The Nature of Salient Outcomes and Referents in the Extended Model

Michael J. Ryan, The University of Alabama
Michael J. Etzel, Utah State University
ABSTRACT - The specification of outcomes and referents, although a crucial aspect in the development of the Extended Model, has received little attention. This paper discusses saliency and empirically investigates the elicitation technique. The issues raised by the findings indicate that a good deal of research is needed to investigate both the nature of saliency itself and methods for ascertaining salient outcomes and referents.
[ to cite ]:
Michael J. Ryan and Michael J. Etzel (1976) ,"The Nature of Salient Outcomes and Referents in the Extended Model", in NA - Advances in Consumer Research Volume 03, eds. Beverlee B. Anderson, Cincinnati, OH : Association for Consumer Research, Pages: 485-490.

Advances in Consumer Research Volume 3, 1976      Pages 485-490

THE NATURE OF SALIENT OUTCOMES AND REFERENTS IN THE EXTENDED MODEL

Michael J. Ryan, The University of Alabama

Michael J. Etzel, Utah State University

ABSTRACT -

The specification of outcomes and referents, although a crucial aspect in the development of the Extended Model, has received little attention. This paper discusses saliency and empirically investigates the elicitation technique. The issues raised by the findings indicate that a good deal of research is needed to investigate both the nature of saliency itself and methods for ascertaining salient outcomes and referents.

INTRODUCTION

The specification of attributes has been considered a weak part of multi-attribute composition models (Hughes, 1974). [Although the operational procedures and model structures addressed in this paper are basically the same, there is a major difference between expectancy value models that incorporate product attributes and those, such as the extended model, that utilize behavioral outcomes. The reader unfamiliar with this distinction is referred to the review provided by Ryan and Bonfield (1975). A useful distinction and empirical comparison among common social psychological and consumer research conceptualizations of attitude models is also furnished by Mazis, Ahtola, and Klippel (1975).] Fishbein and Rosenberg, who are generally credited with providing the structural origins of these models, have been cited for providing little guidance in this manner (Wilkie and Pessemier, 1973). Consequently, consumer researchers have developed (Pessemier and Wilkie, 1974) and criticized (Kerman, 1974) methods for attribute specification. Although these developments may prove useful, it seems appropriate to examine the methods promulgated by the originators of these models.

A perusal of Fishbein's earlier writings (e.g., Fishbein, 1963) reveals that he determined salient concepts for the Ao model with an elicitation procedure based on free association techniques developed by Maltzman (Maltzman, 1955; Maltzman, Bogartz, and Breger, 1958). This method continues to be used in the extended model by Fishbein and his associates (e.g., Jaccard and Davidson, 1972) and has been suggested (Fishbein, 1971) and used successfully in advertising research applications (Cowling, 1973a).

This paper discusses the elicitation technique and empirically examines specific issues related to its use. Behavioral outcomes and referents determined by way of elicitation from similar groups were compared across two brands in the same product class; the order in which outcomes and referents were most frequently elicited was compared to individual importance rankings; and the individual importance rankings were analyzed in terms of consistency by group and brand.

ELICITATION AND SALIENCY

It is indigenous to the extended model that it be based on salient outcomes and referents. An outcome is considered salient if it serves as a determinant of the attitude toward a specific behavioral act and if the individual associates the outcome with the act. Although Fishbein has not been explicit about the source of salient referents, the structural similarity between the attitudinal and normative components suggests that a referent would be salient if it serves as a determinant of the subjective norm and if the individual associated the referent with the act. The position taken in this paper is that saliency issues, usually referred to hereafter in terms of outcomes, also hold for referents. Also, the relationship of outcomes with the act is specified according to variable combination rules exhibited by the model. However, structure and other model variables will be mentioned as little as possible in order to avoid confusion and allow a clearer focus on the issue of saliency.

Fishbein (1971) has stated that the only way to determine salient outcomes is a simple elicitation procedure in which individuals are asked what outcomes they associate with a specific behavioral act. It is possible to elicit outcomes associated with but not determiners of the attitude. However, based on information processing research (Miller, 1956), Fishbein believes that only 5 to 9 outcomes are salient (Fishbein, 1967, 1971) and that these salient outcomes are the first in order of elicitation. In fact, the notion of elicitation and saliency are intertwined.

Saliency refers to the fact that the respondent is aware of or conscious of the attribute, that it's on the "tip of his tongue." In other words, it has a high probability of being elicited by the respondent. (Fishbein, 1971: p. 313).

The execution of the elicitation technique consists of asking respondents from a defined group a series of non-directive questions, such as:

Q. What comes into your mind when you think about buying Crest toothpaste?

Q. When I say to you buying Crest toothpaste what do you think of?

Q. Is there anyone you know who might like or dislike you to buy Crest toothpaste?

The simplicity of this technique will not appeal to those who believe consumers cannot divulge such associations in so straightforward a manner (Dichter, 1960; Martineau, 1957). Yet, in a different context, a direct elicitation technique worked as well when compared to more complicated methods and produced results-superior to those obtained with indirect questioning (Alpert, 1971).

Although a detailed criticism of alternative procedures has been furnished elsewhere (Cowling, 1973; Fishbein, 1971), some of these criticisms will be mentioned briefly. First, factor analysis is not deemed appropriate since salient items, while being conceptually distinct, are not always expected to score differently on an index of strength, magnitude, etc. Factor analysis merely identifies sets of items that are highly intercorrelated. For example, white teeth and decay prevention are measures of some relative characteristic. and if correlated, would indicate one factor although they are obviously conceptually distinct outcomes. Although this example is extreme and many would argue that a skilled interpretation of a factor leadings matrix would not produce this result, there may be more subtle distinctions made in the minds of consumers that would not be distinguishable with factor analysis. Also, if the notion that consumers are limited information processors is accepted, there seems to be little to gain in reducing an already small number of dimensions. The theory assumes that outcomes are combined, according to their respective beliefs and evaluations, in an unweighted additive manner. [Research questioning these assumptions has been reviewed by Ryan and Bonfield (1975) and more recent evidence questioning the additivity assumption has been put forth by Troutman and Shanteau (1975).] Only salient items are included in the model. Therefore, all included outcomes and referents are determiners of attitude and social influences respectively. The relative determining power of these few salient items is indicated only in the combination of their respective beliefs and evaluations.

SOME ISSUES AND PROBLEMS

Use of the elicitation procedure has produced good results in correlational studies (Cowling, 1973); Bonfield and Ryan, 1975, Ryan, 1975) and has been shown to be more effective than the use of predetermined lists (Nazis, et al, 1975). Yet, there are some issues and problems that need to be explored. This research examined four such issues.

First, there is the issue of whether or not the elicitation procedure provided generalizable outcomes salient to different groups and brands. Given the wide academic acceptance of the marketing concept and the resulting segmentation strategies, it seems consumer researchers would avoid applying the same set of attributes or outcomes to different groups. For example, using outcomes elicited from college students in a study of housewife purchasing behavior conflicts with all we know about behavioral differences in marketing. Different brand positioning strategies would also suggest that outcomes may be brand rather than product specific. Consequently, this research elicited outcomes for two differently positioned brands across two independent samples similar in life style but geographically separated. It was expected, based on Cowling (1973, 1973a), that elicited outcomes would be similar for the two groups but different for each brand.

Second, the question of whether the procedure will provide outcomes relevant to group members or merely provide-idiosyncratic items needs to be directly addressed. [Wilkie and Pessemier (1973) have described this problem succinctly and suggested individual level analyses across brands as a possible solution.] Although the majority of Fishbein's current work is group oriented (Fishbein and Ajzen, 1975), he has argued that the best estimate of attitude is obtained from a cognitive structure based on individual subject's own elicited concepts (Kaplan and Fishbein, 1969). Also, the majority of successes in applying elicitation procedures have used correlational analysis which can be misleading in terms of model fitting (Birnbaum, 1973). Consequently, outcomes were tabulated by frequency and order of mention in order to ascertain if outcomes could be identified that were common to the majority of group members.

Third, the relative importance of the outcomes, aside from the information derived from belief and evaluation statements, is of interest. Yet, the unweighted composition rules do not allow a direct assessment of importance. Fishbein (1971, 1975) has argued that concepts that are more important have more polarized belief and evaluation statements. He believes that this explains why importance weights, when added as a third cognitive structure variable, do not increase prediction. That is, their effects are already present in the belief and evaluation statements. Although all strongly held beliefs need not be based on salient outcomes, all salient outcomes involve strong beliefs and elicitation is the only way to ascertain saliency (Fishbein, 1971). From this, it seems reasonable to conclude there may be a direct relationship between importance and order of elicitation. This is not to suggest an importance variable as a replacement or addition to the model; rather, it may be useful in terms of parsimony or verification. If order of elicitation indicated outcome importance this information by itself could be useful before construction of belief and evaluation statements or as a cross check on belief and evaluation measurement procedures. Consequently, this research examined the relationship between the order in which outcomes were elicited and individual rankings of outcome importance.

Fourth, given that the above relationship is supported, the use of individual importance rankings suffers from the same limitations in applying individually elicited concepts to all group members. Namely, are the rankings idiosyncratic or is there consistency among group members? Consequently, the individual rank orderings were examined for consistency.

METHOD

Subjects consisted of two convenience samples of undergraduate students enrolled in upper division business courses. The first group, containing 97 subjects, was located at The University of Alabama and the second group, containing 121 subjects, was located at the University of Kentucky. These groups were chosen to represent similar lifestyles although they were expected to be slightly different due to separate regional influences. The product class chosen was toothpaste and the brands were Crest and Ultra Brite. The product class was chosen since it has been a popular subject of multiattribute model researchers who have sometimes used predetermined attribute lists across brands and widely varying sample groups. The brands were chosen to represent different positioning strategies; Crest being positioned as a decay preventative and Ultra Brite as a cosmetic. All data were collected during the Spring of 1975.

Subjects were asked to write their answers to a series of questions similar to those shown above pertaining to each brand. Responses were analyzed and outcomes and referents were tabulated according to total frequency of mentions and frequency of mentions by rank order. Outcomes and referents were then ordered according to frequency of mentions. Finally, each subject rank ordered the outcomes in terms of importance.

Kendall's Tau statistic was used to compare the rank order of elicitation according to frequency of mentions with each subject's independent ranking. Kendall's Coefficient of Concordance was employed to examine the consistency of the independent rank orderings.

FINDINGS AND DISCUSSION

The frequencies of outcome and referent mentions for the Alabama sample are shown in Table 1 and for the Kentucky sample in Table 2. The total frequency of mentions and the frequency by rank order were consistent for both samples and brands. That is, those outcomes and referents mentioned first were also the most often mentioned. It is interesting to note that consistent with Cowling's (1973 a) findings, few items were mentioned by more than 50% of the respondents. Although no rules have been established concerning the number of mentions necessary before an elicited outcome is usable, it does seem inappropriate to ask respondents to rate outcomes when the majority of the outcomes are not salient for the majority of the sample. The good results that have been obtained in studies using such items may be the result of halo effects (Beckwith and Lehmann, 1975). Cowling (1973) has suggested using only the individual's own elicited beliefs although Kaplan and Fishbein (1969) could only conclude that estimates of attitude based on a subject's own elicited beliefs are at least as accurate as estimates based on standard sets of beliefs. At any rate, an idiosyncratic attitude model has limited use to consumer researchers primarily occupied with the understanding of group behavior.

TABLE 1

OUTCOME AND REFERENT TOTAL FREQUENCY AND FREQUENCY OF MENTIONS BY RANK ORDER FOR ALABAMA SAMPLE (n = 97)

TABLE 2

OUTCOME AND REFERENT TOTAL FREQUENCY AND FREQUENCY OF MENTIONS BY RANK ORDER FOR KENTUCKY SAMPLE (n = 121)

The problem of group saliency becomes more important in realistic settings since the groups under examination are likely to be more heterogeneous than those examined in this study. It may be possible, as suggested by Cowling (1973), to start with such a group using common outcomes as a basis for segmentation. On the other hand, a larger sample in the present study may have allowed the identification of subgroups with a large percentage of commonly mentioned outcomes.

A rule of thumb stating that only items mentioned by the majority of respondents be included in the model seems reasonable. Yet, for the groups under study this rule would allow only one outcome and two referents for the Alabama Crest model, no outcomes or referents for the Alabama Ultra Brite model, and one outcome and referent for the Kentucky Crest and Ultra Brite model. Is it realistic to suppose that so few or possibly no outcomes and referents are determiners of attitude and social influence respectively? Possibly so. The notion that 5 to 9 outcomes are expected to be salient is based on studies involving beliefs and attitudes likely to be central to the individual such as racial attitudes (Fishbein, 1967) and attitudes toward birth control (Jaccard and Davidson, 1972). It seems unlikely that toothpaste brand purchase was a central activity for the subjects employed in this study and thus there may have been a very simple, if any, cognitive structure present. In agreement with Nakanishi and Bettman (1974), the cognitive complexity assumed in using numerous outcomes and referents may be an overkill for a low involvement product class such as toothpaste. [Nakanishi and Bettman (1974) found that the addition of outcomes to cognitive structure is an order determined by an independent importance ranking did not significantly improve Aact prediction. However, a predetermined outcome set developed for a different sample was applied across seven toothpaste brands. Consequently, it can be argued that this result was due to a failure to apply brand and group specific outcomes.] In the interest of both parsimony and greater understanding more work is needed to explore models based on single or few outcomes. It also seems important for those testing or exploring the extended model to provide readers of their research with a detailed rationale for the inclusion of outcomes and referents in the model. This would allow research consumers to judge the saliency of outcomes irregardless of correlational results.

The outcomes and referents shown in Table 1 and 2 are compared across brands and samples in Table 3. In both the Kentucky and Alabama sample, only one outcome was common to both brands. This finding supports the rather commonly held notion that consumers may associate quite different sets of outcomes with different brands. The implication for brand preference attitude research is important since it indicates that respondents may base their cognitive structures on different outcomes for different brands. This phenomenon may not manifest itself in response to belief or evaluative rating scales based on non-salient outcomes due to a number of extraneous influences such as demand artifacts (Sawyer, 1975). Again, this points to the need for researchers to explicitly state their rationale for including outcomes in the model. It also indicates that the application of the same outcomes across brands may not provide an adequate test of the extended model. In fact, the lack of common outcomes would obviate strategies based on identifying outcomes with which consumers differentiate brands. It may be that a gestalt approach, considering the entire salient set, is more appropriate if a segment does not have common outcomes across brands.

TABLE 3

OUTCOMES AND REFERENTS

There were few differences in outcomes across the two samples for the same brand and no differences in referents across brands or samples except for the order of elicitation. Since there was reason to expect the samples to be homogenous, this gives some indication that the elicitation procedure is reliable.

TABLE 4

SUMMARY OF INDIVIDUAL CORRELATIONS BETWEEN ORDER OF ELICITATION AND INDEPENDENT RANK ORDERINGS

The findings from the comparison of order of elicitation and individual rank orderings of outcomes and referents are summarized in Table 4. There was a direct relationship between the majority of independent rankings and elicited order of Crest outcomes for both samples. The Crest findings, indicating that the first few elicited outcomes were also the most important for the groups, tends to support the previous suspicion that only one or two outcomes may be salient. However, the Ultra Brite findings showed no preponderance of direct relationships. It appeared that respondents may rank outcomes mentioned by few group members just as important as those mentioned many times. This indicates that importance and saliency are different. These different findings across the two brands are not readily explainable. Yet, the similar results with two independent samples suggests these findings may be reliable. The order of elicitation of Ultra Brite outcomes was the same for the first three outcomes for both samples (Tables 1 and 2). It may be that respondents were reluctant to rank sex appeal as more important than other more rational appearing outcomes when in fact this would have been an accurate response. Again, this may indicate demand artifacts or halo effects. Perhaps the methods suggested by Sawyer (1975) would be useful in investigating this phenomenon. At any rate, whatever the intended use of outcome importance rankings, these findings indicate they may be misleading if applied across brands.

The referent findings were similar. There was a direct relationship between order of referent elicitation and independent importance rankings for the majority of respondents in the Crest model across both groups and for the Ultra Brite Kentucky sample. There were small differences in the frequency of referent mentions for the Alabama Crest sample (Table 1) indicating that these items were equally salient for group members whereas the elicited mentions produced more clearly defined ordering in the other three instances. Also, although the same referents were elicited, the elicited order was different across both brands in the Kentucky sample and the Crest Alabama sample. This finding suggests that the relative importance of referents is indicated by the order of elicitation, and may be brand or group specific.

The findings investigating the consistency of the importance rankings are presented in Table 5. The Crest outcome rankings were moderately consistent across both samples whereas the Ultra Brite outcome rankings showed no consistency across both samples. This finding supports the previous indication that Ultra Brite outcome importance rankings are not meaningful for the two groups. On the other hand, the moderate consistency in Crest importance rankings indicates that in spite of the direct relationship between order of outcome elicitation and importance rankings, there was a good deal of heterogeneity among the individual rankings. Taken together, these findings suggest that outcome importance rankings, since they appear to be idiosyncratic, have little use when applied to groups.

TABLE 5

CONCORDANCE AMONG RANKING OF OUTCOMES AND REFERENTS

The Crest referent rankings were highly consistent whereas the Ultra Brite rankings were moderately consistent for both groups. When considered with the previous findings, only the Crest referent importance ranks appeared to have meaning in terms of group analysis since they confirmed importance as indicated by order of elicitation and were consistent among respondents in both samples.

SUMMARY AND CONCLUSION

The findings from the independent rankings were largely inconsistent. In general, the consistency of outcome rankings varied across brands and groups as well as did the relationship between independent rankings and order of elicitation as ascertained by frequency of mentions. In addition, there was some indication that respondent importance rankings may be subject to artifact influences. Consequently, independent rankings appeared to be of little value. On the other hand, the suggestion that few or no common outcomes may be held salient by members of homogenous groups is of considerable importance since it implies that commonly used formulations of the model may be overly complex for low involvement products.

Based on this study, the following suggestions are made:

1. The appropriate number of elicited outcomes or referents germane to a target group may be a function of the centrality of the beliefs being studied. A reasonable rule of thumb may be to include only items mentioned by more than 50% of the group members.

2. Outcomes and referents used in studies should always be generated from the group to be studied and should be situation specific. Using items generated from other audiences or for other products or brands may prove misleading.

3. The order of item elicitation may be more useful than importance rankings. Importance ranks may be subject to social desirability bias or other yet to be explained factors that undermine their usefulness.

4. A number of studies using varied groups, products, and brands, and comparative methods are needed to further address the questions raised here. In the meantime, it is advisable to use the elicitation technique in the care and recognition of its potential weaknesses.

REFERENCES

Mark I. Alpert, "Identification of Determinant Attributes: A Comparison of Methods," Journal of Marketing Research, 8 (May, 1971), 184-191.

Nell E. Beckwith and Donald R. Lehmann, "The Importance of Halo Effects in Multi-Attitude Models," Journal of Marketing Research, 12 (August, 1975), 265-275.

Michael H. Birnbaum, "The Devil Rides Again: Correlation as an Index of Fit," Psychological Bulletin, 79 (April, 1973), 239-242.

E. H. Bonfield and Michael J. Ryan, "A Practical Application in the Fishbein Extended Model in Ordering Promotional Strategy," unpublished working paper. College of Commerce and Business, The University of Alabama, 1975.

Anthony B. Cowling, "Use of Elicitation Technique for Producing Dimensions of Brand Choice," Proceedings, British Market Research Society, 1973a, 139-156.

Anthony C. Cowling, "Determining and Influencing Consumer Purchase Decisions," European Research, 1 (January, 1973), 26-31, 43.

Ernest Dichter, The Strategy of Desire. Garden City, N.J.: Doubleday and Co., 1960.

Martin Fishbein, "A Behavior Theory Approach to the Relations Between Beliefs About an Object and the Attitude Toward the Object," in Martin Fishbein, ed., Readings in Attitude Theory and Measurement. New York: John Wiley and Sons, Inc., 1967, 389-400.

Martin Fishbein, "A Consideration of Beliefs, and Their Role in Attitude Measurement, "in Martin Fishbein, ed., Readings in Attitude Theory and Measurement. New York: John Wiley and Sons, 1967, 257-266.

Martin Fishbein, "An Investigation of the Relationships Between Beliefs About an Object and the Attitude Toward that Object," Human Relations, 15 (August, 1963), 322-340.

Martin Fishbein, "Some Comments on the Use of 'Models' in Advertising Research," Proceedings. European Society of Market Research, Seminar on Translating Advanced Advertising Theories into Research Reality, 1971, 297-318.

Martin Fishbein, "Attitude, Attitude Change, and Behavior: A Theoretical Overview," in Philip Levine, Ed., Attitude Research Bridges the Atlantic. Chicago: American Marketing Association, 1975, 3-16.

Martin Fishbein and Icek Ajzen, Belief, Attitude, Intention, and Behavior: An Introduction to Theory and Research. Reading, Mass.: Addison-Wesley, 1975.

G. David Hughes, "Trends in the Development of Multi-Attribute Models of Central Information Processing," in G. David Hughes and Michael L. Ray, eds., Buyer/Consumer Information Processing. Chapel Hill: The University of North Carolina Press, 1974, 277-287.

James J. Jaccard and Andrew R. Davidson, "Toward an Understanding of Family Planning Behaviors: An Initial Investigation," Journal of Applied Social Psychology, 2 (July-September, 1972), 228-235.

Kalman J. Kaplan and Martin Fishbein, "The Sources of Beliefs, Their Saliency, and Prediction of Attitude," The Journal of Social Psychology, 78 (June, 1969), 63-74.

Jerome B. Kernan, "Comments on the Role of Summated Attribute Models in the Central Processing of Information,'' in G. David Hughes and Michael L. Ray, eds., Buyer/Consumer Information Processing. Chapel Hill: The University of North Carolina Press, 1974, 353-355.

Irving Maltzman, "Thinking: From A Behavioristic Point of View," Psychological Review, 62 (July, 1955), 275-286.

Irving Maltzman, William Bogartz, and Louis Breger, "A Procedure for Increasing Word Association Originality and Its Transfer Effects," Journal of Experimental Psychology, 56 (1958), 392-398.

Pierre Martineau, Motivation in Advertising. New York: McGraw-Hill, 1957.

Michael B. Mazis, Olli T. Ahtola, and R. Eugene Klippel, "A Comparison of Four Multi-Attribute Models in the Prediction of Consumer Attitudes," Journal of Consumer Research, 2 (June, 1975), 38-52.

G. A. Miller, "The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information," Psychological Review, 62, (1956), 81-97.

Masao Nakanishi and James R. Bettman, "Attitude Models Revisited: An Individual Level Analysis," Journal of Consumer Research, 1 (December, 1974), 16-21.

Edgar A. Pessemier and William L. Wilkie, "Multi-At-tribute Choice Theory - A Review and Analysis," in G. David Hughes and Michael L. Ray, eds., Buyer/Consumer Information Processing. Chapel Hill: The University of North Carolina Press, 1974, 288-330.

Michael J. Ryan, "A Test of Associations in a Causal Chain Derived from Fishbein's Behavioral Intention Model," unpublished working paper. College of Commerce and Business, The University of Alabama, 1975.

Michael J. Ryan and E. H. Bonfield, "The Fishbein Extended Model and Consumer Behavior," Journal of Consumer Research, 2 (September, 1975), 118-136.

Alan G. Sawyer, "Demand Artifacts in Laboratory Experiments in Consumer Research," Journal of Consumer Research, 1 (March, 1975), 20-30.

C. Michael Troutman and James Shanteau, "Do Consumers Judge Product Quality by Adding or Averaging Attribute Information," Kansas State University Human Information Processing Institute Report #75-4, July, 1975. Also presented at The American Psychological Association Meetings, Chicago, 1975.

William L. Wilkie and Edgar A. Pessemier, "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, 10 (November, 1973), 428-441.

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