Evoked Set Size As a Function of Number of Choice Criteria and Information Variability

Joseph A, Belonax, Jr., Bowling Green State University
Robert A. Mittelstaedt, University of Nebraska - Lincoln
ABSTRACT - If consumers construct an evoked set by a sorting process, applying more choice criteria to information which contains more variability should increase the likelihood of achieving a smaller evoked set. A 2 X 3 factorial experiment to test these hypotheses is reported. Results support both hypotheses.
[ to cite ]:
Joseph A, Belonax, Jr. and Robert A. Mittelstaedt (1978) ,"Evoked Set Size As a Function of Number of Choice Criteria and Information Variability", in NA - Advances in Consumer Research Volume 05, eds. Kent Hunt, Ann Abor, MI : Association for Consumer Research, Pages: 48-51.

Advances in Consumer Research Volume 5, 1978      Pages 48-51

EVOKED SET SIZE AS A FUNCTION OF NUMBER OF CHOICE CRITERIA AND INFORMATION VARIABILITY

Joseph A, Belonax, Jr., Bowling Green State University

Robert A. Mittelstaedt, University of Nebraska - Lincoln

ABSTRACT -

If consumers construct an evoked set by a sorting process, applying more choice criteria to information which contains more variability should increase the likelihood of achieving a smaller evoked set. A 2 X 3 factorial experiment to test these hypotheses is reported. Results support both hypotheses.

PROBLEM

Conceptualizing the consumer as an information processor and decision maker, Chaffee and McLeod (1973) present a generalized model which defines the area of consumer information processing. The consumer is seen as making a buying decision according to some rule which combines information from an attribute-by-brand matrix. Within the context of this broad model, studies of information processing may be classified as structural or functional in their orientations.

Structural issues which have been investigated include the nature and number of product attributes used in making product evaluations and/or choices (a number of such studies are cited and discussed in Myers and Alpert, 1977). Other studies have examined the nature of the "evoked set," i.e., the number of brands actually considered by consumers when making a choice (Campbell, 1969; Gr°nhaug, 1973/74; Jarvis and Wilcox, 1973; May and Homans, 1977; Narayana and Markin, 1975; Ostlund, 1973).

Research on functional aspects of information processing has taken many forms (summarized in Bettman, 1977) but often involves fitting one or more decisions rule models to brand evaluations made by persons who have been exposed to an attribute-by-brand matrix of information. As Bettman (1977) has pointed out, some of these models (affect referral, linear compensatory, disjunctive and conjunctive) assume that persons process information by brand while others (lexicographic, elimination by aspects, semi-order lexicographic and additive difference) assume processing by attribute. The somewhat mixed results of studies attempting to fit various models may reflect situational differences in those studies. Further, as Bettman notes, modes of information presentation may bias results in favor of one model or another.

However, coming from a different perspective, it may be suggested that the attempts to fit a single model to a total decision process, especially when the buying situation has one or more relatively unfamiliar aspects, may be overlooking an important intermediate stage. The concept of evoked set suggests that choices are made only after the consumer has constructed a set of "acceptable'' brands (Howard and Sheth, 1969). The implied two-process model, with information first used to form an evoked set, is consistent with the results reported by Pras and Sun, hers (1975). They found the conjunctive model to be the best predictor of preference rank order when all alternatives were considered, but the linear compensatory model to predict best when only acceptable alternatives were considered.

This suggests that the construction of an evoked set may be an important phenomenon and this study is addressed to that issue. While no decision rule models will be fit to preference orderings, the number of alternatives considered acceptable (evoked set size) will be examined with respect to two variables -- the number of choice criteria employed and the nature of the information presented -- according to hypotheses which follow from a "sorting view" of evoked set construction.

HYPOTHESES

Although all consumers may be exposed to the same set of alternative brands in an experimental setting or in the market place, the concept of evoked set implies that not everyone makes a choice from the same set of brands. Indeed, there appear to be substantial inter-individual differences in evoked set size for particular products (Campbell, 1969; Gr°nhaug, 1973/74; Narayana and Markin, 1975). Confronted with an array of information about a fairly large number of brands, it seems reasonable to suppose that a person would try to simplify the situation. While this could be done by reducing the number of brands or reducing the number of choice criteria, it seems more likely that a person would try to reduce the number of brands actively considered for at least two reasons. First, it would reduce the number of comparisons which would ultimately have to be made. Second, and assuming we are talking about attributes of importance, to simplify by eliminating choice criteria would force the consumer to, in effect, forego some of the consumption goals which he or she is trying to accomplish through the purchase of the product.

The consumer's task, then, is to eliminate brands from active consideration based on information about attributes of importance. Since the purpose of evoked set construction is to simplify the ultimate choice process, it seems likely that a person would use an "elimination rule" and proceed by "knocking out" brands which failed to meet some acceptable minimum level on one or more evaluative criteria. It would follow that, as the consumer applied a larger number of evaluative criteria to the awareness set, the likelihood of finding a reason for eliminating brands would increase. Further, the nature of available information should play a role. If the information about attributes consists of some sort of ratings of alternative brands, as the variability of those ratings increased and/or the general level of their desirability decreased, the likelihood of finding a reason to eliminate brands should also increase. Therefore, specifically hypothesized in this study are the following:

Hypothesis 1: The size of the evoked set is inversely related to the number of choice criteria used in the evaluation task.

Hypothesis 2: The size of the evoked set is inversely related to the variability and directly related to the mean level of the attribute ratings presented in the evaluation task.

METHOD

To test these hypotheses, a 2 x 3 factorial experiment was designed with "number of evaluative criteria used" and "variability and mean levels of presented attribute information" as independent variables and the reported size of the evoked set as the dependent variable. Before discussing the operationalization of the variables, four preliminary design considerations will be discussed.

First, the product for which brand information was to be supplied had to be sufficiently complex to allow the use of several evaluative criteria. Further, as will be argued later, it was desirable to select a product about which participants would be relatively uninformed and, yet, one in which they would be somewhat interested. A product which appeared to meet these criteria, micro-wave ovens, was selected.

Second, since the number of attributes was an independent variable, it was necessary to attempt to insure that brand rating information was presented for attributes of importance. Forty-eight knowledgeable respondents (senior home-economics majors enrolled in a course which had previously discussed microwave ovens) rated the importance of and perceived interbrand differences for 16 attributes. Using the "direct dual questioning technique" described by Alpert (1971), 6 attributes with average importance and difference ratings significantly (p < .01) greater than the grand mean were selected as "salient." The six included browning uniformity, cooking precision, cooking uniformity, cooking versatility, oven capacity and warranty protection. When presented to participants in the experiment, the nature of each attribute was described in a short paragraph. Participants began the evaluation task by reading those descriptions which were then briefly discussed by the experimenter.

Third, since the size of the evoked set appears to be related to the size of the awareness set (Jarvis and Wilcox, 1973), it was necessary to hold constant the number of brands about which information was provided. To manipulate the variability of the attribute ratings "standard squares" were used (for reasons described below.) Since there were six salient attributes, information was presented on six "brands" to each participant.

Fourth, because other studies have suggested a relationship between product and/or brand familiarity and the size of the evoked set (Gr°nhaug, 1973/74; Jarvis and Wilcox, 1973; Ostlund, 1973), it was necessary to try to control for product familiarity. To this end, participants for the study were 300 students in an introductory nutrition course which had not discussed microwave ovens in class. Random assignment of participants to the several treatment levels should have reduced the effects of prior product experience. Finally, the "brands" were presented as "real, but disguised" and designated by the letters "A" through "F."

One independent variable, the variability and mean level of attribute ratings, was manipulated by systematically changing the ratings of the 6 previously determined "salient" attributes for each of the six "brands." Within each treatment level, the mean and variability were held constant by the use of "standard squares" (Kirk, 1968). However, between treatment levels, both the mean level and variability were systematically changed from the "low variability" condition (X = 8.0; s2 = 5.9), through the "intermediate variability" condition (X = 7.0; s2 = 8.4) to the "high variability" condition (X = 6.0; s2 = 14.0). It will be noted that, to provide information which could be used for "sorting," the mean levels were decreased as the variability was increased.

While it is possible to hold the means constant and manipulate the variability, such a design would necessitate choosing a rather "middling" value for the mean and would result, in the lower variability conditions, in fewer (if any)"high" ratings on individual attributes. Since some persons are likely to focus on those attributes with high ratings, manipulating variability without the simultaneous adjustment of mean values runs a serious risk of having a "self canceling" effect.

The other independent variable was a stratification according to the number of choice criteria used. After examining the brand ratings and making an overall evaluation of each brand, participants were asked to indicate: (a) the importance of each attribute to the evaluation decision, (b) the minimum rating on each attribute which an oven would have to receive to be evaluated as "acceptable" and (c) the number (and designating letters) of brands considered to be acceptable for purchase. (The answer to this last question is the dependent variable, the evoked set size.) Answers from questions (a) and (b) were combined and a mean calculated for each attribute. Any participant's combined answer on any attribute which exceeded the mean for that attribute was designated an "employed choice criterion'' and the number of such was determined for each participant. Previously it had been determined to consider those using 1, 2 or 3 criteria as the "few choice criteria" group and those using 4, 5 or 6 as the "many" choice criteria group. Splitting in this way resulted in reasonably equal numbers in each condition, with 164 in the "few" group and 136 in the "many" category.

Since the other independent variable was randomized across the entire group of participants, the two variables are "crossed" and the design factorial. Data were analyzed using the ANOVA subroutine of MULTIVARIATE (Finn, 1972). Although there were some differences in cell size (Table 1), recording the variables for unequal cells did not affect significance.

RESULTS

Table 1 shows the mean evoked set size for the various conditions of the study. An examination of the means shows that, as predicted in Hypothesis 1, those employing "few" choice criteria constructed larger evoked sets than those employing "many" choice criteria. Hypothesis 2 predicted that the evoked set would be smaller as the variability of the presented attribute ratings increased while their mean levels decreased. The means are in the predicted order for both the "many" and the "few" choice criteria groups and, therefore, in total. Table 2 presents the results of the two-way univariate analysis of variance. Both main effects are significant while the interaction is not. The results confirm both hypotheses.

TABLE 1

MEAN EVOKED SET SIZE FOR ALL CONDITIONS

TABLE 2

UNIVARIATE ANALYSIS OF VARIANCE

DISCUSSION

The results confirm the view that, even though relatively few brands were presented for evaluation, participants did not consider all of them to be acceptable. The "grand mean" is an evoked set size of only 2.45 brands. That the size of the evoked set seems to depend on the number of choice criteria used and the nature of the available information is consistent with the expectation that participants would engage in a simplification strategy to construct an evoked set. The evoked set is largest when relatively few choice criteria are applied to information which indicates that the brands are of fairly uniform "high" quality. When many choice criteria are applied to information which represents brands to be of relatively lower quality (but very good on some attributes and very poor on others), the participants responded by shrinking their evoked set size. This suggests at least two observations.

First, that people appear to construct an evoked set of a size dependent on the conditions of the study implies that greater attention should be paid to this phenomenon in information processing studies. Since there appears to be no way to directly manipulate the size of an evoked set, it can operate as an uncontrolled and possible confounding variable in information processing experiments. Although there does not appear to be any evidence bearing on the issue, it seems reasonable to assume that "overall" ratings (or rankings) of brands which are not included in the evoked set (i.e., are in the "inept" or "inert" sets) would be less reliable and/ or less valid than those of brands included in the evoked set. Studies which attempt to fit models to "overall evaluation" ratings (or rankings) for all brands in the awareness set (or about which information is provided in an experimental setting) should probably closely consider the salience of the attributes presented and the variability and mean levels of the information to which participants are exposed.

Second, it has been argued that, for persons using more choice criteria, decisions are more complex (Park, 1976). This seems reasonable as long as decision makers are combining attribute information in a manner simulated by one of the compensatory models. Indeed, the finding that persons using more choice criteria constructed smaller evoked sets may be explained by arguing that they anticipated the large number of comparisons which would have to be made and, thus, felt strongly impelled to reduce their evoked set size. However, as noted by Reilly and Holman (1977), using the "number of choice criteria" as a stratifying variable creates self-selected treatment groups and causal inferences become impossible. On the other hand, because there are substantial inter-individual differences in the number and saliency of choice criteria, manipulation of this variable is, if not impossible, so difficult as to rule out a straight-forward test of the relationship between the number of criteria used and complexity. In the present study, since the evoked set shrank as the variability of attribute information increased (while the mean levels decreased), it appears more parsimonious to infer that the use of more choice criteria increased the likelihood of "knocking brands out" by finding some attribute which failed to achieve a minimum acceptance level. To the extent that persons "sort" brands to construct an evoked set, it appears that the use of more choice criteria does not increase complexity but, rather, facilitates simplification.

REFERENCES

Mark I. Alpert, "Identification of Determinant Attributes: A Comparison of Methods," Journal of Marketing Research, 8 (May, 1971), 184-91.

James R. Bettman, "Data Collection and Analysis Approaches for Studying Consumer Information Processing," in W. D. Perreault, Jr., ed., Advances in Consumer Research, Vol. 4, Proceedings of the 7th Annual Conference of the Association for Consumer Research, 1977, 342-48.

Brian M. Campbell, "The Existence of Evoked Set and Determinants of Its Magnitude in Brand Choice," Unpublished doctoral dissertation, Columbia University, 1969.

Steven H. Chaffee and Jack M. McLeod, "Consumer Decisions and Information Use," in S. Ward and T. S. Robert-son, eds., Consumer Behavior: Theoretical Sources, Englewood Cliffs, N. J.: Prentice-Hall, 1973, 385-415.

Jeremy D. Finn, Multivariance: Univariate and Multivariate Analysis of Variance, Covariance and Regression, (Chicago: National Educational Service, 1972).

Kjelf Gr°nhaug, "Some Factors Influencing the Size of the Buyers Evoked Set," European Journal of Marketing, 7 (Winter 1973/74), 232-41.

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

Lance P. Jarvis and James B. Wilcox, "Evoked Set Size --Some Theoretical Foundations and Empirical Evidence," in T. V. Greer, ed., Combined Proceedings Fall Conference of the American Marketing Association, No. 35, 1973, 236-40.

Roger E. Kirk, Experimental Design: Procedures for the Behavioral Sciences, (Belmont, Calif.: Brooks/Cole, 1968).

Frederick E. May and Richard E. Homans, "Evoked Set Size and the Level of Information Processing in Product Comprehension and Choice Criteria," in W. D. Perreault, Jr., ed., Advances in Consumer Research, Vol. 4, Proceedings of the 7th Annual Conference of the Association for Consumer Research, 1977, 172-75.

James H. Myers and Mark I. Alpert, "Semantic Confusion in Attitude Research: Salience vs. Importance vs. Determinance," in W. D. Perreault, Jr., ed., Advances in Consumer Research, Vol. 4, Proceedings of the 7th Annual Conference of the Association for Consumer Research, 1977, 106-10.

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

Lyman E. Ostlund, "Evoked Set Size: Some Empirical Results," in T. V. Greer, ed., Combined Proceedings Fall Conference of the American Marketing Association, No. 35, 1973, 226-30.

Whan C. Park, "The Effect of Individual and Situation-Related Factors on Consumer Selection of Judgmental Models," Journal of Marketing Research, 13 (May, 1976), 144-51.

Bernard Pras and John Summers, "A Comparison of Linear and Nonlinear Evaluation Process Models," Journal of Marketing Research, 12 (August, 1975), 276-81.

Michael Reilly and Rebecca H. Holman, "Does Task Complexity of Cue Interrelation Affect Choice of an Information Processing Strategy? An Empirical Investigation," in W. D. Perreault, Jr., ed., Advances in Consumer Research, Vol. 4, Proceedings of the 7th Annual Conference of the Association for Consumer Research, 1977, 185-90.

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