they note that most failures that resulted in cor-
rection plus recoveries were generally more severe in na-
ture. Thus, the lower rating of the correction plus
recoveries might have been due to the severity of the initial
failure.
Recovery paradox effects are limited by the require-
ment that in most situations, customers must seek redress
for recovery to occur. In the case of airline delays and can-
cellations, most consumers seek redress or correction of
the problem. However, for many service failures, custom-
ers choose not to complain. In fact, under low-harm failure
conditions, the customer is less likely to complain, and re-
dress seeking may occur less often than under high-harm
failure conditions (Richins 1983, 1987). Conversely, in a
study of postcomplaint satisfaction, Tax, Brown, and
Chandrashekaran (1998) found that most complaints re-
sulted from problems judged by the consumer to be highly
important.
It also should be remembered that harm is specific to
the individual and the context. One customer’s low-harm
failure is another’s high-harm failure. Consider the differ-
ence in harm incurred between a dine-in customer and a
drive-through customer who receive the wrong sand-
wiches at a fast-food restaurant. Likewise, consider the
difference in harm caused by a service failure between an
important business meeting and informal dinner with ca
-
sual acquaintances. Clearly, the issue of the harm caused
by the failure is worthy of future study.
Is Service Recovery
an Opportunity?
Does service recovery present an organization with an
opportunity to improve customer satisfaction as argued by
some (Abrams and Paese 1993; Hart, Heskett, and Sasser
1990)? Our finding that the higher the recovery perfor-
mance, the higher the postrecovery satisfaction supports
the importance of superior service recovery. Inferior re-
covery performance can lead to what Bitner, Booms, and
Tetreault (1990) termed a double deviation from customer
expectations: The firm fails to deliver on the initial service
and the recovery service. Therefore, superior recovery
could be viewed as an opportunity when compared to infe-
rior recovery.
A consistent finding from the literature is that most dis-
satisfied customers never bother to complain. If superior
service recoveries encourage customers who would not or-
dinarily complain to seek redress, then overall postservice
failure customer satisfaction could be enhanced following
outstanding service recovery. Therefore, postrecovery sat-
isfaction following superior recovery is an opportunity
when viewed against the dissatisfaction of those who do
not seek redress. In addition, organizations may capitalize
on the information gained from such complaints to design
more reliable service delivery systems (Tax and Brown
1998). In this respect, service recovery can be viewed as an
opportunity to gain access to superior market intelligence
on the cause of customer dissatisfaction, which can, in
turn, lead to more reliable service offerings.
However, our research indicates that excellent recovery
is not an opportunity when compared to the satisfaction re-
sulting from error-free service delivery. Error-free service
is the better option for a variety of reasons, including cus-
tomer confidence in the company’s reliability and the ab-
sence, by and large, of a recovery paradox. Smith and
Bolton (1998), in reporting evidence supporting a recov-
ery paradox effect, observed that such an effect depended
on achieving consistently high-service recovery perfor-
mance. Uneven- or poor-service recovery risks alienating
and losing customers. Our findings support the work of
quality researchers (e.g., Crosby 1979; Deming 1986)
who conclude that organizations should strive to identify
and eliminate all potential sources of failure before the
consumption experience to maximize both firm financial
performance and customer satisfaction.
LIMITATIONS AND DIRECTIONS
FOR FUTURE RESEARCH
This study relied on scenario-based experiments, an ap
-
proach with strong precedent, and specific steps increased
McCollough et al. / CUSTOMER SATISFACTION 133
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

realism and generalizability. Nevertheless, different meth
-
odologies should be employed to confirm and possibly ex-
tend the conclusions of this research. A survey approach
could replicate many of the findings here and increase
generalizability. The Critical Incident Technique (Bitner,
Booms, and Tetreault 1990) could be useful in identifying
specific actions service providers can take to ensure suc-
cessful recoveries.
The empirical conclusions in this study are based on
findings from one service industry. Caution must be exer-
cised in extending the conclusions of this study to other
services. For instance, it is possible that customer reac-
tions to service failure and recovery differ based on their
involvement in the service. As an example, customers
rarely notice electrical services unless something goes
wrong. Failure for such a service would result in elevating
a “low-involvement” service to a “high-involvement” one.
If recovery is superior, a recovery paradox effect might
emerge when comparing the “unequal” satisfaction of the
no-failure low-involvement group with that of the high-
involvement postrecovery satisfaction group. Therefore,
researchers should investigate customer satisfaction after
low-involvement service failure and recovery.
Furthermore, although adequate, the fit indexes for the
disconfirmation model of service failure and recovery are
modest. Future research should seek improved measures
of the constructs investigated here. In addition, the struc-
tural model might be enhanced by the incorporation of ad-
ditional or alternative constructs such as justice or the
incorporation of service failure harm.
The role of attributions regarding both the service fail-
ure and the recovery effort deserves study. The locus for
the service failure was fixed in this research on the service
provider, and no effort was made to manipulate stability
and controllability attributions. However, Hocutt,
Chakraborty, and Mowen (1997b) report that postrecovery
satisfaction following service failure due to the customers’
actions equals the control condition of no-service failure.
Therefore, more attention should be paid to service fail-
ures contributed to by the consumers’ actions or an act of
God. Likewise, the role of harm in determining
postrecovery satisfaction and a recovery paradox deserves
additional research.
134 JOURNAL OF SERVICE RESEARCH / November 2000
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

REFERENCES
Abrams, Michael and Matthew Paese (1993), “Wining and Dining the
Whiners: Treating Complaints as Opportunities Is One Customer
Strategy That Really Works,” Sales and Marketing Management, 145
(February), 72-75.
Anderson, James C. (1987), “An Approach for Confirmatory Measure-
ment and Structural Equation Modeling of Organizational Properties,”
Management Science, 33 (4), 525-41.
and David W. Gerbing (1988), “Structural Equations Modeling in
Practice: A Review and Recommended Two-Step Approach,” Psy
-
chological Bulletin, 103 (3), 411-23.
Andreassen, Thor Wallin (1999), “What Drives Customer Loyalty with
Complaint Resolution?” Journal of Service Research, 1 (May),
324-32.
Andreeva, Nellie (1998), “Unsnarling Traffic Jams at U.S. Airports,”
Business Week, August 10, 84.
Bagozzi, Richard P. and Youjae Yi (1988), “On the Evaluation of Struc
-
tural Equation Models,” Journal of the Academy of Marketing Sci
-
ence, 16 (Spring), 74-94.
Bearden, William O. and Jesse E. Teel (1983), “Selected Determinants of
Consumer Satisfaction and Complaint Reports,” Journal of Mar-
keting Research, 20 (February), 21-28.
Bentler, Peter M. and Douglas G. Bonett (1980), “Significance Tests and
Goodness of Fit in the Analysis of Covariance Structures,” Psycho-
logical Bulletin, 8 (3), 588-606.
Berry, Leonard L., Valarie A. Zeithaml, and A. Parasuraman (1990),
“Five Imperatives for Improving Service Quality,” Sloan Manage-
ment Review, 31 (Summer), 29-39.
Bies, Robert J. and Joseph S. Moag (1986), “Interactional Justice: Com-
munication Criteria of Fairness,” in Research on Negotiation in Orga
-
nizations, Vol. 1, R. Lewicki, M. Bazerman, and B. Sheppard, eds.
Greenwich, CT: JAI, 57-79.
Bitner, Mary Jo (1990), “Evaluating Service Encounters: The Effects of
Physical Surroundings and Employee Responses,” Journal of Mar
-
keting, 54 (April), 69-82.
, Bernard H. Booms, and Mary Stanfield Tetreault (1990), “The
Service Encounter Diagnosing Favorable and Unfavorable Inci
-
dents,” Journal of Marketing, 54 (January), 71-84.
Blodgett, Jeffrey G., Donald H. Granbois, and Rockney G. Walters
(1993), “The Effect of Perceived Justice on Complainants’ Negative
McCollough et al. / CUSTOMER SATISFACTION 135
APPENDIX
Summary of Manipulations
Study 1 Manipulations
High-recovery expectations • Sign in scenario text promises a voucher for $150 off the purchase price of a round-trip ticket for delays of 2 hours
or more if the airline is responsible for the delay
Low-recovery expectations • Sign in scenario text declaims any responsibility for delayed or canceled flights
High-recovery performance • Agent checks to see if passenger can be rebooked on another airline without prompting
• Three apologies are offered
• Passenger is offered a $150 ticket voucher, a meal voucher, use of a phone for local or long-distance calls, and use
of lounge
• Agent asks if there is anything else the passenger needs
• Rebooks passenger on a later flight
Low-recovery performance • Agent checks to see if passenger can be rebooked on another airline after being prompted
• Two apologies are offered
• With prompting offers use of phone and meal voucher, refuses $150 ticket voucher
• Rebooks passenger on a later flight
Study 2 Manipulations
High distributive justice • Passenger is offered a $150 ticket voucher, a meal voucher, and use of phone for local or long-distance calls
Moderate distributive justice • Passenger is offered meal voucher and use of phone for local and long-distance calls
Low distributive justice • Passenger is refused meal voucher and use of phone
High interactional justice • Agent checks to see if passenger can be rebooked on another airline without prompting
• Three apologies are offered
• Agent asks if there is anything else the passenger needs
• Rebooks passenger on a later flight
Moderate interactional justice • Agent checks to see if passenger can be rebooked on another airline after being prompted
• Two apologies are offered
• Rebooks passenger on a later flight
Low interactional justice • Agent refuses to check to see if passenger can be rebooked on another flight, noting all future flights are booked
• No apology is offered
• Passenger is placed on standby (total length of the delay is the same as in the case of high and moderate
interactional justice)
• Agent is portrayed as in a hurry to move on to the next customer
• The customer must go to the customer service desk for meal voucher and use of the phone (high and moderate
distributive-justice conditions only)
Studies 1 and 2 No-Failure Conditions
No-failure condition • Scenario portrays an uneventful flight with no problems
• Agent is polite and friendly
• Customer arrives at final destination on time
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

Word-of-Mouth Behavior and Repatronage Intentions,” Journal of
Retailing, 69 (Winter), 399-428.
Bolton, Ruth N. and James H. Drew (1991), “A Multistage Model of Cus
-
tomers’ Assessments of Service Quality and Value,” Journal of Con
-
sumer Research, 17 (March), 375-84.
Boshoff, Christo (1999), “RECOVSAT: An Instrument to Measure Satis-
faction with Transaction-Specific Service Recovery,” Journal of Ser-
vice Research, 1 (February), 236-49.
Cohen, Ronald L. (1985), “Procedural Justice and Participation,” Human
Relations, 38 (July), 643-63.
Crosby, Phillip B. (1979), Quality Is Free: The Art of Making Quality
Certain. New York: McGraw-Hill.
Deming, W. Edwards (1986), Out of Crisis. Cambridge: MIT Press.
Etzel, Michael J. and Bernard I. Silverman (1981), “A Managerial Per-
spective on Directions for Retail Customer Dissatisfaction Re-
search,” Journal of Retailing, 57 (Fall), 124-36.
Fisk, Raymond P., Stephen W. Brown, and Mary Jo Bitner (1993),
“Tracking the Evolution of the Services Marketing Literature,” Jour-
nal of Retailing, 69 (Spring), 61-103.
Folger, Robert and Mary A. Konovsky (1989), “Effects of Procedural and
Distributive Justice on Reactions to Pay Raise Decisions,” Academy
of Management Journal, 32, 115-30.
, David Rosenfield, Janet Grove, and Louise Corkran (1979), “Ef-
fects of ‘Voice’ and Peer Opinions on Responses to Inequity,” Jour-
nal of Personality and Social Psychology, 37 (December), 2253-61.
Folkes, Valerie S. (1984), “Consumer Reactions to Product Failure: An
Attributional Approach,” Journal of Consumer Research,10
(March), 398-409.
 and Barbara Kotsos (1986), “Buyers’ and Sellers’ Explanations
for Product Failure: Who Done It?” Journal of Marketing, 50 (April),
74-80.
Fornell, Claus (1992), “A National Customer Satisfaction Barometer:
The Swedish Experience,” Journal of Marketing, 56 (January), 6-21.
 and David F. Larcker (1981), “Evaluating Structural Equation
Models with Unobserved Variables and Measurement Error,” Jour-
nal of Marketing Research, 18 (February), 39-50.
Goodwin, Cathy and Ivan Ross (1992), “Consumer Responses to Service
Failures: Influence of Procedural and Interactional Fairness Percep-
tions,” Journal of Business Research, 25 (September), 149-63.
Greenberg, Jerald (1987), “Reactions to Procedural Injustice in Payment
Distributions: Do the Means Justify the Ends?” Journal of Applied
Psychology, 72 (February), 55-61.
 (1990a), “Looking Fair vs. Being Fair: Managing Impressions of
Organizational Justice,” Research in Organizational Behavior, 12,
111-57.
 (1990b), “Organizational Justice: Yesterday, Today, and Tomor-
row,” Journal of Management, 16 (June), 399-432.
 and Robert Folger (1983), “Procedural Justice, Participation, and
the Fair Process Effect in Groups and Organizations,” in Basic Group
Processes, P. B. Paulus, ed. New York: Springer-Verlag, 235-56.
Gronroos, Christian (1988), “Service Quality: The Six Criteria of Good
Perceived Service Quality,” Review of Business, 9 (Winter), 10-13.
Halstead, Diane and Thomas J. Page, Jr. (1992), “The Effects of Satisfac-
tion and Complaining Behavior on Consumer Repurchase Inten-
tions,” Journal of Consumer Satisfaction, Dissatisfaction, and
Complaining Behavior, 5, 1-11.
Hansen, David E. and Peter J. Danaher (1999), “Inconsistent Perfor-
mance during the Service Encounter: What’s a Good Start Worth?”
Journal of Service Research, 1 (February), 227-35.
Hart, Christopher W. L. (1993), Extraordinary Guarantees. New York:
American Management Association.
, James L. Heskett, and W. Earl Sasser, Jr. (1990), “The Profitable
Art of Service Recovery,” Harvard Business Review, 68 (July-August),
148-56.
Hirschman, Albert O. (1970), Exit, Voice, and Loyalty-Responses to De
-
cline in Firms, Organizations, and States. Cambridge, MA: Harvard
University Press.
Hocutt, Mary Ann, Goutam Chakraborty, and John C. Mowen (1997a),
“The Art of Service Recovery: Fact or Fiction? An Empirical Study
of the Effects of Service Recovery,” in Marketing Theory and Appli
-
cations, Debbie Thorne LeClair and Michael Hartline, eds. Chicago:
American Marketing Association, 50-51.
, , and  (1997b), “The Impact of Perceived Justice
on Customer Satisfaction and Intention to Complain in a Service Re-
covery,” in Advances in Consumer Research, Merrie Brucks, ed. Ann
Arbor, MI: Association for Consumer Research, 457-63.
Jöreskog, Karl G. and Dag Sörbom (1993), LISREL 8: Structural Equa
-
tion Modeling with the SIMPLIS Command Language. Chicago: Sci-
entific Software International.
Kelly, Scott W. and Mark A. Davis (1994), “Antecedents to Customer Ex-
pectations for Service Recovery,” Journal of Academy of Marketing
Science, 22 (Winter), 52-61.
, K. Douglas Hoffman, and Mark A. Davis (1993), “A Typology
of Retail Failures and Recoveries,” Journal of Retailing, 69 (Winter),
429-52.
McCollough, Michael A. and Sundar G. Bharadwaj (1992), “The Recov-
ery Paradox: An Examination of Consumer Satisfaction in Relation
to Disconfirmation, Service Quality, and Attribution Based The-
ories,” in Marketing Theory and Applications, Chris T. Allen,
Thomas J. Madden, Terence A. Shimp, Roy D. Howell, George M.
Zinkhan, Deborah D. Heisley, Richard J. Semenik, Peter Dickson,
Valarie Zeithaml, and Roger L. Jenkins, eds. Chicago: American
Marketing Association, 119.
McFarlin, Dean B. and Paul D. Sweeney (1992), “Distributive and Proce-
dural Justice as Predictors of Satisfaction with Personal and Organi-
zational Outcomes,” Academy of Management Journal, 35 (August),
626-37.
Morris, Susan V. (1988), “How Many Lost Customers Have You Won
Back Today? An Aggressive Approach to Complaint Handling in the
Hotel Industry,” Journal of Consumer Satisfaction, Dissatisfaction,
and Complaining Behavior, 1, 86-92.
Murray, Keith B. and John L. Schlacter (1990), “The Impact of Services
versus Goods on Consumers’ Assessment of Perceived Risk and
Variability,” Journal of the Academy of Marketing Science,18(Win-
ter), 51-65.
Oliver, Richard L. (1980), “A Cognitive Model of the Antecedents and
Consequences of Satisfaction Decisions,” Journal of Marketing Re-
search, 17 (September), 46-49.
 (1981), “Measurement and Evaluation of Satisfaction Processes
in Retail Settings,” Journal of Retailing, 57 (Fall), 25-48.
 (1989), “Processing of the Satisfaction Response in Consump-
tion: A Suggested Framework and Research Propositions,” Journal
of Consumer Satisfaction, Dissatisfaction, and Complaining Behav-
ior, 2, 1-16.
 (1993), “Cognitive, Affective, and Attribute Bases of the Satis-
faction Response,” Journal of Consumer Research, 20 (December),
418-30.
 and William O. Bearden (1985), “Disconfirmation Processes and
Consumer Evaluations in Product Usage,” Journal of Business Re-
search, 13 (June), 235-46.
 and Raymond R. Burke (1999), “Expectation Processes in Satis-
faction Formation,” Journal of Service Research, 1 (February),
196-214.
 and John E. Swan (1989a), “Consumer Perceptions of Interper-
sonal Equity and Satisfaction in Transactions: A Field Survey Ap-
proach,” Journal of Marketing, 53 (April), 21-35.
 and  (1989b), “Equity and Disconfirmation Perceptions
as Influences on Merchant and Product Satisfaction, Journal of Con
-
sumer Research, 16 (December), 372-83.
Perdue, Barbara C. and John O. Summers (1986), “Checking the Success
of Manipulations in Marketing Experiments,” Journal of Marketing
Research, 23 (November), 317-26.
Reicheld, Frederick F. and W. Earl Sasser, Jr. (1990), “Zero Defections:
Quality Comes to Services,” Harvard Business Review, 68 (Septem
-
ber-October), 105-11.
136 JOURNAL OF SERVICE RESEARCH / November 2000
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from

Richins, Marsha L. (1983), “Negative Word-of-Mouth by Dissatisfied
Consumers: A Pilot Study,” Journal of Marketing, 47 (Winter),
68-78.
 (1987), “A Multivariate Analysis of Responses to Dissatisfac
-
tion,” Journal of the Academy of Marketing Science, 15 (Fall), 24-31.
Roos, Inger (1999), “Switching Processes in Customer Relationships,”
Journal of Service Research, 1 (August), 68-85.
Ross, William T., Jr. and Elizabeth H. Creyer (1993), “Interpreting Inter
-
actions: Raw Means or Residual Means?” Journal of Consumer Re
-
search, 20 (September), 330-38.
Singh, Jagdip (1990), “Voice, Exit, and Negative Word-of-Mouth Behav-
iors: An Investigation Across Three Service Categories,” Journal of
the Academy of Marketing Science, 18 (Winter), 1-15.
 and Robert Widing II (1991), “What Occurs Once Consumers
Complain? A Theoretical Model for Understanding Satisfaction/
Dissatisfaction Outcomes of Complaint Responses,” European Jour-
nal of Marketing, 25 (5), 30-46.
Smart, Denise T. and Charles L. Martin (1993), “Consumers Who Corre-
spond with Business: A Profile and Measure of Satisfaction with Re-
sponse,” Journal of Applied Business Research, 9 (Spring), 30-42.
Smith, Amy K. and Ruth N. Bolton (1998), “An Experimental Investiga-
tion of Customer Reactions to Service Failure and Recovery Encoun-
ter: Paradox or Peril?”Journal of Service Research, 1 (August),
65-81.
Stauss, Bernd and Christian Friege (1999), “Regaining Service Cus-
tomers,” Journal of Service Research, 1 (May), 347-61.
Sundaram, D. S., Claudia Jurowski, and Cynthia Webster (1997), “Ser-
vice Failure Recovery Efforts in Restaurant Dining: The Roles of
Criticality of Service Consumption,” Hospitality Research Journal,
20 (3), 137-49.
Swan, John E. and Frederick I. Trawick (1981), “Disconfirmation of Ex-
pectations and Satisfaction with a Retail Service,” Journal of Re-
tailing, 57 (Fall), 49-67.
TARP (1979), Consumer Complaint Handling in America: Summary of
Findings and Recommendations. Washington, DC: U.S. Office of
Consumer Affairs.
 (1986), Consumer Complaint Handling in America: An Updated
Study Part II. Washington, DC: U.S. Office of Consumer Affairs.
Tax, Stephen S. and Stephen W. Brown (1998), “Recovering and
Learning from Service Failure,” Sloan Management Review,39
(Fall), 75-88.
, , and Murali Chandrashekaran (1998), “Customer Eval-
uations of Service Complaint Experiences: Implications for Rela-
tionship Marketing,” Journal of Marketing, 62 (April), 60-76.
Triplett, Tim (1994), “Product Recall Spurs Company to Improve Cus-
tomer Satisfaction,” Marketing News, 28 (April 11), 6.
Tucker, Ledyard R. and Charles Lewis (1973), “A Reliability Coefficient
for Maximum Likelihood Factor Analysis,” Psychometrika, 38 (1),
1-10.
Webster, Cynthia and D. S. Sundaram (1998), “Service Consumption
Criticality in Failure Recovery,” Journal of Business Research, 41,
153-59.
Westbrook, Robert A. (1980), “Intrapersonal Affective Influences on
Consumer Satisfaction with Products,” Journal of Consumer Re
-
search, 7 (June), 49-54.
 (1981), “Sources of Consumer Satisfaction with Retail Outlets,”
Journal of Retailing, 57 (Fall), 68-85.
Yi, Youjae (1990), “A Critical Review of Consumer Satisfaction,” in Re-
view of Marketing, Valarie A. Zeithaml, ed. Chicago: American Mar-
keting Association, 68-123.
Zeithaml, Valarie A., Leonard L. Berry, and A. Parasuraman (1996),
“The Behavioral Consequences of Service Quality,” Journal of Mar
-
keting, 60 (April), 31-46.
Michael A. McCollough is an assistant professor of marketing at
the University of Idaho. He obtained his Ph.D. in marketing from
Texas A&M University. His research interests are services mar-
keting, services quality, service failure and recovery, service
guarantees, and customer satisfaction and dissatisfaction. He has
published in the Journal of Hospitality and Tourism Research,
the Journal of Marketing Education, Marketing Education Re-
view, the Journal of Professional Services Marketing, and Inter-
national Review of Retail, Distribution and Consumer Research.
Leonard L. Berry holds the JCPenney Chair of Retailing Studies
and is a distinguished professor of marketing at Texas A&M Uni-
versity. He is a former national president of the American Mar-
keting Association and founder of Texas A&M University’s
Center for Retailing Studies. His research interests are services
marketing, service quality, and retailing strategy. He has pub-
lished numerous journal articles and books, including Dis-
covering the Soul of Service—The Nine Drivers of Sustainable
Business Success (Free Press, 1999), On Great Service: A
Framework for Action (Free Press, 1995), Marketing Services:
Competing through Quality (Free Press, 1991), and Delivering
Quality Service: Balancing Customer Perceptions and Expecta-
tions (Free Press, 1990).
Manjit S. Yadav is an associate professor of marketing and Cen-
ter for Retailing Studies Faculty Fellow, Texas A&M University.
His current research focuses on strategic issues related to elec-
tronic commerce. His work has been published in a number of
leading journals, including the Journal of Marketing Research,
the Journal of Consumer Research, and Sloan Management Re-
view. He is a member of the Editorial Review Board of the Jour-
nal of the Academy of Marketing Science. He developed and
currently teaches a graduate course (Strategic Foundations of
E-Commerce) and an undergraduate course (Fundamentals of
Internet Marketing) dealing with the strategic challenges and op-
portunities in the emerging electronic marketplace.
McCollough et al. / CUSTOMER SATISFACTION 137
at SAGE Publications on December 2, 2009 http://jsr.sagepub.comDownloaded from