Methodological Issues in Evoked Set Formation and Composition

James H. Myers, University of Southern California
[ to cite ]:
James H. Myers (1979) ,"Methodological Issues in Evoked Set Formation and Composition", in NA - Advances in Consumer Research Volume 06, eds. William L. Wilkie, Ann Abor, MI : Association for Consumer Research, Pages: 236-237.

Advances in Consumer Research Volume 6, 1979      Pages 236-237


James H. Myers, University of Southern California

While all three papers in this session expand upon our limited knowledge of evoked sets, they do so in very different ways. May theorizes a linkage between learning theory and information processing theory in evoked set formation based on parts of McGuire's information processing paradigm; Parkinson and Reilly utilize one aspect of consumer information processing theory as a basis for studying evoked set formation in an empirical setting. Both of these papers purport to focus primary attention upon the dynamics of evoked set formation. Belonax studies yet another set of factors that affect the size (but not composition) of the evoked set.

While all papers add to our knowledge in this area, there is some question that any of the papers, or any other work to date for that matter, provides any real insight as to how evoked sets are formed over time within an individual consumer. Parkinson and Reilly deal briefly with this issue when they state, "In an actual temporal sense what probably happens is that the consumer considers one brand at a time and makes a binary decision as to whether or not the brand is worthy of future consideration." This sounds reasonable, but is it really what happens? If it is, would any of the following three processing models be appropriate to reflect this one-at-a-time evaluation: weighted/unweighted linear compensatory, lexicographic? It would seem that these models in particular would be more appropriate for an evoked set formation process which considers all brands in an awareness set at the same point in time. However, it is entirely possible that evoked sets are formed over time by considering each new entry into the awareness set whenever it comes along. It would seem that a conjunctive or a disjunctive processing model would be much more appropriate for this type of evaluation, yet these models did not do nearly as well in predicting "actual evoked sets" as did the other three in the Parkinson and Reilly study. Does this mean that study results suggest that evoked sets are actually formed by comparing and evaluating all brands in the awareness set at a single point in time? But this does not seem to agree with the earlier statement by Parkinson and Reilly about brands being considered "one at a time," unless the entire existing evoked set is reevaluated each time a new entry comes along.

The point of this matter is not whether or not these authors are correct and consistent. The question is whether or not we really know much about the temporal aspects of evoked set formation: not only what kinds of information processing takes place but in what sequence this takes place and over what periods of time. And, of course, how this varies by such factors as psychological and social considerations, product complexity, the learning that might take place, and so on.


Thus, while two of the papers emphasize the "dynamics" of evoked set formation, they do so within a static context. May considers the dynamics of the formation primarily by relating this to two concepts: stages in the adoption process, and stages in the product life cycle. While these factors are indeed dynamic in nature, they seem more appropriate for aggregates than for individuals. There is certainly nothing wrong with examining individuals at different points within each cycle for clues as to what kinds of processing are going on, and this is what May has done. Within the static context, both papers provide some new insights in an area not very well understood.

However, my own view is that we need: (1) a better theory to describe what is going on over time as a given consumer forms his/her own evoked set, (2) research designs that are longitudinal rather than cross-sectional, to trace the formation process within individuals who are in different stages of the process.

For example, it would not be difficult to trace the formation process over time within college students or mail panel participants. The research design might call for the selection of some product category that is new (entirely or relatively) to each respondent and also of some degree of interest. This would be decided by each person on an individual basis. An inventory of the awareness set, the evoked set, and even inert and inept sets (Narayana and Markin, 1975) could be taken at 1 or 2 month intervals over a designated time period. The respondent would be encouraged to learn as much as possible about the product category, using whatever sources of information and/or trial he/she chooses in the normal manner.

At each inventory period, information is gathered about as many aspects of the evoked set formation as possible: present awareness set; evoked, inept and inert sets; sources of first and subsequent information about each new brand entering the awareness set; trial; attribute ratings for familiar brands; preference rankings, etc. The resulting data could be analyzed to track the movement over time of brands into or out of awareness, evoked, inept and inert sets. This movement could be correlated with such factors as current numbers of brands in each of the sets, attribute ratings, information sources, and many others.

This type of longitudinal data would make it possible to construct temporal information processing networks showing the real dynamics of establishing each of the types of sets, the stability of these sets over time, and the types of factors that appear to have the greatest influence on formation, size and composition of the sets. The data could also be tested against the May paradigm combining the information processing and learning theory approaches to evoked set formation. The various brand-attribute ratings could also be processed by the five models used by Parkinson and Reilly to shed light on why a particular brand went into or out of any of the sets.

Another possible research design would follow the suggestions of Parkinson and Reilly that consumers be presented with "the introduction of a totally new brand for which there is no prior evaluation." This brand would presumably be in the form of a detailed concept statement. Several of these concepts could be introduced to the same individuals at successive time periods, and the resulting information processing could be monitored in the manner described above. For this design perhaps all respondents would be asked to consider the same familiar product category, such as toothpaste or deodorant.


If the above longitudinal information is to be analyzed using either the May or the Parkinson and Reilly design, there are a few methodological improvements that should be considered. It would seem that May could provide better operational definitions for some of the cells in his model; e.g., types of information source users, search and retrieval constructs. It would also be helpful if he expanded upon his taxonomy of evoked sets to include considerations other than trial. Other investigators have dealt with size and composition of evoked sets, and it would seem that these factors could and should be studied in longitudinal designs.

Parkinson and Reilly have pointed out their own potential problems with halo and especially with attempting to measure both conjunctive and disjunctive cutoffs. Belonax has used their same approach to establishing cutoffs and would seem to have this problem also. This problem in particular must be solved if we are to have any meaningful test of the congruence between these models (conjunctive and disjunctive) and evoked set formation. My own suggestion is to ask respondents to rate all brands in the awareness set on each attribute and then to indicate brands that are either unacceptable or extremely acceptable on that attribute. Tying cutoffs to actual brands for each attribute should help in clarifying the different levels of acceptability and in identifying brands that either are or are not in the consideration set.

If conjunctive and disjunctive cutoffs were specified more accurately we could then investigate the possibility of a phasing model, which might postulate the establishment of an evoked set by one of these cutoff models followed by evaluation of each brand within the set using a compensatory model of some sort.

The general notion of phasing seems more appropriate to the dynamics of evoked set formation over time.


One of the problems I see in most studies of evoked sets is the matter of operational definitions of the various types of sets. In particular, the evoked set is defined by most investigators (and by Howard & Sheth) as brands the consumer would "consider buying." But Belonax asked what brands the participant would consider to be "acceptable for purchase." Do these terms mean the same thing? And do they mean the same thing for soft drinks as for automobiles? Are there brands that a consumer would consider acceptable but would not seriously consider buying? I believe this is entirely possible.

In the case of automobiles or other expensive items, it might be that several brands would be within some sort of "acceptable" or "possible" range for a consumer but that only one or two brands would be "seriously considered'' due to a risk aversive purchase strategy which requires prior experience with the brand. In the case of soft drinks there is little risk for any purchase, and consumers often consider many different brands for reasons of variety or because each brand may be most appropriate for a certain time of day or drinking situation.

This is more than simply quibbling over semantics. At present the evoked set is considered monolithic. We need construct analysis studies to tell us if there is more than a single layer within this set. It might be that "consider buying" is at least a continuum or perhaps is multidimensional. This would add greatly to our understanding of the evoked set construct and it would certainly provide guidance for designing future studies on the formation of these sets.


Joseph J. Belonax, "Decision Rule Uncertainty, Evoked Set Size, and Task Difficulty as a Function of Number of Choice Criteria and Information Variability," Paper presented at Association of Consumer Research Annual Meeting, October, 1978.

Chen L. Narayana and Ron J. Markin, "Consumer Behavior and Product Performance: An Alternative Conceptualization,'' Journal of Marketing, 39 (October, 1975), 1-6.

Thomas L. Parkinson and Michael Reilly, "An Information Processing Approach to Evoked Set Formation," Paper presented at Association for Consumer Research Annual Meeting, October, 1978.

Frederick E. May, "Evoked Set Formation and Composition: The Learning and Information Processing Hypotheses," Paper presented at Association for Consumer Research Annual Meeting, October, 1978.