Measurement and Modeling of Variety-Seeking Behavior: Observations and Implications

ABSTRACT - This paper discusses three recent papers dealing with the measurement and modeling of variety-seeking behavior. The first paper proposes a measure of variety-seeking behavior and suggests applications of such a measure. The second paper tests some refinements to Jeuland's model of variety-seeking behavior. The third paper incorporates variety-seeking into probabilistic models of choice. Rased on the discussion, some broad implications are outlined for future research.


P. S. Raju (1983) ,"Measurement and Modeling of Variety-Seeking Behavior: Observations and Implications", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 84-87.

Advances in Consumer Research Volume 10, 1983      Pages 84-87


P. S. Raju, University of Illinois at Chicago


This paper discusses three recent papers dealing with the measurement and modeling of variety-seeking behavior. The first paper proposes a measure of variety-seeking behavior and suggests applications of such a measure. The second paper tests some refinements to Jeuland's model of variety-seeking behavior. The third paper incorporates variety-seeking into probabilistic models of choice. Rased on the discussion, some broad implications are outlined for future research.


Psychologists have long recognized the need for variety as a basic need under the general rubric of exploratory behavior. Several theories have been put forth to account for such exploratory behaviors (Berlyne 1960; Driver and Streufert 1965; Fiske and Maddi 1961). Several authors have discussed these theories and applied them to the consumer context (Raju 1981; Raju and Venkatesan 1980 Venkatesan, 1973). Recent empirical work has also supported the existence of exploratory tendencies and linked such tendencies with important consumer behaviors such as brand-switching, innovativeness, and information seeking (Raju 1980). These developments have influenced the field of consumer behavior in two ways. First, it is well accepted now that variety is often sought for its own sake and not merely for other reasons such as the existence of multiple needs or multiple use situations. Second, there is a gradual shift from the viewpoint that variety seeking is an inexplicable (or random) part of consumer behavior to the viewpoint that it is "explicable" and can therefore, he incorporated as a part of choice models (McAlister and Pessemier 1982).

As a result, the area of variety-seeking has shown considerable research activity recently. Several new approaches and ideas are being advanced. In this light, the objectives of this paper are as follows:

1. To discuss three recent papers in the area by Handelsman (1983), Hagerty (1983), and McAlister (1983).

2. To outline, on the basis of the discussion, the implications for future research on variety-seeking.

The three papers to be discussed are related in different ways to an earlier model of variety-seeking proposed by Jeuland (1978). Hence this model is first briefly reviewed so that it could serve as a useful basis for discussion.


Jeuland (1978) was the first to explore the possibility of modeling variety-seeking behavior mathematically. In his model the utility of a given brand is a function of the m st experience with it. The mathematical relationship suggested by Jeuland (1978) is:



Ui(t) - the utility of brand i at time t,

Ei(t) - the experience -the consumer has accumulated with brand i up to time t, and

Gi = preference for brand i due to its unique characteristics.

K = a parameter.

As experience with a brand increases, therefore, the utility decreases, accounting for the variety-seeking behavior. The experience Ei(t) at time t is determined by the experience function. This function consists of two components, a forgetting component and a brand experience component.

Ei (t) = C Ei (t-1) + di(t)

where C is a parameter between 0 and 1 accounting for the forgetting of previous experience, and di(t) is 1 when brand i is chosen and 0 if some other brand is chosen. Hence, as time since the last purchase of a brand increases, Ei(t) decreases and the utility of a brand increases.


The paper by Handelsman (1983) proposes a new measure of varied behavior called VBM and points to some managerial applications of the measure. The measure itself is not discussed in depth by Handelsman (1983) in this paper although such a discussion is available elsewhere (Handelsman, 1982). The measure incorporates two types of variety, namely structural and temporal. Structural variety is represented by the total dissimilarity of objects in the set and temporal variety is represented by purchase behavior over time. The latter is conceptually related to the Jeuland model in that experience with a brand is assumed to decay over time and the utility of a brand is inversely related to the experience. The measure proposed by Handelsman (1983) ranges between" and I with" representing the variety avoider and I the variety seeker.

In the paper Handelsman (1983) also relates the VBM measure to the brand assortment in a product class. More specifically, in product classes characterized by a narrow assortment one would assume VBM of consumers to be skewed toward" and in product classes characterized by a broad assortment one would expect VBM to be skewed toward 1. The results for the three produce classes studied (liquid household cleaner, toothpaste, and cake mix) appear to confirm this relationship.

The major contribution of the paper, however, is not the relationship of VBM and brand assortment but the development of VBM itself. There are no existing behavioral measures of variety-seeking. Past studies have primarily used paper and pencil measures employing attitudinal or intention statements (Raju 1980). The construction of the VBM, employing perceptual and behavioral data, is therefore a unique contribution to the field. The measure, however, is not without drawbacks. It requires data that is not easily available, namely a combination of perceptual and panel data. The situation would be particularly complex for household products where several members may he involved in the purchase decision. Also, the measure is specific to a product category and not a general variety seeking measure. Since the measure is not elaborated on in the paper, it would be inappropriate to point out other specific limitations at this time. Although by no means an ideal measure of variety-seeking, the VBM represents a first step in the direction of developing a better measure.

With regard to the application of the VBM measure, the linkage between brand assortment and the skewness of VBM is very tenuous. As stated by Handelsman (1983), the major assumption here is that the market adapts to the variety-seeking needs of consumers by adjusting the brand assortment. Since no evidence of any kind is provided in support of this assumption, the results are by no means conclusive. It is possible to think of situations where there are multiple market segments and the brand assortment is large mainly to cater to these segments although each segment by itself may not be variety seeking. Such situations would be contrary to -the assumption in the paper. There also seems to be a lack of consistency among the three products in the way brand assortment was measured. For example, in the case of cake mixes, flavors were considered as different offerings hut in the case of toothpaste this does not seem to he the case.

Hagerty's (1983) paper presents an empirical test of two possible refinements to the Jeuland model. The first refinement is the "similarity effect." As noted earlier in the discussion of the Jeuland model, the parameter di(t) in the experience function can only take on two values, 1 and 0, depending on whether brand i or some other brand was chosen. Hagerty (1983) suggests that if some other brand is chosen, di(t) should take on some value between 0 and 1 depending on how similar that brand is to brand i. Hagerty, thus, accounts for the similarity between brands which Jeuland did not do.

The second refinement is the "inhibition effect." Jeuland's experience function considers only forgetting due to the passage of time. It is possible that forgetting is accelerated by the trial of other brands, especially when these brands are very similar to the tried brand. As Hagerty (1983) explains, the "similarity effect" and "inhibition effect" counter each other in terms of the implications for variety-seeking.

Hagerty (1983) tests the two refinements using an experimental method in which the stimuli are songs. He finds a significant, although small, magnitude for the similarity effect and only a marginally significant inhibition effect. In effect, the refinements offer only very minor improvements over the simpler Jeuland model.

Although the results did not support the hypotheses very strongly, the paper is interesting from a methodological viewpoint. The statistical solution Hagerty (1983) provides to the problem of making comparisons between trials is quite innovative. He correctly points out the relevance of Helson's (1964) "adaptation level theory" in such situations. The statistical method proposed by Hagerty should prove useful to other experimenters in this area who are interested in analyzing data ranging over several trials. One alternative method, however, is also worth exploring in this regard. It may be possible to present subjects with a standard stimulus at the beginning of the experiment and have them make all ratings relative to this standard stimulus. This could prevent the -anchor-- from changing when evaluations are made over the different trials. Also, the problem of finding two standard stimuli which are assumed to remain stable in preference can be eliminated by this method.

Among the major problems in Hagerty's paper, one is that the "inhibition effect" is not conceptualized sufficiently well. Brand purchase situations are in many ways different from verbal learning situations and it is not clear if similar effects occur in both situations. It is particularly difficult to believe the consumers get confused and forget their previous purchases when they buy similar brands. It is in fact more likely that they remember past purchases and consciously seek out variety or similarity in future purchases. It is therefore not surprising that the "inhibition effect was not strong. On the other hand, it is surprising that the "similarity effect' was not strong. The exclusion of similarities between brands is definitely a limitation with the Jeuland model and one would expect the predictions to improve with the incorporation of that effect into the model. The low -similarity effect- may have been caused by some aspect of the experimental design. For instance, the song sequences were forced upon the subjects as opposed to having a situation where subjects could freely select what they wanted to hear. Also, the stimulus category of songs may be one where subjects have a high tolerance for repetition. Hence, preferences may not change after just a few playings of the same song. As an indication of this tolerance one has to only look at some of the popular radio stations which repeat the a me hit songs at very close intervals. Needless to a y, these drawbacks also jeopardize the external validity of the experiment especially when extending the findings to other types of stimuli.

The third w Per by McAlister (1983) discusses the important aspect of dependence of choices over time. She first introduces two classes of choice models, namely the "random decision rule" models and the random utility models. Both types of models assume independence among the choice alternatives. Hence, similarity among alternatives is not assumed to affect choice. McAlister (1983) points out that Tversky's "elimination by aspects" and "additive" random aspects" models, while incorporating dependence among choice alternatives at one point in time, still do not capture dependence across time. The major contribution of the paper, therefore, is in extending these models further to incorporate dependence over time. This dependence is modeled in terms of the attributes shared by the alternatives in the choice set.

Since much of the paper is devoted to extending the "random decision rule" model, only this will be discussed here. McAlister (1983) assumes a first order Markov process. Hence, in assessing the probability of choosing an alternative at a point in time, the direct impact of only the last choice is considered. This is a limitation of the model especially since variety-seeking and satiation are intimately connected with the number of past trials of a brand. However, the use of the Markov model could perhaps be justified on the basis of simplicity. The conditional probability of choosing brand i, given that brand j was chosen last (Pi/ j, is assessed by discounting the attributes contained in j from the attributes contained in i. The extent of the discounting is dependent on a variety-seeking intensity parameter, V, which varies between 0 and 1. This parameter represents an individual characteristic and recognizes the fact that individuals are different in their need for variety. The discounting of attributes of j would, of course, be higher for those with a greater need for variety.

As pointed out by McAlister (1983), there are two forces that influence Pi/j. First, there is the force due to the shared attributes of j and i, which will decrease the preference for i. However. there is also the force due to the shared attributes of j with all the other alternatives. If i depresses the probability of choosing these other alternatives more than it does for i, the probability of choosing i could, in fact, he enhanced relative to these other alternatives. The conditional probability of choosing i (Pi/j) would therefore he higher than the unconditional probability of choosing i, i.e., when it was not preceded by the choice of i. McAlister (1983) derives this condition mathematically in the paper. The roles of the variety seeking intensity (V), unconditional probability of choosing i (ni), and the configuration of attributes comprising i are also elaborated in the paper.

It might he useful at this point to compare McAlister's model with the Jeuland model. Both models estimate the probability of choice of an alternative over time based on past experience. However, they approach this problem very differently. The crucial difference is that Jeuland considers only the aggregate utility of an alternative whereas McAlister considers an alternative in terms of its constituent attributes. Both approaches have advantages and disadvantages. In McAlister's model, the similarity between alternatives plays a key role because of the attributes contained in each alternative, whereas in the Jeuland model the "similarity effect" is completely ignored. Hence, in Jeuland's model, experience or nonexperience of a brand only directly affects its own utility and not the utility of other alternatives. As a result, Jeuland is able to construct an experience function over time in terms of aggregate utility of each brand whereas McAlister has to use a Markov model to accommodate similarities between alternatives. The use of the Markov model, although appealing, is somewhat restricting in that only the direct impact of the last choice is considered. Also, as stated in relation to the paper by Hagerty, a simple way of incorporating the similarity effect into the Jeuland model is to let di range between 0 and 1, thereby adjusting the utility of an unchosen alternative depending on how similar it is to the chosen one. An advantage of considering attributes is that McAlister (1983) is able to introduce the parameter v in the attribute discounting process, thereby recognizing individual differences in the need for variety. The Jeuland model does not explicitly consider this need for variety. The consideration of attributes, however, does present an added problem for McAlister (1983) in that any interactions among the attributes are ignored in determining the probability of choice. An "interesting point put forth by Jeuland (1978) in formulating his model is the concept of "minimum perceivable difference" (MPD). The MPD recognizes the threshold effect, in that the difference in utility should exceed the MPD before a consumer will switch from one Brand to another. McAlister does not appear to use such a concept although it appears to be relevant. Finally, in terms of the mathematical treatment, McAlister's model is very appealing especially since it is rigorous and extends well established probabilistic models of choice. The functional form of Jeuland's model on the other hand, is only a preliminary attempt at modeling variety-seeking as claimed by the author himself (Jeuland 1978).


The comments here will be restricted to some of the fundamental issues raised by the papers. First, the papers are only looking at the issue of "varied behavior" as opposed to the broader issue of exploratory behavior." Exploratory behavior can be manifested in various ways, not just in the form of brand-switching. Further, exploration can result from various motivations, the major ones appearing to be risk-taking, variety-seeking, and curiosity (Raju 1980). Not only are the papers restricted to one facet of exploratory behavior, but they are also looking only at behavior and not the motivations behind the behavior. Unfortunately by doing so we are not gaining a better understanding of exploratory behavior or even variety-seeking per se. Unless we pay more attention to these motivations the field will probably face problems similar to that faced by research on brand-loyalty. In other words, we would he repeating brand-loyalty research from the opposite perspective of varied-behavior.

A second issue facing the area of variety-seeking behavior is that it does not make sense to assume that people seek variety always. However, this assumption is implicitly made in the Jeuland Model and the McAlister model. These models ignore the learning effect, which is the backbone of brand loyalty. Of course, it is possible to think of situations where a person has an acceptable set of alternatives, and consciously seeks to maximize variety by switching among these alternatives, e.g., breakfast cereals. However, if the learning effect-and the variety seeking effect can he integrated into one model it is likely to be a far better predictor of brand purchase behavior and applicable over a greater variety of situations.

A third issue is that of the consideration of individual differences in variety seeking. Although McAlister uses the parameter V to represent variety seeking intensity, there is no mention of how this will be measured and ho; rigorous the measurement will be. The Jeuland model accounts for this, if at all, only indirectly through differences in the experience function for different individuals. Future research should try and incorporate individual differences more explicitly in models of choice behavior.


Berlyne, D. E. (1960), Conflict, Arousal, and Curiosity, New York: McGraw-Hill Book Company.

Driver, Michael J., and Streufert, Siegfried (1965), "The General Incongruity Adaptation Level (GIAL) Hypothesis: An Analysis and Integration of Cognitive Approaches to Motivation, Working Paper No. 114, Institute for Research in the Behavioral, Economic, and Management Sciences, Purdue University, Lafayette, IN.

Fiske, Donald W., and Maddi, Salvatore R. (1961), Functions of Varied Experience, Homewood, IL: The Dorsey Press.

Hagerty, Michael R. (1983), "Variety Seeking among Songs which Vary in Similarity, in Advances in Consumer Research, Volume In, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Arbor, MI Association for Consumer Research.

Handelsman, Moshe (1989), Measurement of Individual Varied Behavior using Perceptual and Purchase Diary Data, working draft.

Handelsman, Moshe (1983), Measurement of Individual Varied Behavior Across Product Classes: Results and Applications," in Advances in Consumer Research, Volume in, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Arbor. MI: Association for Consumer Research.

Helson, H. (1964), Adaptation-Level Theory, New York: Harper and Row.

Jeuland, Abel P. (1978), Brand Preference over Time: A Partially Deterministic Operationalization of the Notion of Variety-Seeking, in Research Frontiers in Marketing: Dialogues and Directions, Proceedings of the American Marketing Association, ed. Subhash C. Jain, Chicago, IL: American Marketing Association, 33-37.

McAlister, Leigh (1983), "Extending Probabilistic Models of Choice to Incorporate Variety Seeking," in Advances in Consumer Research, Volume In, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Arbor, MI: Association for Consumer Research.

McAlister, Leigh and Pessemier, Edgar (1992), "Variety Seeking Behavior: An Interdisciplinary Review," Working Paper No. 1297-82, Alfred P. Sloan School of Management, Massachusetts Institute of Technology, Cambridge, MA.

Raju, P. S. (1980), "Optimum Stimulation Level: Its Relationship to Personality, Demographics, and Exploratory Behavior," Journal of Consumer Research, Volume 7, 272-82.

Raju, P. S. (1991), 'Theories of Exploratory Behavior: Review and Consumer Research Implications," Research in Marketing, Volume 4, 223-49.

Raju, P. S. and Venkatesan, M. (1980), "Exploratory Behavior in the Consumer Context: A State of the Art Review," in Advances in Consumer Research, Volume 7, ed. Jerry C. Olson, Ann Arbor, MI: Association for Consumer Research. 258-63.

Venkatesan, M. (1979), "Cognitive Consistency and Loyalty Seeking," in Consumer Behavior: Theoretical Sources, eds. Scott Ward and T. S. Robertson, Englewood Cliffs, NJ: Prentice-Hall, 354-84.



P. S. Raju, University of Illinois at Chicago


NA - Advances in Consumer Research Volume 10 | 1983

Share Proceeding

Featured papers

See More


The Psychology of Fun: Inquiry into Consumers' Fun Experiences

Travis Tae Oh, Columbia University, USA
Michel Tuan Pham, Columbia University, USA

Read More


L14. Christmas Decorations in September – What Happened to Halloween? The Effect of Prospective Event Markers on Time Perceptions and Attitudes Towards Promotions

Chaumanix Dutton, University of Southern California, USA
Kristin Diehl, University of Southern California, USA

Read More


A Simple Step to Go Beyond Present: How Visual Entropy Cues Influence Temporal Focus and Consumer Behavior

Gunes Biliciler-Unal, University of Texas at Austin, USA
Raj Raghunathan, University of Texas at Austin, USA
Adrian Ward, University of Texas at Austin, USA

Read More

Engage with Us

Becoming an Association for Consumer Research member is simple. Membership in ACR is relatively inexpensive, but brings significant benefits to its members.