The Four Faces of Aggregation in Customer Satisfaction Research

ABSTRACT - This paper applies Epstein's (1980) four faces of aggregation to customer satisfaction research. While existing studies tend to focus on individual subjects and consumption experiences, there is a recent trend toward aggregation. Studying customer satisfaction in the aggregate (over subjects, occasions, stimuli, and measures) should result in more reliable empirical generalizations in an area where disparate empirical findings are common. The discussion illustrates the importance of taking a macro-psychological perspective on customer satisfaction where satisfaction is itself a cumulative, abstract construct on which a variety of products and services may be compared.

Citation:

Michael D. Johnson (1995) ,"The Four Faces of Aggregation in Customer Satisfaction Research", in NA - Advances in Consumer Research Volume 22, eds. Frank R. Kardes and Mita Sujan, Provo, UT : Association for Consumer Research, Pages: 89-93.

Advances in Consumer Research Volume 22, 1995      Pages 89-93

THE FOUR FACES OF AGGREGATION IN CUSTOMER SATISFACTION RESEARCH

Michael D. Johnson, University of Michigan

ABSTRACT -

This paper applies Epstein's (1980) four faces of aggregation to customer satisfaction research. While existing studies tend to focus on individual subjects and consumption experiences, there is a recent trend toward aggregation. Studying customer satisfaction in the aggregate (over subjects, occasions, stimuli, and measures) should result in more reliable empirical generalizations in an area where disparate empirical findings are common. The discussion illustrates the importance of taking a macro-psychological perspective on customer satisfaction where satisfaction is itself a cumulative, abstract construct on which a variety of products and services may be compared.

INTRODUCTION

Research on customer satisfaction relies heavily on the experimental tradition established in psychology. Satisfaction studies typically use individuals as the level of observation and often involve a limited set of both stimuli and product or service use occasions. This paper describes how this approach limits the ability of research to generate empirical generalizations regarding satisfaction. Using Epstein's (1980) "four faces of aggregation," it is argued that greater application of adequate sampling procedures over individuals, stimuli or situations, trials or occasions, and measures facilitates the establishment of generalizable results. Application of the four faces of aggregation in a customer satisfaction context requires that greater attention be paid to: (1) a market level or macro-psychological perspective on satisfaction, (2) defining satisfaction around customers' cumulative experience with a product or service to-date, and (3) operationalizing and modeling satisfaction accordingly.

Nature of the Problem

Although a volume of satisfaction experiments are reported in the literature, empirical generalizations remain elusive (see Yi 1991 for a comprehensive review). A number of studies support expectancy-disconfirmation as a principal determinant of satisfaction while others support direct effects of both performance and expectations on satisfaction. Given the range of performance and expectation related effects on satisfaction, the trend has been to include multiple antecedents in satisfaction models (e.g., Anderson and Sullivan 1993; Churchill and Surprenant 1982; Oliver 1993). However, the size of these effects varies widely across studies.

Endemic to this problem is the concern that experimental studies in the behavioral sciences are inherently limited in their ability to establish reliable generalizations (Ajzen and Fishbein 1974; Cronbach 1975; Epstein 1980; Greenwald 1975; Koch 1959). Epstein (1979, 1980) describes these limitations in his essays on the "stability of behavior," the key points of which are summarized here. One major limitation pertains to the issue of control. The purpose of experimentation is to provide a high level of control. However, adequate control in a behavioral study may be impossible given the wide range of variables that can not be controlled in any given experimental design (Campbell and Stanley 1963). A second limitation is that, when one attempts to increase experimental control in social science research, the result is often an increase in the problem one seeks to reduce. "The achievement of a small error term by controlling the variables that can be controlled magnifies the contribution of uncontrolled situation-specific variables no less than that of experimental variables" (Epstein 1980, p. 794). That is, as one increases control, the effect of highly idiosyncratic or situational factors becomes even more salient which reduces generalizability.

Another limitation relates to the lack of effort expended conducting replication studies, and the relatively small proportion of such studies that are published (Greenwald 1975; Smith 1970). Unless an experiment demonstrates something different, rather than the same result in a different context or situation, it is considered uninteresting. Results that replicate across different contexts or situations are often judged as "not a sufficient contribution" to the field of consumer research. This response says, in effect, that consumer researchers are not interested in establishing empirical generalizations.

Epstein describes two solutions to these problems, the study of higher order interactions and the use of aggregation or sampling. The first, advanced notably by Cronbach (1957), is to pay greater attention to higher order interactions. Modeling individual difference, situation, and stimulus related interactions should, presumably, control for factors idiosyncratic to a particular study and allow researchers to look across studies to see common, generalizable effects. One problem with this solution again lies in the solution itself. "Once we attend to interactions, we enter a hall of mirrors that extends to infinity. However far we carry our analysis - to third order or fifth order or any other - untested interactions of a still higher order can be envisioned" (Cronbach, 1957, p. 119). Years later, it is not surprising that Cronbach (1975) was far more pessimistic in his evaluation of this solution. Epstein's second option is aggregation.

THE FOUR FACES OF AGGREGATION

Aggregation provides two important research benefits. It reduces error in measurement, and provides more generalizable results. Epstein describes four different forms of aggregation: (1) aggregation over subjects, (2) aggregation over stimuli and/or situations, (3) aggregation over trials and/or occasions, and (4) aggregation over measures. After describing the latter three points and their application to satisfaction research, the discussion will focus on aggregation over subjects as having particular relevance.

Aggregation Over Stimuli and/or Situations

A wealth of evidence demonstrates how laboratory findings vary widely from stimulus set to stimulus set and context to context. Nevertheless, many satisfaction studies continue to investigate a particular stimulus set or context, or a limited range of stimuli or contexts, relative to the domain to which the researcher wishes to generalize. Epstein argues that "the need for replication in various settings before a relationship can be accepted as of general theoretical interest can hardly be overstated" (1980, p. 799). As noted, researchers are often biased against replication studies. Aggregation over varying stimuli or situations within an experiment is one solution. That is, individual studies should seek a more broad based set of stimuli and situations as the basis for empirical study. This is not the same as Brunswik's (1947) concerns over the ecological validity of experiments. Rather, aggregation over stimuli and situations is seen as a way of increasing replicability and generality by canceling out uncontrolled unique effects.

It is not surprising that the satisfaction studies that do cut across a wide range of stimuli yield a similar, generalizable result. For example, Andreasen and Best (1977) examined customer satisfaction and complaint data across 35 product and service categories and found greater satisfaction, on average, with products than with services. A primary reason for this difference is that it is inherently more difficult to provide consistent service quality than it is to provide consistent product quality (Gr÷nroos 1984; Zeithaml et al. 1988). Service production involves more of the human resources of the firm and customers themselves. Not surprisingly, Fornell (1992) examined 32 Swedish industries and also found services lagging behind products on customer satisfaction.

Aggregation Over Trials and/or Occasions

Aggregating over trials and/or occasions serves the same purpose. It cancels out any uniqueness due to particular trials or occasions. Epstein groups trials and occasions together because they both involve aggregation over time. Yet aggregating over occasions is seen as more important. Aggregating over specific trials in an experimental session increases "concurrent" reliability. In contrast, aggregating over occasions cancels out incidental effects associated with particular occasions and thus increases "temporal" reliability. The personality literature offers an illustration, where Mischel (1968) notes the relative failure of measured personality traits to predict objective events or behaviors. The objective data in this case consists of single behavior observations or events, usually obtained in a laboratory. In four studies, Epstein (1979) argued and showed that the stability of such relationships can be demonstrated so long as the behavior is averaged over a sufficient number of occasions or events. Although personality traits are not strong predictors of what an individual will do on any given occasion or event, they predict an individual's behavior over several occasions quite well.

The application of occasion aggregation to satisfaction research centers on the distinction between "transaction specific" and "cumulative" satisfaction (Johnson, Anderson and Fornell 1995). Transaction specific satisfaction is a customer's evaluation of a particular product or service experience (Cronin and Taylor 1992). Cumulative satisfaction is the customer's evaluation of their entire purchase and consumption experience to-date (Johnson and Fornell 1991). Epstein's arguments suggest that adopting a more cumulative operational definition of the satisfaction construct will result in more generalizable findings. Oliver (1993) emphasizes that interest in satisfaction research stems, in large part, from the fact that most customer decisions are not initial decisions. Rather, they are decisions influenced by purchase and consumption histories. It is this history, or aggregation over occasions, that drives current and future behavior. An important implication is that aggregation over occasions should produce an operationalization of satisfaction that explains customer loyalty. This aggregation does not necessarily involve taking occasion specific measures and averaging them into an overall experience. It may consist of self-ratings by customers, on a single occasion, that are based on experiences gathered over a period of time.

Aggregation Over Measures

Aggregation over measures is commonplace in consumer research, and specifically research on customer satisfaction. The use of multiple measures to reduce measurement error in causal model estimations has become almost routine (e.g., Churchill and Surprenant 1982; Cronin and Taylor 1992; Fornell 1992; Oliver 1980; Westbrook and Reilly 1983). There is less consensus as to just what the multiple measures of satisfaction should be. Johnson and Fornell (1991) argue for an expanded view of these measures to properly tap cumulative satisfaction. Properly measured, this cumulative satisfaction allows for meaningful comparisons across very different individuals and product categories. Satisfaction, viewed as an abstract construct, allows one to compare seemingly noncomparable products and services (Johnson 1984). Because this satisfaction is an overall evaluation of the customer's consumption experience (a.k.a. consumption utility), satisfaction measures should be chosen that reflect the inherently abstract nature of the construct.

Cumulative satisfaction should thus be operationalized as a latent variable or index using a variety of proxies. Specifically, measures that evaluate or compare performance relative to the different standards that customers use in the course of their purchase and consumption experience provide good reflective measures of latent satisfaction. Fornell (1992) uses three measures of satisfaction in the Swedish Customer Satisfaction Barometer (SCSB): overall satisfaction, confirmation of expectations, and the product's distance from the customer's hypothetical ideal product in the category. These same satisfaction measures are being used in the American Customer Satisfaction Index (ACSI) which is being developed by the National Quality Research Center at the University of Michigan. These measures offer different angles from which customers may express satisfaction. The satisfaction index is itself embedded in a system of cause and effect relationships where expectations and perceived performance affect satisfaction while satisfaction, in turn, affects complaining behavior or "voice" and loyalty or "exit." Importantly, this latent satisfaction index only extracts that portion of the disconfirmation ratings, satisfaction ratings, and ideal point ratings which all three measures have in common and which predicts customer behavior (Johnson and Fornell 1991).

One implication is that the satisfaction index is not confounded by either disconfirmation or performance. It is only the psychological difference between performance and expectations and between performance and the customer's ideal that are used as reflective measures of satisfaction. Another implication is that antecedent measures within a transaction or occasion specific view of satisfaction can be used to provide multiple indicators of cumulative satisfaction. On any given occasion, for example, expectancy-disconfirmation ratings and/or comparisons to an ideal product or service in a category are logical antecedent to satisfaction (Boulding et al. 1993). Across occasions, however, a broader, more abstract satisfaction construct is partially reflected in a subject's rating of performance versus expectations, performance versus their ideal, and overall satisfaction. Operationalizing satisfaction as an index based on the shared variance among these measures is thus consistent with a cumulative view of the satisfaction construct. Naturally, there is a trade-off inherent in this approach. Individual level process details are lost in favor of an approach that balances description and prediction.

Aggregation Over Subjects

Finally, aggregation over subjects is particularly interesting and relevant to customer satisfaction research. Averaging responses over a large sample of subjects increases the stability of findings and their generality. Epstein views aggregation over subjects as common practice in psychological research with proven value. Despite early research by Pfaff (1977; Lingoes and Pfaff 1972), this aggregation has only recently reemerged in customer satisfaction research.

Research on the "time series - cross section paradox" illustrates the value of aggregate level analyses (Adams 1965; Bouwen 1977; Katona 1979). This research examines the ability of consumers' attitudes toward the economy, as expressed in consumer confidence measures, to predict subsequent purchase behavior. Aggregate analyses show a clear attitude-behavior relationship; the more optimistic (pessimistic) customers are as a whole regarding economic conditions, the more likely they are to increase (decrease) their subsequent spending on major durables. Analogous relationships do not emerge from individual level data, where stated buying plans, not attitudes, predict individual purchase behavior.

Bouwen's (1977) study provides insight into this paradox. He introduced two important individual level variables that serve to either mediate or mask the attitude-behavior relationship in this context. One is the consumer's own base-level of optimism-pessimism toward the economy which must be controlled for. The other is the consumer's future time orientation, or the degree to which they plan their purchases into the future. Bouwen predicted an attitude-behavior relationship for those consumers with a future time orientation, controlling for individual differences in optimism-pessimism. His results reveal that a relationship between attitudes and behavior does emerge from individual level data once these variables are introduced. These results are important in that they support the argument that the predictive effect of attitudes observed in aggregate data has a parallel at the individual consumer level which is obscured by individual differences. Katona (1979) concludes that understanding the influence of attitudes on behavior is much simpler when studied in the aggregate where individual differences are often self-canceling random factors. With respect to satisfaction, another advantage of aggregating over subjects is the ability to focus on market segment level data and phenomena. Managers, product planners, and development teams typically focus on entire markets or market segments when making decisions that affect customer satisfaction.

One should not conclude that individual level studies should be replaced by aggregate studies. Rather, the discussion and research described here illustrates the value of augmenting existing individual level studies with studies based on aggregates. As WSrneryd (1988) observes, understanding what is possible is a prerequisite for prediction. Individual level studies provide a rich description of the types of phenomena that are possible. Aggregate level studies help us understand what is probable. It is also important to note that aggregating over individuals may be problematic unless one also aggregates over situations and/or occasions. As more subjects are aggregated into more and more reliable variable means, smaller differences are needed to find a statistically significant result. Minute, incidental differences may emerge that are specific to a particular situation or occasion. The implication for satisfaction research is straightforward. Aggregation over individuals is more valuable when satisfaction is defined and operationalized as cumulative (across occasions) rather than transaction specific (within a particular occasion). Otherwise, a highly context dependent result may appear to be more general than it actually is.

Emerging Generalizations

Recent studies using the SCSB data (Anderson, Fornell and Lehmann 1994; Fornell and Johnson 1993; Johnson, Anderson and Fornell 1995) illustrate the application of the four faces of aggregation to customer satisfaction. These studies examine market segment (firm) level satisfaction (aggregation over individuals). Satisfaction is defined, and measured, as cumulative in nature. Customers are asked to reflect back over their recent experiences when assessing performance and satisfaction rather than focus on a particular product or service encounter (aggregation over occasions). The analyses cut across a wide range of product-oriented to service-oriented industries (aggregation over stimuli). Finally, satisfaction is an index composed of a broad-based set of multiple measures (aggregation over measures).

Fornell and Johnson (1993) examined the effects of differentiation on satisfaction across industries. Differentiation, or the degree to which customers choose among predictably different options in an industry, is shown to have very systematic effects on aggregate expectations, perceptions of performance, and subsequent satisfaction. As differentiation increases, perceived performance increases which, in turn, increases aggregate satisfaction. Differentiation also increases customers' aggregate expected level of performance in an industry. Following Van Raaij (1989) and Katona (1979), these increased expectations have a separate positive effect on aggregate satisfaction. Overall the study supports the generalizability of basic marketing principles. That is, when a heterogeneous population of customers has a wider variety of predictably different options to choose from, satisfaction increases.

Anderson, Fornell and Lehmann (1994) focus on the recent debate regarding the financial payoff from increasing quality and satisfaction. As satisfaction increases, customer retention should increase thereby reducing one's marketing costs and increasing revenues. Thus satisfaction should increase profitability. Using ROA (return on asset) data in combination with the SCSB, their findings support a positive impact of customer satisfaction on profitability. Their results also call into question the traditional assumption that market share and profitability are positively related. In fact, they find a negative relationship; increasing market share may actually decrease satisfaction suggesting that they are not necessarily compatible goals.

In the third study, Johnson, Anderson and Fornell (1995) used the aggregate data to test alternative models of market level expectations, perceived performance, and customer satisfaction. The authors argued that aggregate performance expectations, like price expectations, should be largely rational in nature. Unlike price expectations, however, these performance expectations should remain adaptive to changing market conditions. Performance information is revealed over a longer time period than is price information making the aggregate performance expectations more adaptive. Their results support the adaptive nature of performance expectations. They also show that market level satisfaction, expectations, and perceived performance are relatively stable constructs over time.

Overall these studies suggest that an aggregate level of analysis provides a number of emerging generalizations. Providing a heterogeneous population of customers with a variety of predictably different options increases satisfaction. Because it is more difficult to differenticate a service, service satisfaction is generally lower than product satisfaction. Satisfaction, in turn, has a significant positive impact on customer retention and profitability. Increasing satisfaction may also be inconsistent with a firm's market share goals. Finally, aggregate performance expectations are not completely "rational" in a strict economic sense. Rather, they remain adaptive as performance information is revealed over time.

There are several directions for future research using aggregates in this area. One is the study of cross cultural differences in satisfaction. National satisfaction indices are now in place in Sweden, Germany and the U.S. while indices are being developed in Taiwan and New Zealand. The degree to which similar industries show similar levels of satisfaction across these cultures will provide an interesting test of the degree to which industrial organization variables versus cultural differences drive aggregate satisfaction. A second research direction is to explore those industrial organization variables, beyond differentiation, that likely affect aggregate satisfaction. A third research direction is to study performance drivers across industries and cultures. Specifically, to what degree does "fitness for use" or customization of a product or service affect aggregate satisfaction versus its reliability, or "things gone wrong."

CONCLUSIONS

As interest in customer satisfaction research continues to grow, researchers face an important choice. Some will choose to conduct experimental studies focusing on individual subjects, events, stimuli, and/or measures. These studies will provide a level of detail that is essential to our understanding of the nature and antecedents of satisfaction. At the same time, these studies will not necessarily generate empirical generalizations. To do so, researchers should also study satisfaction from an aggregate perspective. Using Epstein's (1980) four faces of aggregation, this paper argues that aggregation over stimuli, occasions, measures, and individuals will help provide empirical generalizations in this area. Existing studies of aggregate satisfaction reveal a number of these emerging generalizations.

Yet the implications of aggregation stem far beyond empirical findings. Taking an aggregate perspective on satisfaction affects how we conduct satisfaction research, how we define satisfaction, and how we measure it. Aggregation entails a market level or macro-psychological perspective on satisfaction. This satisfaction is defined as customers' cumulative experience with a product or service to-date. This, in turn, requires researchers to take a broad view of the measures used to operationalize satisfaction as an inherently abstract construct on which different people and product categories may be compared.

REFERENCES

Adams, F. Gerard (1965), "Prediction with Consumer Attitudes: The Time Series - Cross Section Paradox," The Review of Economics and Statistics, 4, 367-378.

Ajzen, Icek and Martin Fishbein (1977), "Attitude-Behavior Relations: A Theoretical Analysis and Review of Empirical Research," Psychological Bulletin, 84, 888-918.

Anderson, Eugene W., Claes Fornell, and Donald R. Lehmann (1994), "Customer Satisfaction, Market Share, and Profitability: Findings from Sweden," Journal of Marketing, 58 (July), 53-66.

Anderson, Eugene W. and Mary Sullivan (1993), "The Antecedents and Consequences of Customer Satisfaction for Firms," Marketing Science, 12 (Spring), 125-143.

Andreasen, Alan R. and Arthur Best (1977), "Consumers Complain - Does Business Respond?," Harvard Business Review, July-August 1977, 93-101.

Boulding, William, Richard Staelin, Ajay Kalra, and Valerie Zeithaml (1993), "A Dynamic Process Model of Service Quality: From Expectations to Behavioral Intentions," Journal of Marketing Research, 30 (February), 7-27.

Bouwen, RenT (1977), "Anticipation and Realization: Attitudes and Buying Plans in the Future Time Orientation of Consumer Behavior," Psychologica Belica, 17 (2), 113-134.

Brunswik, Egon (1947), Systematic and Representative Design of Psychological Experiments, Berkeley: University of California Press.

Campbell, Donald T. and Julian C. Stanley (1963), Experimental and Quasi-Experimental Designs for Research, Chicago: Rand McNally.

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

Cronbach, Lee J. (1957), "The Two Disciplines of Scientific Psychology," American Psychologist, 12, 671-684.

Cronbach, Lee J. (1975), "Beyond the Two Disciplines of Scientific Psychology," American Psychologist, 30, 116-127.

Cronin, J. Joseph, Jr. and Steven A. Taylor (1992), "Measuring Service Quality: A Reexamination and Extension," Journal of Marketing, 56 (July), 55-68.

Epstein, Seymour (1979), "The Stability of Behavior: I. On Predicting Most of the People Much of the Time," Journal of Personality and Social Psychology, 37, 1097-1126.

Epstein, Seymour (1980), "The Stability of Behavior: II. Implications for Psychological Research," American Psychologist, 35 (9), 790-806.

Fornell, Claes (1992), "A National Customer Satisfaction Barometer: The Swedish Experience," Journal of Marketing, 56 (January), 6-21.

Fornell, Claes and Michael D. Johnson (1993), "Differentiation as a Basis for Explaining Customer Satisfaction Across Industries," Journal of Economic Psychology, 14 (4), 681-696.

Greenwald, Anthony G. (1976), "An Editorial," Journal of Personality and social Psychology, 33 (1), 1-7.

Gr÷nroos, Christian (1984), "A Service Quality Model and Its Marketing Implications," European Journal of Marketing, 18 (1), 36-44.

Johnson, Michael D. (1984), "Consumer Choice Strategies for Comparing Noncomparable Alternatives," Journal of Consumer Research, 11 (December), 741-753.

Johnson, Michael D., Eugene W. Anderson, and Claes Fornell (1995), "Rational and Adaptive Performance Expectations in a Customer Satisfaction Framework," Journal of Consumer Research, forthcoming.

Johnson, Michael D. and Claes Fornell (1991), "A Framework for Comparing Customer Satisfaction Across Individuals and Product Categories, Journal of Economic Psychology, 12 (2), 267-286.

Katona, George (1979), "Toward a Macropsychology," American Psychologist, 34 (2), 118-126.

Koch, Sigmund (1959), "Epilogue," in S. Koch (ed.), Psychology: A Study of a Science, Vol. 3, New York: McGraw Hill.

Lingoes, James C. and Martin Pfaff (1972), "The Index of Consumer Satisfaction: Methodology," in M. Venkatesan (ed.), Association for Consumer Research: 3rd Annual Conference Proceedings, 689-712.

Mischel, Walter (1968), Personality and Assessment, New York: John Wiley.

Oliver, Richard L. (1980), "A Cognitive Model of the Antecedents and Consequences of Satisfaction Decisions," Journal of Marketing Research, 17 (November), 460-469.

Oliver, Richard L. (1993), "Cognitive, Affective, and Attribute Bases of the Satisfaction Response," Journal of Consumer Research, 20 (December), 418-430.

Pfaff, Martin (1977), "The Index of Consumer Satisfaction: Measurement Problems and Opportunities," in H. Keith Hunt (ed.), Conceptualization and Measurement of Consumer Satisfaction and Dissatisfaction, Cambridge, MA: Marketing Science Institute, 36-71.

Smith, N. C., Jr. (1970), "Replication Studies: A Neglected Aspect of Psychological Research," American Psychologist, 25, 970-975.

Van Raaij, Fred (1989), "Economic News, Expectations, and Macro-Economic Behavior," Journal of Economic Psychology, 10 (4), 473-493.

WSrneryd, Karl-Erik (1988), "Economic Psychology as a Field of Study," in W. Fred Van Raaij, Gery M. Van Veldhoven, and Karl-Erik WSrneryd (eds.), Handbook of Economic Psychology, Dordrecht, The Netherlands: Kluwer Academic Publishers, 2-41.

Westbrook, Robert A. and Michael D. Reilly (1983), "Value-Percept Disparity: An Alternative to the Disconfirmation of Expectations Theory of Consumer Satisfaction," in Richard P. Bagozzi and Alice M. Tybout (eds.), Advances in Consumer Research, Vol. 10, Ann Arbor, MI: Association for Consumer Research, 256-261.

Yi, Youjae (1991), "A Critical Review of Customer Satisfaction," in V. A. Zeithaml (ed.), Review of Marketing 1990, Chicago: American Marketing Association.

Zeithaml, Valerie A., Leonard L. Berry, and A. Parasuraman (1988), "Communication and Control Processes in the Delivery of Service Quality," Journal of Marketing, 52 (April), 35-48.

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