You are on page 1of 20
Customer Satisfaction: A Meta-Analysis of the Empirical Evidence David M. Szymanski Texas A&M University David H. Henard North Carolina State University The growing number of academic studies on customer sat isfaction and the mixed findings they report complicate ef- forts among managers and academics to identify the ‘antecedents to, and outcomes of, businesses having more- versus less-satisfied customers. These mixed findings and the growing emphasis by managers on having satisfied customers point to the value of empirically synthesizing the evidence on customer satisfaction to assess current knowledge. To this end, the authors conduct ameta-analysis of the reported findings on customer satisfaction. They document that equity and disconfirmation are most strongly related to customer satisfaction on average. They also find that measurement and method factors that char- acterize the research often moderate relationship strength between satisfaction and its antecedents and outcomes. The authors discuss the implications surrounding these ef fects and offer several directions for future research. Our focus is customer satisfaction. Gulfstream Aeronautics Our customers will be totaly satisfied withthe prod- ucts, services and technology we supply. —Shell Chemical Company Satisfaction Guaranteed. —Wal-Mart Stores, Inc. Journal ofthe Academy of Marketing Science. ‘Volume 29, No.1 pages 16-35, (Copyright © 2001 by Academy af Marketing Sclnce. As these business slogans make apparent, customer sat- faction has come to represent an important cornerstone for customer-oriented business practices across a multi- tude of companies operating in diverse industries. This emphasis on businesses’ having satisfied customers fur- ther serves to accentuate the potential value resulting from an empirical synthesis of the documented findings on the antecedents and consequences of customer satisfaction ‘These findings vary considerably in terms of statistical sig nificance, direction, or magnitude. Consequently, a ‘meta-analysis ofthe evidence can advance managers’ and academic researchers’ understanding of customer satis tion principles by documenting the statistical significance, direction, and magnitude of the effects that can be expected ‘on average. A meta-analysis can also document the degree to which the variance in effect sizes is real versus artifactual and can further identify moderating variables that account for the variance in the satisfaction relationships. (Our objective in this study is to advance understanding bby conducting a meta-analysis of the satisfaction findings and discussing the results, To accomplish this objective, ‘we proceed with an overview of the rationale behind the ‘general antecedents, outcomes, and potential moderators of customer satisfaction. We then describe the methodol- ‘ogy used for identifying the population of empirical stud- ies on customer satisfaction. The findings from ourstaisti- cal analysis of 50 empirical studies’ reporting 517 correlations involving customer satisfaction and related elements are presented immediately thereafter, We con- clude the meta-analysis by discussing implications of the findings, limitations ofthe study, and directions for future research. ANTECEDENTS, CONSEQUENCES, AND MODERATORS OF CUSTOMER SATISFACTION ‘The framework guiding our theoretical discussion and ‘empirical investigation focuses on the relationships from Oliver's (1997) conceptual model of customer satisfaction that have been empirically examined in the literature.’ ‘Although subsequent discussions center on anticipating the sign of the association that will emerge on average, of ‘more substantive foci in the meta-analysis are the estima- tion of the magnitude of the respective relationships and the identification of factors accounting for the variance in reported effect sizes. Antecedents to Satisfaction Research on customer satisfaction has focused predom- {nantly on modeling the effects of the following factors on buyers’ level of satisfaction: expectations, disconfirmation of expectations, performance, affect, and equity. Each of these effects is discussed in tur, Expectations. The role of expectations in satisfaction levels has traditionally been modeled in one of two ways. One isthe role of expectations as anticipation, which we will discuss here. The other isthe role of expectations as comparative referents, which we will discus later inthe context of disconfirmation effects. With regard to expecta- ‘ions as anticipation, the thinking is that consumers’ ex- pectations have a direct influence on satisfaction levels. This direct influence is inthe absence of any assessment of, or comparison to, actual outcomes or performance lev- els of the stimulus (LaTour and Peat 1979; Oliver and DeSarbo 1988). Rather, consumers are thought to adapt to 1 certain level of performance. They form expectations consistent with these performance levels, and these expec- tations serve as the baseline for satisfaction assessments (Oliver 1981, 1993). Consumers are thought to assimilate satisfaction levels to expectation Tevels in order to avoid the dissonance that would arise when expectations and sat- {sfaction levels diverge. This assimilation effect results in satisfaction judgments’ being higivlow when expectations are high/low (Oliver 1997), andthe majority of the empiri- cal findings support a positive relationship between expec- tations and satisfaction (c.g., Bearden and Teel 1983; Oliver and Linda 1981; Swan and Trawick 1981) Disconfirmation of expectations. Besides support for ‘expectations as anticipation, there is support for expecta- tions as comparative referents. Subsumed under the disconfirmation paradigm, this perspective to satisfaction judgments is a central focus in early studies on satistac- tion. For example, well over half of the correlations (64%) (on disconfirmation effects are from studies published be- fore 1984 (e.g., Bearden and Teel 1983; LaBarbera and ‘Seymanski, Menard CUSTOMER SATISFACTION 17 Mazursky 1983; Oliver 1980). In these studies, we find the first conceptualizations of expectations asthe stan- dard against which performance outcomes are assessed. Here consumers are said to be satisfied when actual out: comes exceed expectations (positive disconfirmation), dissatisfied when expectations exceed outcomes (negative disconfirmation), and just satisfied (zero or simple disconfirmation) when outcomes match expectations (Oliver 1981; Oliver and DeSarbo 1988). Hence, disconfirmation and satisfaction are thought to be posi- tively corelated and we expect ths relationship tobe evi- ddenced in the meta-analysis. Performance. In addition to performance as a compo- nent of disconfirmation, performance has been modeled as directly affecting satisfaction (e.g., Churchill and Surprenant 1982; Halstead, Hartman, and Schmidt 1994; Oliver and DeSarbo 1988; Tse and Wilton 1988) Modeling performance as a separate predictor of satisfac- tion follows directly from the notion of a value-percept di- versity; that is, customers are likely to be more satisfied. with the offering as the ability of the offering to provide consumers what they need, want, or desire increases rela- tive to the costs incurred (lohnson 1998), Because this positive relationship between performance and satisfac- tion is documented frequently in the literature (exceptions include Gilly and Gelb 1982; Swan and Oliver 1991; ‘Westbrook 1981), we anticipate that performance and sat isfaction will be positively correlated on average in the cu- rmulative findings. Affect. The possiblity that satisfaction isnot just cog tive but includes an affective component has also been a focus of research attention. Studies in this gente (e2., Mano and Oliver 1993; Westhrook 1987; Westbrook and Oliver 1991) explicate the dimensionality of affect and ex- amine the role of affect in satisfaction judgments. They find that affect is two-dimensional with overall affec’s having an impact on satisfaction levels above and beyond classical expectancy-disconfirmation effects, for example. ‘These effects ae often discussed in the context of affec- tive-processing mechanisms. That is, emotions elicited during consumption ae proposed to leave affective traces in memory, traces that are availabe for consumers 0 3e- ‘cess and integrate into their satisfaction assessments (Westbrook and Oliver 1991). A second explanation grounded in attribution theory, specifically, Weiner’s (1986) locus-stability-controllablility matrix, suggests that affect canbe attribution dependent (Oliver 1993; Oli ver and DeSarbo 1988). Attributions can evoke specific af- fect depending on whether the outcome of the consumption experience isa successor failure, and there- fore, affect is presented as another component of postpurchase expression that feeds positively into satisfac tion assessments. These theories imply @ positive relation- ship between affect and satisfaction, and a positive 18 JOURNAL OF THE ACADEMY OF MARKETING SCIENCE relationship between these two factors is widely supported inthe literature, Our corresponding expectation, therefore, is a positive relationship between affect and satisfaction in the aggregate data, Equity. In addition to expectations, disconfirmation, performance, and affect, satisfaction has been modeled as the direct outcome of equity. Equity isa fairness, rightness, or deservingness judgment that consumers make in refer ence to what others receive (Oliver 1997:194). Con- sumers’ calculation of equity implicitly if not explicitly assumes the following form: o where O is outcomes, 1s inputs, cis the consumer, ris the referent person or group, and «is a proportional operator. Based in theory on distributive justice (individuals expect to get what they deserve based on their inputs [e.g., Oliver 1993; Oliver and Swan 1989a, 1989b)), procedural justice (he relative mannerin which the outcomes were delivered [e.g Swan and Oliver 1991), and interactional justice (he relative manner in which the consumer is treated in terms of respect, politeness, and dignity [e.g., Clemmer 1988), consumers are presented as being satisfied (posi- tive inequity) when their equity ratio is proportionately ‘greater than the ratio achieved by the referent person or group (see Goodwin and Ross 1992; Oliver 1997). The positive relationship between equity and satisfaction that is typically supported in the literature (e.g, Clemmer 1988; Oliver 1993; Oliver and Swan 1989a, 1989b, Swan and Oliver 1991) leads us to expect a positive relationship ‘on average in the meta-analysis. Relationships among predictors of satisfaction. An as- sessment of the nomological network among the predic~ {ors of customer satisfaction suggests that several pais of antecedents may be related (see Figure 1). For one, it is possible that performance and expectations are positively related. Johnson (1998) describes this relationship both in terms of expectations’ or past performance information's predicting current performance levels and strong expecta- tions’ affecting perceptions of performance. Hence, we anticipate expectations" having a positive effect on percep: tions of performance on average. ‘Second, arguments can be advanced for performance and expectations being separately related to disconfirma- tion (see Figure 1). However, the net direction of these effects is difficult to predict. On one hand, improved per- formance can have a positive effect on disconfirmation when expectations remain constant, and higher expecta tions can have adetrimental affect on disconfirmation when performance remains constant. On the other hand, improv. ed performance does not guarantee positive disconfirm. WINTER 2001 FIGURE 1 Model of the Antecedents and Consequences of Customer Satisfaction NOTE: WOM = word of mouth ation when expectations rise proportionately or more than proportionately to performance. Likewise, higher expec- {ations alone do not guarantee negative disconfirmation ‘when performance increases proportionately or more than proportionately compared with expectations. These multi pple outcomes on disconfirmation that are possible and that ccan be associated with a change in performance and/or expectations imply that resolving the directionality issue is better relegated to the empirical portion of the meta-analysis. Another determination that is better relegated to the empirical portion of the meta-analysis is the direct effect of performance on consumer equity. On one hand, perfor- mance and equity may be positively related. This would ‘occur when the consumer's outcomes improve in the face Of inputs’ staying the same, declining, or increasing less than proportionately to outcomes, all else being constant. ‘On the other hand, higher performance by itself unlikely to guarantee that equity perceptions will improve. Equity ‘can remain the same or decline inthe face of improved per- formance when the consumer's inputs increase in propor- tion to, oF more than proportionately to, their outcomes. Equity also can remain the same or decline when (1) the referent party's outcomes increase in proportion to, or ‘more than proportionately o, the consumer's outcomes; or (2) the other party's inputs decrease to the extent that the party's equity ratio is now equal to, or higher than, the con- sumer's equity ratio. In other words, multiple states of equity (positive, negative, ora steady state) can be associ- ated with a change in performance. Identifying the state that emerges on average is one of the objectives in the meta-analysis. Consequences of Customer Satisfaction Few studies have investigated the outcomes of satisfac tion and only a few outcomes of satisfaction have been investigated in these studies. These outcomes are com- plaining behavior, negative word of mouth (WOM), and repurchase intentions. Complaining behavior. Consumers’ tendency to com- plain to sellers has been discussed in the literature as one ‘mechanism available to consumers for relieving cognitive dissonance when the consumption experience is dissatis- {ying (Oliver 1987). Complaining has also been discussed fas a mechanism for venting anger and frustration and a ‘mechanism for initiating or secking redress for failed con- sumption experiences (Nyer 1999). Although consumers have the option of voting with their feet (exiting) or re- ‘maining loyal (staying) in the face of adissatisfying expe- rience (Andreasen 1988; Day 1984; Hirschman 1970), greater dissatisfaction is traditionally thought to prompt complaining, especially when the problem leading to dis- satisfaction is severe, the degree of external attribution of blame isto the retailer or manufacturer, or the likelihood of redress is relatively high (e.g., Folkes 1984; Richins 1983: Ursic 1985). In other words, given proper ability (e.g. ‘channel knowledge, access and communication skills) and motivation (¢-., cultural norms, willingness to confront) to complain (“behavioral model of complaining”), as well asa favorable alignment of perceived costs (cg., time, ef- fort, and monetary importance), benefits (¢-g., money back, replacement) and assessments of success (€.g., firm's reputation, threat to business) in regard to complain- ing (‘economic model of complaining” (see Oliver 1997)), the expectation is that greater dissatisfaction would be ‘manifested in more complaining to sellers. Negative WOM behavior. Negative WOM behavior to ‘other consumers represents yet another form of complain- ing behavior that is expected to increase in the face of adis- satisfying experience. This effectis especially likely when the product or service failure is severe, attributions for the failure are external, orhigh levels of social activity charac lerize the disappointed consumer (Folkes 1984; Richins 1983). Negative WOM offers consumers a mechanism for releasing tension, getting back at the entity by informing others of disappointing encounters, regaining control over distressing situation, gaining sympathy from others, and ‘conveying to others that the consumer has high standards (Nyer 1999). These motivations for telling potential buy- cers about a particularly dissatisfying experience, in turn, suggest that negative WOM and satisfaction would be in- versely related on average. Repeat purchasing. Satisfaction is further thought to af- fect the likelihood that consumers will buy the offering Szymanski, Henard/ CUSTOMER SATISFACTION. 19 again. Oliver (1997), for example, discusses loyalty as an outcome of customer satisfaction. He proposes three phases of satisfaction—cognitive, affective, and conative— that culminate in action loyalty (operationalized as repeat usage). Tis positive relationship between satisfaction and repeat purchasing is evidenced in the extant data (©. Bearden and Teel 1983), and we anticipate the same rela: tionship in the meta-analysis. Relationships between the outcomes of satisfaction. In addition to the direct effects of satisfaction on complain- ing, negative word of mouth, and repeat purchasing, rela- tionships can exist between the outcomes of customer satisfaction (see Figure 1). In this regard, both complain ing and negative WOM activities have been discussed for theireffects on repeat purchasing. Specifically, an increase in complaining or negative WOM behaviors is viewed as, hhaving a detrimental effect onthe likelihood of repatronage. ‘This train of thought is grounded in self-perception the- ory—public disclosure of a position increases commit: ment to that position—and dissonance theory, which predicts that consumers strive for actions that are consis tent with cognitions (Tax and Chandrashekaran 1992). ‘Those theoretical perspectives imply that consumers strive for consistency across voiced feelings, held emotions, and purchasing actions. As a consequence, more complaining to sellers and other consumers (which represents an ex- plicit, negative position toward an offering by the con- sumer is likely to reduce repatronage (which represents a negative consumer behavior toward the offering). This negative relationship should be evidenced when the data are aggregated across studies. Potential Moderators of Customer Satisfaction As documented later in the meta-analysis, there often is wide variation in the magnitude of the correlations reported for the same correlate with satisfaction. One objective of the meta-analysis, therefore, isto identify the moderators of satisfaction effects. These moderators include the comparison standard, measurement level, methodology, subject population, and type of offering. ‘Comparison standard. In formulating a level of satis- faction, consumers might arrive at different conclusions depending on the reference being used. For example, com- parison-level theory (LaTour and Peat 1979; Thibaut and Kelly 1959) suggests that satisfaction is an additive func- tion of both experience-based disconfirmation of attribute levels obtained from a brand and the corresponding com- parison levels of that brand. Satisfaction modeled as the discrepancy between outcomes and some standard of ex- pectation implies that overall satisfaction may differ when Gifferent references are being used across studies. This could happen when the expectancy-based (c.g. expecta 20 JOURNAL OF THE ACADEMY OF MARKETING SCIENCE tions formed through vicarious learning) or experi- ence-based (actual performance) norms are used as references in the satisfaction studies (see Oliver 1997; Yi 1990). In fact, the use of different references across studies implies that satisfaction estimates would differ when con- ssumers’ expected and experience-based norms are not the same, Moreover, because itis likely that they will not al- ‘ways be the same, we anticipate that the comparison stan- dard will emerge as a significant moderator of satisfaction effects. ‘Measurement level. By measurement level, we refer to ‘whether satisfaction is captured through an aggregate (sin- le-item) or attribute (multi-item) level of measurement (ee Yi 1990). To differentiate, an attribute level of mea- surement attempts to first capture buyers’ satisfaction to- ‘ward a specific aspect or dimension of the offering and then aggregates the assessments into an overall satsfac- tion score. In contrast, an aggregate measurement would inquire only about a buyer’s overall or global satisfaction with a product or service encounter. ‘The aggregate measurement subsumes the attribute ‘measurements. One presumption, therefore, is that the two assessments—aggregate and individual atribute—would ‘yield similar estimates of overall satisfaction, On the other hhand, studies have begun to address causal relationships between the two measures (see Mittal, Ross, and Baldasare 1998). This research implies that using one ‘measure in lieu of the other would lead to different esti- ‘mates of association when the aggregate and attribute assessments are not perfectly correlated. Moreover, itis conceivable that the two measures would diverge and that the aggregate measure may be amore accurate measure of customer satisfaction. For example, the estimates would diverge when consumers engage in partial information processing and partial satisfaction assessments (.c., satis- faction assessments based on evaluations of only some of the features) or when consumers disproportionately weight the attributes incorporated into satisfaction judg~ ‘ments (Oliver 1997). An aggregate assessment would allow consumers to impose their weighting criteria on the ‘elements before responding with an overall assessment of satisfaction. Adding or averaging item scores, as is typi- cally done in an attribute-by-attribute approach, would preclude the capturing of consumers’ disproportionate ‘weighting schemes and could incorporate factors into the satisfaction judgment that consumers would not naturally consider when making satisfaction assessments (¢.2.. selected factors that are included in the survey instrument by the researcher). This difference in what is or can be cap {ured when using aggregate versus attribute measures sug- gests that the measurement level could be an important explanator of the differences in satisfaction effects reported in the literature, ‘WINTER 2001 ‘Methodological approach, An additional element that ‘could account for the variance in the magnitude of the ef- fect evidenced in the literature is whether researchers use ‘an experimental or survey approach. Each approach has characteristics that could contribute to the variance ob- served across estimates of relationship strength. Experi- ‘ments, for example, can control the levels of the factor to which a participant is exposed and can offer the control necessary to eliminate potential confounds. However, ex- periments compromise realism when they use fictitious stimuli under artificial consumption conditions. A survey approach, meanwhile, may offer less control over the as- signment of subjects to the levels of a fuctor but may be ‘mote realistic because it is based on real offerings under natural consumption conditions (Tabachnick and Fidell 1996). The degree to which the differences that character- ize surveys and experiments bias estimates of association ‘will be examined explicitly inthe meta-analysis. Participants. Researchers have raised concems about the generalizability of student-hased findings across the ‘consumer population (Burnett and Dunne 1986; Park and Lessig 1977). Students” restricted age range, limited con- sumption experiences, and relatively low income have re- sulted in students” being portrayed as atypical consumers. Students have also been portrayed as having yet-to-be- solidified cognitive structures that make them more sus- ceptible to reference group influences (Park and Lessig 1977). More important, these distinctions could translate intodifferences.in how the two groups of consumers assess satisfaction or how they react to satistying and dissatisty- ing experiences. For example, having less-defined cogni- tive structures and being more outward focused (i.e, more susceptible to reference group influences) could mean that the more cognitive and inward-focused factors such as ex- ppectations,disconfirmation, and affect play less of arolein student participants’ satisfaction assessments. Satisfac- tion assessments might also play a lesser role in students” purchasing intentions if students are influenced more by their peers. Furthermore, the comparisons to others im- plied by the reported findings on reference group influ- ences suggest that equity assessments—the comparison of outputs and inputs to other consumers or groups of con- sumers—could play a heightened role in the satisfaction assessments of student participants relative to nonstudent participants. These possibilities suggest that the types of participants could account for some variance in the satis- faction correlations. Type of offering. While early research on consumer sat- isfaction sometimes contrasts the processes and outcomes associated with evaluating durable versus nondurable goods (e.g., Churchill and Surprenant 1982), more recent perspectives have focused on the distinction between ser- vices and products. Services are distinguished from prod- ucts in at least four ways: perishability, tangibility, separability in production and consumption, and standard- ization (Parasuraman, Zeithaml, and Berry 1985). Individ- ually and collectively, these inherent differences are thought to have an impact on how satisfaction is assessed ‘and what the final assessments look like. Johnson (1998), for example, proposes that expectation effects on satisfac- tion are generally weaker in a service context because the intangible nature of services makes information on expec- tations less concrete and less useful. Halstead etal (1994) find that consumers’ satisfaction formation processes are distinct for services compared with products. They refer- ‘ence the services literature for theoretical support for their findings, This literature argues that evaluations for services are more difficult than evaluations for products because service evaluations are based on different expec- tations and grounded in processes as well as outcomes (Gronroos 1982; Zeithaml, Berry, and Parasuraman 1993), Although a separate focus on satisfaction effects for ser- vices versus products is only now gaining momentum, the product-service contexts found in early studies will be captured in the meta-analysis. METHODOLOGY To assess the validity of the satisfaction effects pro. posed here, steps were taken to first develop a database of the satisfaction findings. The first step in this process was the delineation of the criterion for including studies in the review. Candidates for inclusion were empirical studies that specified customer satisfaction as a measured variable in their empirical models. These studies were identified through keyword searches of electronic databases (ABVInform, WILS, UML, among others) using customer- consumer satisfaction and buyer satisfaction as identifying terms, searches of the references found in the available studies, and manual searches of leading academic journals in which studies on customer satisfaction would most likely be published, namely, the Journal of the Academy of Marketing Science; the Journal of Consumer Research; the Journal of Customer Satisfaction, Dissatisfaction and Complaining Behavior, the Journal of Marketing; the Journal of Marketing Research; Management Science, and Marketing Science. In addition, we requested working papers on satisfaction from authors of previous satisfac: tion studies and asked them for leads on working papers by other authors. The search process was terminated in July 1998, when it became clear that further efforts were not yielding additional studies. After gathering the studies on customer satisfaction, ‘our attention turned toward identifying the measure of association (correlation, regression coefficient, etc.) that ‘would permit the greatest number of effects to be included Stymanski, Hen / CUSTOMER SATISFACTION 21 in the meta-analysis, that is, the correlation coefficient. ‘The correlation coefficient is used most often in the litera- ture to report satisfaction relationships, itis the metric to which many satisfaction findings can be converted (see Glass, McGaw, and Smith 1981), and the correlation pre- serves the continuous properties in the satisfaction mea- sure and its correlates.? ‘Not all the empirical studies, however, reported corre- lations or measures that could be converted to correlations. ‘We therefore asked the authors of satisfaction studies not reporting correlations for correlational data, In a few instances, the authors were able to provide the correlations wwe requested so that in the end, data from 50 of 85 empiri- cal studies on customer satisfaction could be included in the meta-analysis (see studies with an asterisk in the Ref- ‘erences section for included studies). The $0 studies con- sist of 44 published studies and 6 dissertations reporting a total of $17 correlations involving satisfaction or satisfac- tion-related variables. The correlation values and the methods and measures associated with each correlation ‘were coded into the database. Two individuals independ- ently coded all the data. Coding consistency was achieved in 96% of the instances, and the few discrepancies that ‘occurred were resolved through discussions in reference tothe coding scheme, META-ANALYTIC FINDINGS ‘The analysis of the data and the reporting of findings proceed in three phases. First, we describe the correlations in terms of range, direction, statistical significance, and sample size. These data accentuate the nature and diversity of the findings on customer satisfaction. Second, we pre- sent the findings from the univariate analysis ofthe corre- lations. The purpose here is to offer insights into the con- tral tendencies of the individual correlates with customer satisfaction. Multivariate analysis of the correlations is also presented to offer additional evidence in support of bivariate findings. The final stage of data analysis centers con identifying the statistically significant moderators of the satisfaction effects. Descriptive Analysis ‘The data in Table 1 make apparent the diversity in the satisfaction effects reported in the literature. For one, the data reveal that the range of the reported values ean be quite broad for certain correlates with satisfaction. AS ‘examples, the correlations for disconfirmation with satis- faction range from 24 to .87, the values of the correla- tions for performance with satisfaction range from -37 to 81, and the values ofthe correlations for expectations with disconfirmation range from ~36 to .34. In addition, the 50" oye ye ogra Sens 2a WOW ont vN OWN va ener 1 radar onary Suture WOW op 4 YN ow WN OWN oor wo t Aupepsnd reed pau Jo pro ae, Wa cre ers wove eae cl 6 used yeday-oreIES £ WWW YN WNW, w-mg- gs now yo pow aneu-voe , WN OWN, YR WN 8 OO £ soe Suu doo pou yo ap 03190 t 1 wz uy sm 9 souranod-suonedg WN. 1 * 90S z ‘no soem pg WN. 1 mz ose oor L ‘soumuguoaspsouRauopng 9 6 5 srr woe ee L € ws eee a s L + ure u a a € osos8 st fa t n sures re of t 1 a aes wag- ot a “onasay—aquny slong —oqumy pion —aquyy —pSouaaIay —aquRN NR —samyA—oNaUE —_-FapMIS FReD sanojmuny fo a8ury —fosoqany fo srquny mor uaa “nonaua ood oyu nosey ‘ume onan i0g puosyudisuny alls aumsag usa ‘SUONEIeO9 PeleleL-UORDEISHeS oy) UO UONEULIO}U aAnd}OseG bane. correlations reported for the same correlate with satisfac- tion often contain positive and negative correlations as well as correlations that are statistically significant in the face of other correlations that are not significant. Although ‘most ofthe correlations have signs that are consistent with overriding expectations, we do find instances where the disparity in direction and statistical significance is notable. For example, for the expectations-disconfirmation link, ‘we find that 26 percent ofthe correlations are positive and statistically significant, another 26 percent of the correla- tions are negative and statistically significant, and the remaining 48 percent of the correlations are not statist cally significant in either direction, ‘The data in Table I further document that far less atten- tion in the empirical literature has been devoted to under- standing the outcomes of customer satisfaction. Only 29 (5.6%) of the correlations in our database pertain to a vari- able that is a consequence of customer satisfaction. In con- trast, 488 (94.4%) correlations pertain to an antecedent of customer satisfaction, Moreover, among the antecedents,