Toward a Replication Tradition in Consumer Behavior: Cross-Cultural Replication of Bennett and Mandell's Study of the Learning-Information Seeking Hypothesis

ABSTRACT - Two replications of Bennett and Mandell's classic study of the effects of prior product/brand experience on the information-seeking behavior of new car buyers are reported. Bennett and Mandell found significant support for the notions of classical learning theory but only nonsignificant evidence for a hypothesis drawn from the Howard-Sheth theory. Here none of the hypotheses were supported in the first replication conducted in the Midwest. However, the results of the second replication in Norway conformed with the findings of Bennett and Mandell.

Citation:

Johan Arndt, Kjell Gr°nhang, and Richard E. Homans (1981) ,"Toward a Replication Tradition in Consumer Behavior: Cross-Cultural Replication of Bennett and Mandell's Study of the Learning-Information Seeking Hypothesis", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 564-567.

Advances in Consumer Research Volume 8, 1981      Pages 564-567

TOWARD A REPLICATION TRADITION IN CONSUMER BEHAVIOR: CROSS-CULTURAL REPLICATION OF BENNETT AND MANDELL'S STUDY OF THE LEARNING-INFORMATION SEEKING HYPOTHESIS

Johan Arndt, Norwegian School of Economics and Business Administration

Kjell Gr°nhang, Norwegian School of Economics and Business Administration

Richard E. Homans, University of Missouri, St. Louis

R. Nail Maddox, University of Missouri, St. Louis

Frederick E. May, University of Missouri, St. Louis

ABSTRACT -

Two replications of Bennett and Mandell's classic study of the effects of prior product/brand experience on the information-seeking behavior of new car buyers are reported. Bennett and Mandell found significant support for the notions of classical learning theory but only nonsignificant evidence for a hypothesis drawn from the Howard-Sheth theory. Here none of the hypotheses were supported in the first replication conducted in the Midwest. However, the results of the second replication in Norway conformed with the findings of Bennett and Mandell.

INTRODUCTION

The purpose of the study reported here is to a conduct a cross-cultural replication of the often-quoted study of Bennett and Mandell (1969) on the relationship between learning and information-seeking.

As pointed out some years ago by Kollat and his colleagues (1970, p. 329), progress in consumer behavior research is hindered by the lack of a strong replication tradition. Many propositions have shaky empirical foundations as they are based on findings in a single study conducted at a special time, on a special location, in a special research setting, and by a special researcher. Nevertheless, despite their humble origin, such single-shot findings often take the form of "generalized knowledge" as they are quoted and requoted. Reservations as to the generality of the results tend to be leveled out.

OBJECTIVES OF REPLICATION

In general, a purpose of replication is to isolate universals from phenomena which are merely idiosyncratic to particular populations or methods of measurement.

More specifically, the role of replication may be discussed in terms of Campbell's distinction between internal and external validity of empirical findings (Campbell, 1957). Internal validity relates to the impact of the experimental manipulation on the set of independent variables. Hence, the validity may be threatened by lack of controls for extraneous variables producing spurious effects. Replication studies concerned with internal validity would emphasize rigorous design, particularly better controls and allowance for other explanatory variables.

The replication study to be reported here, however, focuses essentially on the external validity of the findings in the original investigation. Therefore the critical issue is the generalizability of the findings. It is true from the philosophy of Hume that induction or generalization can never really be justified logically. However, it is possible to approach the problem pragmatically by examining the sensitivity of the findings originally uncovered to variations in population, physical setting, and different operational definitions of the variables.

All replication studies involve a dilemma. To simplify, the situation is as follows: If the replicator is too "conservative" and the design of the new study is too similar to the original one, the findings risk being only trivial reconfirmations of what is already known. Such research is unlikely to be interesting to the scientific community. If on the other hand the replicator yields to the temptation of extending the study too far, deviating too much from the original design, such "radicalism'' may produce non-comparable data. In this case, a middle road was chosen.

CHOICE OF STUDY FOR REPLICATION

At least three important criteria should be used in the choice of study to be replicated. First, it is imperative that the study has addressed a substantial and central theory area within the field of interest. Second, there should be some minimum of consensus among researchers in the area that the particular study is a significant contribution. These two substantiality criteria are needed as safeguards to avoid the risk of having the Journals flooded by trivial, quick-and-easy replications. A third (and practical) criterion is that the methodology should be reported in sufficient detail to allow the desired comparability of design.

The study chosen for replication was Bennett and Mandell's (1969) investigation of the effects of learning or prior product/brand experience on the information-seeking behavior of new car buyers.

Learning theories have lodged themselves firmly in marketing and consumer behavior as explanations of brand choice and information-seeking and handling (Alderson, 1952; Kotler, 1965; Ray, 1973). The models employed have ranged in complexity from Watson's (1913) simple Stimulus-Response formulation to the elaborate Howard-Sheth model (Howard & Sheth, 1969). In addition to being more complex, the Howard-Sheth model differs from the main body of learning theory in a major way which is of importance here.

Howard and Sheth argue that all experience with a class of objects, whether positively or negatively reinforced, are instructive. The consumer's information requirements and the intensity of his search efforts decline as experience in buying and using the generic product accumulates.

In classical learning theory, learning is portrayed as a function of the number of reinforced trials or the number of successive reinforced trials with a specific stimulus object (brand) (Rail, 1966, Hilgard and Bower, 1966).

The Bennett and Mandell (1969) study was chosen as it appears to be a classic in its own light. It was one of the first major studies addressing the Howard-Sheth theory. As such it is routinely quoted in standard textbooks such as Hansen (1972, p. 333), Howard (1977, p. 141), Engel, Blackwell, and Kollat (1978, p. 239), and Berkman and Gilson (1978, p. 409-10) and comprehensive reviews of consumer information-seeking behavior, such as Newman (1977). A third point was that the methodology seemed to be basically sound.

Fourth, the reporting of the method was detailed enough for the purposes of the present study.

BENNETT AND MANDELL'S STUDY

In their study Bennett and Mandell formulated the following hypotheses:

H1:  The buying experience itself is instructive, whether or not the choice is positively reinforced. As this experience increases, evidenced by the total purchases in the individual's history, the amount of effort expended on information search will decrease.

H2:  As the total number of reinforced purchases of a brand increases, the amount of information-seeking before purchase of that brand will decrease.

H3:  As the number of sequential reinforced purchases of a brand increases, the amount of information-seeking before purchase of that brand will decrease (Bennett and Mandell, 1969, p. 431).

The first reflects the Howard-Sheth model. The other two were drawn from the nations of classical learning theory.

Bennett and Mandell tested the hypotheses with self reports of information-seeking collected from 146 recent new ears buyers chosen from new car registrations in the Spring of 1967 in Harrisburg, Pennsylvania. The respondents were requested to reconstruct their entire car ownership history, were given ample time, and were also encouraged to examine any legal or financial records that might assist their recall. The independent variable, "number of trials", was defined in three ways. The first (T₁), which was used to examine H1, was simply the total number of cars the respondent had owned. For T₂, the total number of purchases of the brand currently owned were counted. T₃ was the length of the current sequence of purchases of the brand currently owned.

To illustrate, consider the history COOCOCC where C is the brand of automobile currently owned and O is any other brand. The values for this individual would be T₁ = 7, T₂ = 4, and T₃ = 2.

The second set of critical data was related to the extensiveness of information-seeking. The measurement instrument was a structured questionnaire specifying ten classes of information sources, see Table 1. Responses were weighted on a judgmental basis (by a panel of experts) according to a scheme believed to reflect the effort involved in consulting various information sources (Table 1). The first hypothesis was not supported (X² =4.46, df = 6, p < .62).

Strong support was found for H2 (X² = 17.80. df = 6, p<.007) and H3 (X² = 22.12, df=6, p<.001), the relationships being monotonic and in the predicted direction.

REPLICATION

As previously mentioned, an objective of this replication was to test the robustness of the original findings to variations in the operational definitions and to changes in physical setting or culture.

The first replication study which was also conducted in the U.S. (St. Louis) introduced some variations in field design.

To test the spatial validity of the results, analysis was also performed on a second set of data collected in Norway. As far as we can judge, the design of the Norwegian study coincided with the Bennett and Mandell study in most respects, the only main difference being the one of location. It should be noted that at that time, Norway was lagging somewhat in national income per capita as compared with the U.S. Because of high fiscal taxes making cars very expensive, the incidence of car ownership was smaller, while the length of ownership for each car acquired was much longer. Hence Norwegian consumers were expected to have experienced less learning as car buyers than American counterparts.

ST. LOUIS STUDY

The first replication study which used data collected from a cluster probability sample of 130 St. Louis households during the Spring of 1966, introduced some variations in the measurements made.

Bennett and Mandell employed a structured questionnaire, while the data in the St. Louis study were derived from responses to a series of open-ended questions. Bennett and Mandell sampled current new car registrations. In the St. Louis replication randomly selected households were asked about their most recent new car purchase.

Responses indicating that the buyer had consulted any of the information sources in Table 1 were recorded. Two different summated measures of information seeking were developed for each subject. In the first, the effort weights reported by Bennett and Mandell were applied. In the second score, no weights were employed; all sources were given an equal value. Bennett and Mandell recorded the total amount of information-seeking for all cars considered combined. The first replication, however, only obtained measures of information sources used for the car bought.

The measures of prior product/brand experience were identified to those employed by Bennett and Mandell.

BERGEN STUDY

The source of the data in the second replication was personal interviews conducted among a sample of 96 recent new car buyers, selected by random sampling among all new car registrations in a given time period in the late Fall of 1969 in Bergen (the second largest city in Norway).

Like in the original study a structured questionnaire was used, and the buyers were asked about their information-seeking for all the brands considered. Also other parts of the procedure paralleled the Bennett and Mandell investigation. The only main difference was that auto shows (see Table 1) was not included as a possible source of information (owing to the absence of this source in Bergen at that time).

Therefore, while the three data sets are contemporaneous, there were certain differences in procedure. These differences are summarized in Table 2.

TABLE 1

SOURCES OF INFORMATION AND IMPORTANCE WEIGHTS

TABLE 2

SUMMARY OF MAIN DIFFERENCES IN DESIGN IN THE THREE STUDIES

THEORETICAL EXPECTATIONS

At this point it is appropriate to indicate the theoretical expectations for the replications and to comment on the meaning of confirmation and disconfirmation in this respect.

Since the procedure was practically the same (and since there was no reason to believe that the Harrisburg and Bergen car buyers would differ significantly in terms of relationships among the variables of interest), the Bergen replication was expected to produce findings supporting the original results.

For the St. Louis replication the expectations were less clear. In this case it is necessary to speculate about the impact of the variations in procedure introduced. The use of open-ended questions (instead of a structured questionnaire) in the St. Louis replication would seem to mean a downward bias in reported information seeking and more error due to inaccurate recall. Hence, the correlations obtained would be suppressed. Limiting the measure of extent of information-seeking to information gathering about the car actually bought (rather than about all cars in the evoked set) was believed to have similar effect. Therefore, the St. Louis study was expected to show weaker relationships for the two hypotheses (H2 and H3) supported by the original study.

In this case, confirmation of the results originally found would strengthen the belief that there is a relationship between learning and extent of information seeking. Spatial validation by the Bergen data would make it more likely that the Harrisburg findings did reflect an underlying behavioral law, and that the findings were not a chance result of some idiosyncrasies of a given setting. Disconfirmation of the findings by the St. Louis study would imply that the underlying "true relationships" were not strong enough to withstand major variations in procedure weakening the measures.

FINDINGS

The correlations of the measures of product/brand experience with the information seeking indices for the St. Louis and Bergen samples are shown in Table 3.

As seen in Table 3, the St. Louis and Bergen data were analyzed by computing product moment correlation coefficients instead of the Chi Square statistic as in the original study. This did not affect the conclusions, but was done to simplify reading and easing comparison. As shown in Table 3, all the signs for the St. Louis data, with the exception of one, were as predicted. However none of the correlation coefficients were significantly different from zero. There was clearly no compelling evidence for any of the three hypotheses.

For the Bergen sample, on the other hand, the results showed a remarkable similarity to the Bennett and Mandell findings. While no support was obtained for H1, the data conformed with the theoretical expectations for H2 (p < .05) and H3 (p < .05).

DISCUSSION

The main finding common to all three studies was the lack of support for the H1 hypothesis based on the Howard-Sheth formulation that all buying experiences in the product category are instructive. A possible explanation is that the development of learning theory as applied here has mainly been concerned with low-involvement or low complexity issues, such as repetitive buying or frequently purchased convenience goods. It may be that for a big-ticket items such as cars (being presumably higher in importance, complexity, and involvement) there is a non-monotonic relationship between learning and information-seeking. At low levels of learning there may be little search for information because consumers, though objectively "needing" information, lack ability to process and evaluate information.

TABLE 3

CORRELATIONS OF PRODUCT/BRAND EXPERIENCE WITH INFORMATION SEEKING: (PRODUCT MOMENT CORRELATION COEFFICIENTS)

A second finding was that the weights assigned to the sources of information did not make any difference.

The main difference in results in the two replication studies was that the Bergen study as expected did support the patterns uncovered in the original study. However, also in line with our expectations, there was no relationship for H2 and H3 in the St. Louis replication.

Hence, it appears in conclusion that the Bennett and Mandell findings were not sensitive to change in physical setting and culture but very sensitive to changes in the operational definitions and way of measurement. The fact that the previous findings were reproduced in a different setting, increases our confidence in the generality of the original results.

The findings of this attempt to replicate the Bennett and Mandell study underscore the need for a replication tradition in consumer research. Further replications should bring in more variables and/or alternative hypotheses to establish under what conditions the original effects found exist. Hence, comprehensive systematic replication is necessary before it is possible to obtain a high degree of confidence in a lawlike generalization.

One solution is to carry out what Lykken (1968) calls "constructive replication", letting second and third researchers start from the conclusion of the original researcher using only the fact or law that the former claims to have established and then let the replicators formulate their own conceptualization and methods for sampling, measurement, and data analysis. To illustrate, instead of using a cross-sectional design, the Bennett and Mandell findings could he put to a mere fundamental test by designing a longitudinal study. A panel study could follow a sample of persons over time through several car purchases.

REFERENCES

Alderson, W. (1952), "Psychology for Marketing and Economics," Journal of Marketing, 17, 119-35.

Bennett, P. D. and Mandell, R. M. (1969), "Prepurchase Information Seeking Behavior of New Car Purchasers - The Learning Hypothesis," Journal of Marketing Research, 6, 430-3.

Berkman, H. W. and Christopher, C. G. (1978), Consumer Behavior: Concepts and Strategies, Encino, California: Dickenson Publishing Company

Campbell, D. T. (1957), "Factors Relevant to the Validity of Experiments in Social Settings," Psychological Bulletin, 54, 297-312.

Engel, J. P., Blackwell R. D., & Kollat, D. T. (1978), Consumer Behavior, Third Edition, Hindsdale, Ill.: The Dryden Press.

Hall, J. F. (1966), The Psychology of Learning, Philadelphia: Lippincott.

Hansen, F. (1972), Consumer Choice Behavior, New York: The Free Press.

Hilgard, E. R., and Bower, G. H. (1966), Theories of Learning, Third Edition, New York: Appleton-Century-Crofts.

Howard, J. A. (1977), Consumer Behavior: Application of Theory, New York: McGraw-Hill Book Co.

Howard, J. A., & Sheth, J.N. (1969), The Theory of Buyer Behavior, New York: John Wiley & Sons.

Kollat, D. T., Engel, J. F., & Blackwell, R. D. (1970), "Current Problems in Consumer Behavior Research," Journal of Marketing Research, 7, 327-32.

Kotler, P. (1965), "Behavioral Models for Analyzing Buyers," Journal of Marketing, 29, 37-45.

Lykken, D. T. (1968), "Statistical Significance in Psychological Research," Psychological Bulletin, 70, 151-9.

Newman, J. W. (1977), "Consumer External Search: Amount and Determinants," in A. G. Woodside, J. N. Sheth, & P. D. Bennett (Eds.), Consumer and Industrial Buying Behavior, New York: Elsevier, 79-94.

Ray, M. L. (1973), "Psychological Theories and Interpretations of Learning," in S. Ward & T. S. Robertson (Eds.), Consumer Behavior: Theoretical Sources, Englewood Cliffs, NJ: Prentice-Hall, 45-117.

Watson, J.B. (1913), "Psychology as the Behaviorist Views It," Psychological Review, 20, 158-77.

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