Testing Comparison Level and Predictive Expectations Models of Satisfaction

John E. Swan, University of Alabama in Birmingham
Warren S. Martin, University of Alabama in Birmingham
ABSTRACT - Based on past literature and current developments in the field of consumer satisfaction, models were developed relating to the prediction of satisfaction from initial levels of expectations, disconfirmation of expectations and disconfirmations of past experience on similar products. In order to test these propositions, 8 different regression models were created. Data from 67 student subjects relating to the satisfaction of an automobile were analyzed. The disconfirmation of past performance expectations and initial expectations were the strongest predictors of satisfaction with an adjusted R2 of .47.
[ to cite ]:
John E. Swan and Warren S. Martin (1981) ,"Testing Comparison Level and Predictive Expectations Models of Satisfaction", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 77-82.

Advances in Consumer Research Volume 8, 1981      Pages 77-82

TESTING COMPARISON LEVEL AND PREDICTIVE EXPECTATIONS MODELS OF SATISFACTION

John E. Swan, University of Alabama in Birmingham

Warren S. Martin, University of Alabama in Birmingham

ABSTRACT -

Based on past literature and current developments in the field of consumer satisfaction, models were developed relating to the prediction of satisfaction from initial levels of expectations, disconfirmation of expectations and disconfirmations of past experience on similar products. In order to test these propositions, 8 different regression models were created. Data from 67 student subjects relating to the satisfaction of an automobile were analyzed. The disconfirmation of past performance expectations and initial expectations were the strongest predictors of satisfaction with an adjusted R2 of .47.

INTRODUCTION

This paper briefly reviews the comparison level theory of satisfaction proposed by LaTour and Peat (1979b) and reports an empirical test of this theory along with empirical tests of other models of satisfaction. While the comparison level theory, of satisfaction has intuitive appeal it has only been subject to a single empirical test (LaTour and Peat 1980, 1979a). A secondary purpose of this paper is to compare the LaTour and Peat model to other models in terms of their ability to explain satisfaction. In particular the contributions of the disconfirmation of predictive expectations and initial level of expectations as predictors of satisfaction were tested.

MODELS OF SATISFACTION

Most of the recent models of satisfaction have been oriented towards a paradigm relating initial expectations to perceived performance of the product (or service) (Day 1977). The main proposition is that satisfaction is directly related to the disconfirmation of expectations (Oliver 1977, 1979, Howard and Sheth 1969, Swan 1977). Disconfirmation is a process in which the user subjectively compares post purchase product performance to prior expectations. Positive disconfirmation results if post purchase performance exceeds expectations. Confirmation occurs if performance equals expectations and negative disconfirmation results if performance is short of expectations. Disconfirmation has been conceptualized as continuous variables that can range from a high positive value (as performance exceeds expectations) to a low or zero value (as performance approaches and equals expectations) to a high negative value (as performance decreases relative to expectations (Oliver 1979, 1977; Swan 1977). Satisfaction has also been thought of as a continuous variable that can range from high satisfaction to indifference to dissatisfaction and will be hereafter referred to as "satisfaction" (Day 1977). It should be noted that researchers have not claimed that interval levels of measurement have been achieved for disconfirmation and satisfaction. In addition, whether satisfaction is a continuous variable or whether satisfaction and dissatisfaction are two different states is a point that has not been fully resolved (Leavitt 1977). The main prediction of the Day model, that satisfaction is directly related to disconfirmation has been supported by research findings (Swan 1977; Oliver 1979, 1977; Swan and Combs 1976).

LaTour and Peat (1979b) have discussed the basic disconfirmation of expectations model along with similar approaches. They have critically evaluated much of the past work in this area and offered an alternative conceptualization of consumer satisfaction. LaTour and Pear's basic approach is that satisfaction is a function of how well a product performs on a set of attributes and for each attribute the customer compares performance to a comparison level (Thibaut and Kelly 1959). Favorable performance in relation to the comparison level for an attribute would contribute to satisfaction due to a positive disconfirmation of expectations.

The important theoretical contribution of the comparison level approach is to three possible sources are identified as forming the comparison level: (1) past experience with different levels of the attribute for similar products used for similar purposes; (2) levels of the attribute that the customer believes similar customers have received; and (3) expectations from information provided by marketers or the expected attribute level based on unique characteristics of the specific purchase situation.

As an example, an automobile owner's satisfaction with the attribute miles per gallon (MPG) achieved with a new car would be related to the difference between the obtained MPG compared to the mileage of his previous car, the mileage his friends claim for their cars, and manufacturers, dealer, and EPA mileage estimates. Overall satisfaction is predicted to be a function of the difference between obtained performance compared to the comparison level which is composed of the three sources of performance standards, past experience, experience of others, and the present situation.

HYPOTHESES AND MODELS TESTED IN THIS STUDY

As noted above, prior research has found that satisfaction is directly related to disconfirmation. Implicitly the literature has defined expectations as predictive expectations. Miller (1977a) as well as Summers and Granbois (1977) have argued that consumers may have different kinds of expectations including what Summers and Granbois (1977) have termed predictive expectations. Predictive expectations have been thought of as preusage anticipations of how veil the product will perform (Swan and Trawick 1979). The first model in this study tested the hypothesis that:

H1: Satisfaction will be directly related to the disconfirmation of predictive expectations.

Disconfirmation was measured as the after usage attribute ratings of the test product (Ai) minus predictive expectations (PREDi) summed across 12 attributes. The model used to test the first satisfaction (S) hypothesis was:

EQUATION   (1)

The set of hypotheses tested included some basic ones plus the same hypotheses with an expectation term added because Oliver (1979, 1977) and Swan (1977) have found To expectations explained some variation in satisfaction above the attributable to disconfirmation of expectations. Expectations may be related to satisfaction due to a level of performance effect. The argument is as follows. Expectations indicate how well or poorly the user anticipates the product will perform. Assume that in two cases performance matches expectations. In the first case expectations and satisfaction were both at a high level, the users should be satisfied. In the second case both expectations and performance were low, so user satisfaction should be low. Swan and Trawick (1979) have found that performance meeting high expectations results in higher satisfaction than performance equal to low expectations. The second hypothesis and model was:

H2: Satisfaction will be directly related to both predictive expectation (EXP) and disconfirmation of predictive expectations.

EQUATION    (2)

The full LaTour and Peat theory explains satisfaction as follows:

EQUATION    (3)

Where S = satisfaction; Ai = after usage subjectively experienced attribute level; PREDi = the expected value of the attribute level based on the characteristics of the particular situation; PASTi = past experience with attribute levels for similar products; Ri = attribute levels experience by reference persons; n = number of salient attributes; W1W2W3 = empirically derived weights. Due to some limitations that will be explained more fully below, this study was not designed to test the LaTour and Peat model so that a reduced form of the LaTour and Peat model was tested:

EQUATION   (4)

The PREDi and PASTi terms constitute the comparison level, the third hypothesis combined with this model was:

H3: Satisfaction will be directly related to perceived product performance (Ai) minus the comparison level (PREDi B PASTi)

EQUATION    (5)

The fourth hypothesis added expectation as an independent variable:

H4: Satisfaction will be directly related to expectation and perceived product performance minus the comparison level.

EQUATION    (6)

The LaTour and Peat approach raises the interesting possibility that satisfaction may be related in some way to the disconfirmation of both past performance and predictive expectations. How the variables should be related is not clear from the literature. Because satisfaction theory and research is in its formative stages, we felt that it would be useful to test for relationships between satisfaction and disconfirmation of past performance (hereafter disconfirmation-past) and predictive expectations (hereafter disconfirmation-predictive) including and excluding an expectation term as variations in the LaTour and Peat framework.

The following hypotheses and models were analyzed:

H5: Satisfaction is directly related to disconfirmation-past performance expectations.

EQUATION    (7)

H6: Satisfaction is directly related to expectation and disconfirmation past performance expectations.

EQUATION   (8)

The seventh hypothesis analyzed satisfaction as a function of the disconfirmation of both past performance and predictive expectations and the eighth model included the expectation term:

H7: Satisfaction is directly related to disconfirmation of past performance and predictive expectations:

EQUATION  (9)  and  (10)

METHOD

Design of the Study

The design of the study was a longitudinal approach where expectations (PREDi, PASTi, EXP) were measured prior to a test drive of a Pinto automobile and after the test drive ratings of perceived performance (Ai) and satisfaction (S) were obtained. The 67 subjects were all students in basic marketing classes at a State University.

An "after only" group (N = 9) was given the questionnaire only after the test drive in order to check for pretest effects in the main before-after measures. A comparison of the after scores in the before-after group to the after only group yielded no significant difference between the groups. This limited sample suggests that the experience of rating the Pinto before the test drive did not bias the after drive ratings.

The data for this paper were originally collected as part of a study of the effects of test driving three different distances on ratings of a Pinto. The three drive groups were combined for this paper as the present analysis was by subject, not by experimental group (length of the test drive) and ratings of the Pinto did not vary by experimental group. The data are useful in terms of an analysis of the hypotheses tested in this paper and fit the hypotheses veil; however, it must be recognized that a limitation of the data was that it is a secondary analysis.

CONCEPTUAL AND OPERATIONAL MEASUREMENT

Predictive expectations (PREDi) have been conceptually defined as the anticipated benefits from the product (Howard and Sheth 1969). The anticipated benefits were measured as the before ratings of the Pinto on 12 attributes (prestige, room of interior, engine power, number of options and accessories, attractiveness of interior styling, ride comfort, fun to drive, durability, well engineered, well made, handling, exterior styling). An example of an item appears below:

I EXPECT THE PINTO WILL:

Roomy Interior :__:__:__:__:__:__:__:__:__:__:__:__: Small Interior

Disconfirmation has been defined in two ways: perceived disconfirmation is the consumers post usage perception of how well the product met his expectations, while the concept used in this study, inferred disconfirmation has been defined as the difference between after usage judgments of product performance compared to expectations (Oliver 1980, Trawick and Swan 1980). Disconfirmation of predictive expectations was measured as the difference between the after minus before ratings of the Pinto on the 12 attributes noted above or EQUATION.

One potential problem with the disconfirmation measure is that the same scale was used twice which could lead to similar before and after ratings if the subjects attempted to be consistent. A factor that may have mitigated that problem is the time and driving experience between ratings which was about 15 minutes (short drive), 30 minutes (intermediate drive) or overnight usage of the Pinto (long drive). A second perspective on the disconfirmation measure is that Trawick and Swan (1980) have found that disconfirmation measured as in this paper and disconfirmation measured as the users perception of how well the product met expectations gave essentially the same results as correlates of satisfaction and were strongly intercorrelated.

In order to provide a measure of predictive expectations that was not a part of the disconfirmation measure, predictive expectation (EXP) was measured as the subjects anticipated overall satisfaction with the Pinto.

The LaTour and Peat (1979b) model incorporates a comparison level based on three components of which were included in this study. One component is the anticipated attribute level, which the current authors believe is equivalent to predictive expectations, PREDi. The second component is past experience, PASTi with attribute levels for similar products. This was measured by the subject's ratings of the automobile that he (she) had driven most frequently.

The adequacy of ratings of the car driven most frequently as an operational measure of PASTi could be questioned especially if the car driven most frequently was quite different from a Pinto. About 40% of the subjects had compact or subcompact cars which would be similar to the Pinto. Another reason for using the most frequently driven car is that it should be a very salient experience.

In the LaTour and Peat model PREDi and PASTi are orthogonal. An examination of the data suggested that PREDi and PASTi were orthogonal, as the average simple correlation between the twelve attributes was r = .08 (without respect to sign).

The LaTour and Peat model includes attribute weights (W1, W2, W3):

EQUATION    (11)

The attribute weights are assumed to be equal in this paper. That assumption was made for two reasons. It seems reasonable given that past findings suggest "differential weights provide only limited benefits, compared to the use of equal weights." (Beckwith and Lehnumn, 1973, pg. 141). Therefore for an exploratory examination of these models, this approximation should be adequate. In addition, a method for measurement of the weights has not been developed and to do so was beyond the scope of this paper.

The dependent variable, satisfaction has been defined in a number of ways, yet a common theme has been that satisfaction involves a post usage evaluation of a product or service (Day 1977). Satisfaction has been commonly measured using a simple rating of satisfaction and such methods have given results that agree with more complex measures (Miller 1977). Satisfaction was measured here after the Pinto was driven on a "very satisfactory" to "very unsatisfactory" 7 point semantic differential scale.

Analysis

The hypotheses predicted relationships between satisfaction and other variables. A simple correlation matrix was used to evaluate the hypotheses in a preliminary fashion and to see how the independent variables were related. Next, a set of multiple regression models were used to test the assumed predictors of satisfaction. The regression models were chosen such that no pair of predictors were correlated at a .5 level or higher. The .5 level is arbitrary but seems conservative (Green and Tull 1978, pg. 334). When the predictors were significantly intercorrelated the models were limited to an interpretation of only R2, not the Beta's. Intercorrelation can result in unstable Beta's, however the R2's are meaningful.

RESULTS

The simple correlations in Table I suggest that satisfaction is most closely related to disconfirmation-past (r = .60) followed by initial expectation (r = .51), and disconfirmation of comparison level expectations (r = .22).  A surprising finding was the lack of a significant disconfirmation of predictive expectations and satisfaction relationship. However, an examination of the chits found a negative correlation between expectation and disconfirmation predictive expectations (r = -.43) and the positive expectation, satisfaction relationship (r = .51) suggests that expectation could be surpressing a disconfirmation-predictive and satisfaction relationship. Such may have been the case, as the regression analysis will show below, when expectation and disconfirmation predictive were included as predictors, both variables were important in determining satisfaction.

The first model (I) of Table 2 indicated that disconfirmation predictive was not a significant predictor of satisfaction unless included in a model, Model II, with expectation. Most of the influence on satisfaction was due to the expectation tern, Beta = .70 for expectation versus .44 for disconfirmation predictive. Reasons for that finding were noted above. Disconfirmation of comparison level expectations alone (Model III), did not significantly [P(F) =.1] predict or explain much variation in satisfaction (r2 = .05). Expectation and disconfirmation comparison exp. (Model IV) were able to explain a little less than a third (r2 = .31) of the variation in satisfaction.

Model V with the disconfirmation of past performance expectations as a predictor indicated that disconfirmation-past alone explained about 36% of the variation in satisfaction. Adding expectation, Model VI, increases satisfaction explained (R2 = .42). A Model (VII) with disconfirmation of both predictive and past performance expectations suggest that without expectation satisfaction is only related to disconfirmation-past. However, when expectation, disconfirmation of predictive and pest are tested as predictors, all three are factors influencing satisfaction (Model VIII) and expectation is the most important predictor (Beta = .48). However a limitation of Model VIII is the high intercorrelation between expectation and the other predictors.

DISCUSSION

The results gave very modest support of the LaTour and Peat model where satisfaction is directly related to the disconfirmation of a comparison level composed of subjectively perceived product performance from which predictive expectations and past performance on similar products are subtracted. The simple correlation (r - .22) between satisfaction and disconfirmation of a comparison level was significant but explained little variation in satisfaction. However, a part of the basic concept advanced by LaTour and Peat, that satisfaction is sensitive to perceived product performance minus past performance on similar products was strongly supported. The results are congruent with findings of LaTour and Peat who created expectancies due to past experience with the product, information from the manufacturer and the experience of other users. They found a past experience effect but no effect of manufacturer or ocher users on satisfaction. LaTour and Peat's basic finding has been supported by a study that was quite different from theirs in the product, subjects and setting. A major conclusion is that models of consumer satisfaction need to incorporate the disconfirmation of past performance expectations.

TABLE 1

SIMPLE CORRELATION (R) BETWEEN FIVE VARIABLES

TABLE 2

REGRESSION OF SATISFACTION ON PREDICTORS IN EIGHT MODELS

In contrast to other research (Swan 1977, Oliver 1979, 1977) satisfaction was not related to the disconfirmation of predictive expectations unless that variable was included in a model with expectations. The negative relationship between expectations and the disconfirmation of predictive expectations shows that whose with high expectations had low disconfirmation scores and vice versa. A potential problem with the operational measure of confirmation is that of a ceiling effect on the measurement scales. If an individual expected the Pinto to perform well on an attribute (before rating = 7) and it did so (after rating = 7) the disconfirmation score would be zero (disconfirmation = after-before). A low expectation allows more room on the after scale for improvement to yield a higher disconfirmation score. However, slight satisfaction could be expected from a customer who expected poor performance and received mediocre. Another potential measurement problem was the use of the same scale as the before and after measures.

It appears from this study and prior research (Oliver 1977, 1979; Swan 1977) that the initial level of expectation is a predictor of satisfaction in addition to a disconfirmation effect. This may be due to the ability of expectation to capture an effect of the level of performance that is independent of confirmation. The argument is similar to what was noted above. If an individual had past performance on an auto of 26 miles per gallon and another auto was expected to and did yield 28 miles per gallon, such a customer should be more satisfied than an auto owner who had past experience of 18 miles per gallon in contrast to another car that expected and gave 20 miles per gallon.   The confirmation is equivalent, +2 miles per gallon but the level of performance is higher in the first case.

Work on expectations and consumer satisfaction is proceeding along at least three lines. Oliver has investigated components of and the process of consumer satisfaction. LaTour and Peat have suggested that expectations may involve different reference points; in particular this paper supports the hypothesis that the disconfirmation of past performance expectations is related to satisfaction. Finally, Swan and Trawick (1980) have found that satisfaction is related to the disconfirmation of two different kinds of expectations: (1) predictive expectations, which were not significant in explaining satisfaction in this paper and (2) desired expectations which represent the level of product performance that is necessary to please or satisfy the consumer. Work is needed to integrate and empirically test the different approaches to understanding consumer satisfaction.

An approach to the topic of disconfirmation that has received some attention in the literature is Assimilation/ Contrast effects (Olson and Dover 1979). That approach is beyond the scope of this work. We feel that it was reasonable to omit assimilation/contrast for two reasons. First, assimilation/contrast would have their effects on the after measures of product performance and would be useful in terms of understanding why a particular objection level of product performance received a certain subjective racing. The subjective after ratings were measured and used in this study to define disconfirmation. This study was not designed to explain the after ratings; rather they were used as basic data. A second reason why assimilation/ contrast was omitted is that the approach has criticized on a number of grounds which are too numerous to cover here, (Oliver 1977, Oleos and Dover 1979).

CONCLUSIONS

In conclusion, this study found that the disconfirmation of expectations based on the past performance of similar product expectation were the strongest predictors of satisfaction. A surprising result was that the disconfirmation of predictive expectations (how the product was anticipated to perform) were not related to satisfaction. Except for the work of LaTour and Peat, the effect of different sources of expectations such as past experience or the experience of others has not been investigated. Sources of expectations could be an important topic for future research in consumer satisfaction.

REFERENCES

Beckwith, Neil and Lehmann, Donald (1973), "The Importance of Differential Weights in Multiple Attributes Models of Consumer Attitude," Journal of Marketing Research, 10, 141-145.

Day, Ralph L. (1977), "Extending the Concept of Consumer Satisfaction," in Advances in Consumer Research, IV, ed., William D. Perreault, Jr., Atlanta, Ga.: Association for Consumer Research, 149-154.

Green, Paul E. and Tull, Donald S. (1978), Research for Marketing Decisions, 4th ed., Englewood Cliffs, New Jersey: Prentice-Hall.

Howard, John A. and Sheth, Jagdish N. (1969), The Theory of Buyer Behavior, New York: Wiley.

LaTour, Stephan A. and Peat, Nancy C. (1980), "The Role of Situationally-Produced Expectations, Others' Experiences and Prior Experience in Determining Consumer Satisfaction," Advances in Consumer Research, VII, ed., Jerry C. Oleos, Ann Arbor: Association for Consumer Research, 588-592.

LaTour, Stephan A. and Peat, Nancy C. (1979a), "Determinants of Consumer Satisfaction: A Field Experiment," Proceedings 1979 American Psychological Association Convention Division 23, forthcoming.

LaTour, Stephan A. and Peat, Nancy C. (1979b), "Conceptual and Methodological Issues in Consumer Satisfaction Research," Advances in Consumer Research, VI, ed., William L. Wilkie, Ann Arbor: Association for Consumer Research, 431-437.

Leavitt, Clark (1977), "Consumer Satisfaction and Dissatisfaction: Bi-Polor or Independent," Conceptualization and Measurement of Consumer Satisfaction and Dissatisfaction, ed., H. Keith Hunt, Cambridge, Mass.: Marketing Science Institute.

Miller, John A. (1977a), "Studying Satisfaction, Modifying Models, Eliciting Expectations, Posing Problems, and Making Meaningful Measurements," Conceptualization and Measurement of Consumer Satisfaction and Dissatisfaction, ed., H. Keith Hunt, Cambridge, Mass.: Marketing Science Institute, 72-91.

Miller, John A. (1977b), "Exploring Some Alternative Measures of Consumer Satisfaction," Marketing: 1776-1976 and Beyond, ed., Kenneth L. Bernhardt, Chicago: American Marketing Association, 661-664.

Oliver, Richard L. (1980), "Conceptualization and Measurement of Disconfirmation Perceptions in the Production of Consumer Satisfaction," in Proceedings Fourth Annual Conference on Consumer Satisfaction/Dissatisfaction and Complaining Behavior, forthcoming.

Oliver, Richard L. (1979), "Product Satisfaction as a Function of Prior Expectation and Subsequent Disconfirmation: New Evidence," New Dimensions of Consumer Satisfaction and Complaining Behavior, eds., Ralph L. Day and R. Keith Hunt, Bloomington, Indiana: School of Business, Indiana University, 66-71.

Oliver, Richard L. (1977), "Effect of Expectation and Disconfirmation on Postexposure Product Evaluations: An Alternative Interpretation," Journal of Applied Psychology, 62, 480-486.

Olson, Jerry C. and Dover, Philip A. (1979), 'Disconfirmation of Consumer Expectations Through Product Trail," Journal of Applied Psychology, 64, 179-189.

Summers, John O. and Granbois, Donald H. (1977), "Predictive and Normative Expectations in Consumer Satisfaction and Complaining Behavior," in Advances in Consumer Research IV, Ed., William D. Perrault, Atlanta, Georgia: Association for Consumer Research, 155-158.

Swan, John E. and Trawick, I. Fredrick (1979), "Satisfaction Related to Predictive vs. Desired Expectations," Proceedings Fourth Annual Conference on Consumer Satisfaction, Dissatisfaction and Complaining Behavior, forthcoming.

Swan, John E. and Longman, Douglas S. (1978), "Satisfaction with New Automobile Repair Experience: An Exploratory Study," Proceeding Southern Marketing Association 1978 Conference, eds., Robert S. Franz and Robert M. Hopkins and A1 Toms, University of Southwestern Louisiana, 25-28.

Swan, John E. (1977), "Consumer Satisfaction with a Retail Store Related to the Fulfillment of Expectations on an Initial Shopping Trip," in Consumer Satisfaction, Dissatisfaction and Complaining Behavior, ed., Ralph L. Day, Bloomington, Indiana: School of Business, Indiana University, 10-17.

Swan, John E. and Combs, Linda Jones (1976), "Product Performance and Consumer Satisfaction: A New Concept," Journal of Marketing, 40, 25-33.

Trawick, I. Fredrick and Swan, John E. (1980), "Inferred and Perceived Disconfirmation in Consumer Satisfaction," in Proceedings 1980 American Marketing Association Educators' Conference, forthcoming.

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