The Determinants of Satisfaction: an Experimental Verification of the Moderating Role of Ambiguity

Prashanth U. Nyer, Chapman University
ABSTRACT - This study extends the findings of Yi (1993) in which he finds that performance ambiguity plays a moderating role in the way consumer satisfaction is determined. This study investigates the role of both performance ambiguity and the ambiguity of expectations in the consumer satisfaction process using an experiment. When perceived performance is ambiguous, the effect of expectations on satisfaction is increased, while the effect of perceived performance on satisfaction is decreased. When expectations are ambiguous, the effect of expectations on satisfaction decreases while the effect of performance on satisfaction increases. The implications of these findings are discussed.
[ to cite ]:
Prashanth U. Nyer (1996) ,"The Determinants of Satisfaction: an Experimental Verification of the Moderating Role of Ambiguity", in NA - Advances in Consumer Research Volume 23, eds. Kim P. Corfman and John G. Lynch Jr., Provo, UT : Association for Consumer Research, Pages: 255-259.

Advances in Consumer Research Volume 23, 1996      Pages 255-259

THE DETERMINANTS OF SATISFACTION: AN EXPERIMENTAL VERIFICATION OF THE MODERATING ROLE OF AMBIGUITY

Prashanth U. Nyer, Chapman University

ABSTRACT -

This study extends the findings of Yi (1993) in which he finds that performance ambiguity plays a moderating role in the way consumer satisfaction is determined. This study investigates the role of both performance ambiguity and the ambiguity of expectations in the consumer satisfaction process using an experiment. When perceived performance is ambiguous, the effect of expectations on satisfaction is increased, while the effect of perceived performance on satisfaction is decreased. When expectations are ambiguous, the effect of expectations on satisfaction decreases while the effect of performance on satisfaction increases. The implications of these findings are discussed.

The importance of consumer satisfaction (CS) as a subject of research lies in it's ability to influence various consumer phenomena such as repurchase, brand loyalty, word-of-mouth and complaint behavior. The expectancy-disconfirmation model of consumer satisfaction (Oliver, 1980) has received much attention in the past decade and a half. According to this model CS is determined by prior expectation, perceived performance and by the disconfirmation of expectation - the subjective difference between perceived performance and expectation.

Various studies have investigated the robustness of the model under differing conditions. The direct effect of perceived performance on CS has been confirmed by many studies including those by Churchill and Suprenant (1982), Oliver and DeSarbo (1988) and Tse and Wilton (1988). According to Tse and Wilton (1988), perceived performance was the single most important determinant of CS. Perceived performance was also shown to have an indirect effect on CS by influencing disconfirmation. However, studies conducted by Cadotte, Woodruff and Jenkins (1987) and Oliver (1980) do not report a significant effect of perceived performance on CS.

The effect of disconfirmation on CS has also been verified by many studies. Oliver and DeSarbo (1988) found disconfirmation to be the most significant predictor of CS. What is of more relevance to this paper is the effect of expectation on CS. While some studies have found a significant direct effect of expectation on CS (Bearden and Teel 1983; Swan and Trawick 1981, Tse and Wilton 1988, Westbrook and Reilly, 1983), others have not reported a direct effect of expectation on CS (Cadotte, Woodruff and Jenkins 1987, Oliver and Bearden 1983).

Clearly the effects of expectation and perceived performance on CS is not constant under all circumstances. Under certain conditions expectation fails to have a significant direct effect on CS, while under other circumstances perceived performance may have no significant direct effect on CS. These findings point to the moderating influence of an external variable.

Yi (1993) suggests that the ambiguity with which product performance is evaluated is capable of moderating the effects of both expectation and perceived performance on CS. Consumers may be unable to unambiguously evaluate the performance of some products. For example a light bulb may be advertised as having an extra long life of 3000 hours, causing consumer expectation to be formed with little ambiguity. On the other hand very few consumers bother to measure the actual life of their light bulbs, leading to a great deal of ambiguity as far as the product performance is concerned. In such a situation, expectation will play a bigger role in determining CS while perceived performance will take on a diminished role. Performance ambiguity can be high when the performance is being judged mostly on subjective criteria (e.g., most fashion products, music, or art) or when there is difficulty in measuring the performance as in the example above. Yi (1993) draws upon the research by Hoch and Deighton (1989) and Hoch and Ha (1986) and upon the self perception theory to find theoretical support for such a phenomenon.

As discussed earlier, the expectancy disconfirmation model of CS has two primary antecedents for satisfaction - prior expectation and perceived performance. The third predictor of satisfaction - disconfirmation of expectation - is the subjective difference between perceived performance and expectation. Since Yi (1993) has shown that performance ambiguity moderates the effect of expectation and performance on CS, the next step then is to investigate whether expectation ambiguity has a similar moderating effect on the role of expectation and performance on CS.

The expectancy disconfirmation model of CS can be seen as a form of information integration, where two sets of information - prior expectations and perceived performance - are integrated to form satisfaction judgments. The two information integration models that have received much attention are the adding model and the averaging model (Anderson 1981). The averaging model posits that an attribute's importance or weight will change as the importance of other attributes change. With the adding model, the weights of attributes are independent of each other. Various studies including those by Anderson and Lopes (1974) and Birnbaum and Stegner (1979) have tested these two models and found support for the averaging model. Under an averaging model scenario, it is easy to see how ambiguous information (information capable of multiple interpretations) is weighted less, leading to a greater emphasis on unambiguous information.

Consumers may have difficulty forming unambiguous prior expectations under many circumstances. Lack of information, complex information or inability to understand the information, are all factors that could lead to high expectation ambiguity. Thus products using new and complex technology may present a challenge to the naive consumer trying to form expectations. Thus when expectations are ambiguous, it could be hypothesized that the role of expectations in determining CS will be diminished while the effect of perceived performance on CS will be enhanced.

Since disconfirmation is a function of expectation and perceived performance, and since the focus of this paper is on these two variables, this paper's initial focus is on the total effects of expectation and perceived performance on CS. The total effect of expectation on CS would include not only the direct effect of expectation on CS, but also the indirect effect of expectation on CS through disconfirmation. While the first set of analyses examine the total effects of expectation and perceived performance on CS, the second set of analyses (using a multiplicative regression model) examine the direct effects of expectation, perceived performance and disconfirmation on CS.

The following hypotheses are proposed:

H1. When performance is ambiguous and when expectation is unambiguous, expectation will have a stronger total effect and performance will have a weaker total effect on CS, compared to the situation in which both expectation and performance are unambiguous.

H2. When performance is unambiguous and when expectation is ambiguous, expectation will have a weaker total effect and performance will have a stronger total effect on CS, compared to the situation in which both expectation and performance are unambiguous.

METHOD

A full factorial experiment was designed in which two factors, performance ambiguity (high and low) and expectation ambiguity (high and low) were manipulated. Subjects consisted of 132 undergraduate students enrolled in an introductory marketing course at a large mid-western university. The subjects were randomly assigned to the four experimental conditions. Subjects were informed that they were about to evaluate one of the many test formulations of a new stain remover, and that other subjects would be evaluating different formulations.

Each of the 132 subjects were provided with coded plastic containers containing the stain remover. In reality all subjects received the identical stain remover, which was a diluted solution of a popular chlorine based bleach mixed with some perfume to mask it's smell.

The subjects were provided with an instruction sheet that contained the manipulation for expectation ambiguity. Subjects in the low expectation ambiguity condition read the following statement:

The formulation that you have been given was tested by an independent product testing laboratory and was rated as good on the following 5 point scale.

Excellent, Good, Average, Fair, Inferior.

The subjects in the high expectation ambiguity condition read the following:

The formulation that you have been given was tested by an independent product testing laboratory and was rated as follows:

"The cleansing power of the test sample was estimated using the Modified Photometric Test (MPT) using refracted light at 3457+. The difference between the reflectiveness of a stained cellulose medium before and after administration of the test sample was measured. This sample obtained a differential score of 2.37 Lumens."

The instruction sheet also included measures of expectation and the manipulation check for expectation ambiguity.

Subjects were then provided with detailed instructions on testing the stain remover. Each subject was provided with a piece of stained cloth. Subjects in the low performance ambiguity condition received a piece of cloth whose smooth white surface made it relatively easy for subjects to evaluate the effectiveness of the stain remover. Subjects in the high performance ambiguity condition received a lightly colored cloth with an irregular pattern and a relatively rough texture which made it difficult for them to evaluate the effectiveness of the stain remover.

These manipulations of expectation and performance ambiguities were selected on the basis of a pilot study. They were shown to have significant effects on the variable being manipulated (expectation ambiguity or performance ambiguity) while having no significant effect on other variables (expectation and performance).

After treating the stained cloth with the stain remover for a specified amount of time, subjects were instructed to inspect the cloth for traces of the stain, and to complete the next questionnaire which included measures of perceived performance, performance ambiguity and satisfaction.

Multiple measures were used to assess all the concepts. All ambiguity concepts were measured using three variables. These included a measure of the difficulty of evaluation, a measure of the confidence in the evaluation and a measure of how sure the respondent was with his or her evaluation. The first two measures were based on those used by Yi (1993). The latter two scales were reverse coded to reflect the ambiguity with which the evaluations were being made.

Perceived performance was measured using three 7 point bipolar scales (very low - very high performance, inferior - superior performance, bad - good performance). Similarly expectations and satisfaction were measured with three scales each while disconfirmation was assessed using two measures. The lowest Chronbach was 0.77 for the disconfirmation construct. All other constructs attained s well above 0.8 thereby demonstrating high levels of reliability.

Comparison of regression coefficients using dummy variables was used to test the hypotheses that the relative importance of expectations and performance in determining CS will be different under different levels of expectation ambiguity and performance ambiguity. This methodology permits a direct comparison of the regression coefficients for two regression models. Hypothesis 1 can be tested by comparing the regression model under conditions of unambiguous expectation and unambiguous performance to the regression model under conditions of unambiguous expectation and ambiguous performance.

Similarly hypothesis 2 can be tested by comparing the regression model under conditions of unambiguous expectation and unambiguous performance to the regression model under conditions of ambiguous expectation and unambiguous performance.

Two dummy variables (D1, D2) were added to the data set to represent the three situations being investigated - both performance and expectation being unambiguous (0,0), ambiguous performance, unambiguous expectation (1,0) and unambiguous performance, ambiguous expectation (0,1).

RESULTS

Manipulation checks showed that the manipulations of the two factors were successful. The mean values of expectation ambiguity in the high and low expectation ambiguity conditions were 5.11 and 3.16 (F=118.64130,1 p=0.00). Similarly the mean value of performance ambiguity in the high and low performance ambiguity conditions were 5.04 and 3.03 (F=87.02130,1 p=0.00).

An ANOVA was conducted to ensure that the manipulations of expectation ambiguity and performance ambiguity did not affect the levels of expectation and perceived performance. As the pilot study had indicated earlier, neither manipulation had a significant effect on the levels of expectation or perceived performance. This is important since a significant effect of the manipulations of the ambiguity variables on expectation or performance would have caused confounding, leading to a less than rigorous testing of the hypotheses.

To test hypothesis 1, a dummy variable regression was conducted using dummy variable D1 and observations where expectation ambiguity was low. D1 took the value of 0 where neither performance nor expectation were ambiguous, and was 1 when performance alone was ambiguous. A regression model was run in which the predictor variables were Expectation, Performance, D1*Expectation, D1*Performance and D1. The dependent variable was CS. The result of the regression is depicted in Table 1.

TABLE 1

TABLE 2

The coefficients of Expectation and Performance represent the baseline model (where both performance and expectation are unambiguous) while the variables including the term D1 denote the adjustments made to reach the model in which performance alone is ambiguous. If hypothesis 1 were true, D1*Expectation should be positive and significant (i.e. expectation should exert a stronger influence on CS), and D1*Performance should be negative and significant (i.e. perceived performance should have a weaker influence on CS).

An examination of the coefficients in Table 1 shows that as hypothesized D1*Expectation is positive and significant (at the 0.10 level). Thus as performance ambiguity goes from low to high, the effect of expectation on CS increases (B=0.40). The coefficient of D1*Performance was hypothesized to be negative and significant. Table 1 shows that though D1*Performance is not significant, it is close to attaining significance and is of the correct polarity.

To test hypothesis 2, a dummy variable regression was conducted using dummy variable D2 and observations for which performance ambiguity was low. D2 took the value of 0 where neither performance nor expectation were ambiguous, and was 1 when expectation alone was ambiguous. A regression model was run in which the predictor variables were Expectation, Performance, D2*Expectation, D2*Performance and D2. The dependent variable was CS. The results of the regression are depicted in Table 2.

As before, the coefficients of Expectation and Performance represent the baseline model (where both performance and expectation are unambiguous) while the coefficients beginning with D2 denote the adjustments made to reach the model in which expectation alone is ambiguous. If hypothesis 2 is true, D2*Expectations should be negative and significant (i.e. expectation should exert a weaker influence on CS), and D2*Performance should be positive and significant (i.e. perceived performance should have a stronger influence on CS).

An examination of the coefficients in Table 2 shows that as hypothesized D2*Expectation is negative and significant. Thus as expectation ambiguity goes from low to high, the effect of expectation on CS decreases (B=-0.69). As hypothesized, the coefficient of D2*Performance is positive and significant. As expectation ambiguity goes from low to high, the effect of performance on CS increases (B=0.41).

The above analysis has provided some evidence for the moderating role of performance ambiguity and expectation ambiguity in influencing the effects of perceived performance and expectation on CS.

The CS model which is typically expressed as shown in equation 1, can now be expressed as shown in equation 2 below.

CS=f (Exp, Per, Dis) (1)

CS=f (Exp, Per, Dis, APPer, APExp, AEPer, AEExp) (2)

where Exp refers to expectation, Per refers to perceived performance, Dis refers to disconfirmation of expectation, AP refers to the performance ambiguity, AE refers to expectation ambiguity, and terms such as APPer refer to the product of AP and Per. The multiplicative terms in eq. 2 represent the moderating effect of the ambiguity variables.

TABLE 3

TABLE 4

As suggested by Yi (1989), the regression model using multiplicative terms was analyzed using mean-centered independent variables. This reduces the effects of multi-collinearity which are sure to be present in such models.

Table 3 represents the analysis of the model of CS represented by eq. 1

Table 4 shows the analysis of the multiplicative model represented by eq. 2.

All the variables in the multiplicative model in Table 4 are significant (one barely so) and in the right direction. The negative coefficient for APPer denotes that as performance ambiguity (AP) increases, the effect of perceived performance (Per) on CS reduces. Similarly the negative coefficient of AEExp implies that as the expectation ambiguity (AE) increases, the effect of expectation on CS decreases. The positive coefficients of APExp and AEPer can also be interpreted in a similar manner.

An F test was used to test whether the model indicated by eq. 1 (R square=0.58) is significantly inferior to the model indicated by eq. 2 (R square=0.75) in predicting CS. The resulting F statistic (F4, 128=2.79) was significant at the 0.05 level, indicating that the multiplicative model is significantly superior to the expectancy disconfirmation model in predicting CS.

DISCUSSION

This study demonstrates that the ambiguity with which expectation and performance are evaluated have a significant impact on the influence of prior expectation and perceived performance on CS. As the ambiguity of expectation increases, the influence of expectations on CS decreases, while the influence of perceived performance on CS increases. Conversely as the ambiguity of performance increases, the influence of perceived performance on CS decreases while the impact of prior expectation on CS increases.

Marketers of products with high performance ambiguity should focus on creating very high and unambiguous expectation since CS will be strongly influenced by expectation. Similarly, marketers of products with high expectation ambiguity should focus on achieving very high and unambiguous performance evaluations since CS in such situations will be determined mostly by perceived performance.

This paper extends the research done by Yi (1993) by studying the ambiguity of expectation in addition to the ambiguity of performance. It also uses an experimental design rather than Yi's survey methodology.

This study has limitations. While data was collected from subjects exposed to both high performance ambiguity and high expectation ambiguity, this data has not been used in the analyses reported here. This was because there was no theoretical basis to form hypotheses for this condition. How do consumers who are subject to both ambiguous expectations and ambiguous performance judge satisfaction? This question has not been answered in this paper.

REFERENCE

Anderson, Norman H. (1981), Foundations of Information Integration Theory, New York: Academic Press.

Anderson, Norman H. and Lola Lopes (1974), "Some Psycholinguistic Aspects of Person Perception," Memory and Cognition, 2 (January), 67-74.

Bearden, William O. and J. E. Teel (1983), "Selected Determinants of Consumer Satisfaction and Complaint Reports," Journal of Marketing Research, 20 (February), 21-28.

Birnbaum, Michael H. and Steven E. Stegner (1979), "Source Credibility in Social Judgment: Bias, Expertise, and the Judge's Point of View," Journal of Personality and Social Psychology, 37 (January), 48-74.

Cadotte, Ernest R., R. B. Woodruff, R. L. Jenkins (1987), "Expectations and Norms in Models of Consumer Satisfaction," Journal of Marketing Research, 24 (August), 305-314.

Churchill, Gilbert A., Jr. and C. Suprenant (1982), "An Investigation into the Determinants of Customer Satisfaction," Journal of Marketing Research, 19 (November), 491-504.

Hoch, Stephen J. and J. Deighton (1989), "Managing What Consumers Learn from Experience," Journal of Marketing, 53 (April), 1-20.

Hoch, Stephen J. and Y. Ha (1986), "Consumer Learning: Advertising and the Ambiguity of Product Experience," Journal of Consumer Research, 13 (September), 221-233.

Oliver, Richard L. (1980), "A Cognitive Model of the Antecedents and Consequences of Satisfaction Decisions," Journal of Marketing Research, 17 (September), 46-49.

Oliver, Richard L. and W. O. Bearden (1983), "The Role of Involvement in Satisfaction Processes," in Advances in Consumer Research, Ann Arbor, MI: Association of Consumer Research.

Oliver, Richard L. and W. S. DeSarbo (1988), "Response Determinants in Satisfaction Judgments," Journal of Consumer Research, 14 (March), 495-507.

Swan, John E. and F. I. Trawick (1981), "Disconfirmation of Expectations and Satisfaction with a Retail Service," Journal of Retailing, 57 (Fall), 49-67.

Tse, David K. and P. C. Wilton (1988), "Models of Consumer Satisfaction: An Extension," Journal of Marketing Research, 25 (May), 204-211.

Westbrook, Robert and M. D. Reilly (1983), "Value-Percept Disparity: An Alternative to the Disconfirmation of Expectations Theory of Consumer Satisfaction," in Advances in Consumer Research, Ann Arbor, MI: Association for Consumer Research.

Yi, Youjae (1989), "On the Evaluation of Main Effects in Multiplicative Regression Models," Journal of Marketing Research Society, 31 (January), 133-138.

Yi, Youjae (1990), "A Critical Review of Consumer Satisfaction," in Review of Marketing 1990, ed. Valerie A. Zeithaml, Chicago: American Marketing Association, 68-123.

Yi, Youjae (1993), "The Determinants of Consumer Satisfaction: The Moderating Role of Ambiguity," in Advances in Consumer Research.

----------------------------------------