Semantic Confusion in Attitude Research: Salience Vs. Importance Vs. Determinance

ABSTRACT - Salience, importance, and determinance have often been incorrectly used in attitude research, with consequent confusion and lowered explanatory power in several studies. This paper is written to clarify these key concepts, compare their predictive validity, and suggest methods for dealing with unresolved issues in the attitude research areas for which these concepts are most relevant.

Citation:

James H. Myers and Mark I. Alpert (1977) ,"Semantic Confusion in Attitude Research: Salience Vs. Importance Vs. Determinance", in NA - Advances in Consumer Research Volume 04, eds. William D. Perreault, Jr., Atlanta, GA : Association for Consumer Research, Pages: 106-110.

Advances in Consumer Research Volume 4, 1977   Pages 106-110

SEMANTIC CONFUSION IN ATTITUDE RESEARCH: SALIENCE VS. IMPORTANCE VS. DETERMINANCE

James H. Myers, The University of Southern California

Mark I. Alpert, The University of Texas at Austin

ABSTRACT -

Salience, importance, and determinance have often been incorrectly used in attitude research, with consequent confusion and lowered explanatory power in several studies. This paper is written to clarify these key concepts, compare their predictive validity, and suggest methods for dealing with unresolved issues in the attitude research areas for which these concepts are most relevant.

INTRODUCTION

Over the past few years there has been semantic confusion in the usage of three terms: salience, importance, and determinance. The first two appear to have been used almost interchangeably in many studies involving attitude measurement and/or product positioning, especially the latter. At least one paper seems to use all three terms interchangeably (Ryan & Etzel, 1975). This paper is written with several objectives; to (a) clarify these key concepts, (b) review research studies that compare these terms, and (c) suggest unresolved issues in the salience-importance-determinance controversy, along with promising directions for research to deal with these issues. It will he shown that semantic confusion has exacerbated problems with using these concepts in multi-attribute models and other attitude measurement applications. Future research may be improved through more rigorous use and definition of these terms.

Much of the confusion arises because all terms refer in some way to the "importance" of a feature or attribute in forming an individual's overall disposition toward a product/service, as manifested by: overall evaluation ratings, usage frequency, choice among brands, etc. What the consumer means by importance is that a feature or attribute in some way "makes a difference'' in brand choice or product usage. Yet there are different kinds of importance, and professional scholarship requires that these differences be identified and defined. To this end the present authors propose the following glossary.

Salience

The earliest definition of salience the present authors were able to find comes from a discussion of attitude measurement in Krech and Crutchfield (1948). They state (p. 163) "Saliency refers to the fact that not all of a man's beliefs stand out with equal prominence in his cognitive field. He may be more acutely aware of certain of his beliefs than others, they may enter his thoughts more readily, they may be more frequently verbalized--they are, in a word, salient." Ryan and Etzel (1975), following the operationalization suggestions of Fishbein (1971), measured saliency by noting the order in which attributes of a product were verbalized when respondents were asked to name the things that "come to mind when you think about buying . . ." (in their study, Crest toothpaste).

Importance

Perhaps the most useful definition of importance would come from a current dictionary, since this term has apparently not been rigorously defined by attitude measurement scholars. A suitable definition would be, "weighty, momentous, of great consequence, significance, or value" (Webster's, 1961). When a feature or attribute is "important" to people it presumably has some consequence or significance in making choices among brands or in forming overall evaluations or rankings of products.

Importance has typically been operationalized by asking respondents to indicate, "How important are each of these features in choosing a . . . ?" (Ryan and Etzel, 1975; Alpert, 1971; Anderson, Cox, and Fulcher, 1976). Numerous applications of this stated importance measure in attempts to validate modified versions of Fishbein/ Rosenberg multiattribute models are mentioned in Wilkie and Pessemier (1973). Among the methodological difficulties encountered in this research stream, using stated importance may inaccurately reflect the attributes' real contributions to overall attitude towards the product or service. In some instances attributes may have been inappropriately included, or excluded, from the model due to using stated importance and/or salience measures. Aside from the usual problems of respondent lack of knowledge and/or unwillingness to reveal true reasons, there is the question of importance vs. determinance.

Determinance

This term was first defined for the marketing literature by Myers and Alpert (1968). They state, "Attitudes toward features which are most closely related to preference or to actual purchase decisions are said to be determinant; the remaining features or attitudes--no matter how favorable--are not determinant" (p. 13).

It is clear that determinance includes but goes beyond "importance." A feature or product attribute can be of extreme importance (e.g., safety of automobiles) yet have no real effect on choice among competing products/services if all are perceived to be equal for this feature. It is therefore difficult to know what a respondent means when he assigns importance weights, unless the dual-question approach (asking them both importance and perceived differences) is used.

One method for operationalizing determinance involves (Alpert, 1971):

D_i = I_i x Y_i where

D_i = Determinance of i^th attribute

I_i = Stated importance of i^th attribute

Y_i = Perceived differences among products, in i^th attribute.

To counter remaining problems of knowledge and willingness to respond truthfully, alternatives to this direct, dual-question method have been proposed (myers and alpert, 1968). For example, a straight-forward method of inferring determinance by covariation between ratings of attributes and overall evaluations uses the simple correlation between them. This can usually be done only on an aggregative basis, of course, unless the number of products rated by each individual is quite large. This measure avoids multicollinearity problems of inferring determinance from beta-coefficients, yet still distinguishes between attributes that correlate with overall attitude and those that are not related. Problems of halo-effects (beckwith and lehmann, 1975) and interactions among attributes may hinder the use of this measure in some instances (myers and gutman, 1974), although it is often valid and useful, given the requisite data base of product attribute ratings.

It is important to distinguish between two possible alternative definitions of determinance:

1. Determinance as a threshold model; i.e., no feature can be important unless the various competing products are perceived to differ to at least some meaningful extent in terms of this feature. Beyond these minimal differences, the amount of variation among products is not necessarily related to the degree of influence the feature has upon overall evaluation or choice.

2. Determinance as a parametric model; i.e., influence of a feature is positively related to the degree of variation among products that feature; the more the variation the greater the influence.

The matter of covariation is the crucial issue here--if a feature is "important" it must "covary" with overall evaluation or choice. But if a feature has no effective variance, it cannot have any covariance; hence, it would seem that the threshold model would have to be valid.

The parametric model states that the greater the variation in a feature, the greater the covariation with some dependent variable. While this is not at all necessary from a purely statistical standpoint, it may turn out to be true from a behavioral standpoint. Some evidence on this matter will be presented later in this paper.

COMPARISONS AMONG TERMS

Since the terms salient and important are most often used synonymously let us first examine the relationship between these two. Salience, in current usage, has to do with the order of elicitation of product/service features that are considered "important" by consumers. Are the moat important features elicited first, the next most important features second, and so on?

Ryan and Etzel (1975) studied the relationship between order of elicitation and importance for both outcomes (e.g., flavor, color, decay prevention) and referents (e.g., dentist, family, close friends). They concluded, "In general, the consistency of outcome rankings varied across brands and groups as well as did the relationship between independent rankings (of importance) and order of elicitation as ascertained by frequency of mentions .... The order of elicitation may be more useful than importance rankings." Clearly, they find differences between importance and order of elicitation (salience), even though their study provides no clear evidence as to which is the more useful concept in explaining consumer product evaluations or choice decisions.

These findings would not surprise Krech and Crutchfield; they observed: "Attitudes, too, differ in importance and in saliency, and here again the correlation between the two is not perfect," (1948, p. 164). While noting a generally positive correlation between importance and salience, they believed there would be numerous instances in which salient attitudes would be not important, and vice versa. Among the reasons they cite include repression of "antisocial attitudes," and those attributes that are so taken for granted that they are seldom spoken of, hence they would be interpreted as non-salient in most studies (1948, p. 257).

This latter notion suggests a distinction between "important" attributes and "determinant" attributes. A seemingly important attribute may be non-salient, due to the expectation that all brands may have satisfactory levels of the attribute. If such is true, it will be left off the list of salient attributes, but this is not serious, for in such conditions an attribute will not determine brand choices (e.g., purity, in a toothpaste). However, some toothpaste attributes, for example, may be salient but non-determinant, since they come to the "top-of-mind" because they are heavily advertised slogans. One may think of Ultra-Brite as having "sex appeal" but buy it in spite of this attribute or avoid it entirely. Lacking a dependent variable, the Ryan/Etzel study did not explore the predictive validity of attributes measured as "salient," "important," or "determinant," although they did show differences in the measures of the first two.

Several examples of differences between "importance" and "determinance" are described in Myers and Gutman (1974). Table 1 shows the results of a study in which fifty Los Angeles businessmen rated airlines with which they were familiar on each of fourteen attributes plus an overall evaluation. They were also asked to state on a five-point scale how important they considered each attribute. The "correlated" column measures determinance as the correlations between attribute ratings and overall evaluations. The rank correlations between this measure and stated importance ranks was .04 (Kendall tau), indicating little agreement. It is also worth noting that safety record was stated as most important yet ranked next to last in "determining" overall evaluation, probably because respondents perceived little difference among airlines in this trait. Two other comparisons of stated vs. correlated importance estimates reported in this same paper produced similar results.

Finally, "salience" has also appeared in the literature of multi-dimensional scaling, particularly for applications of INDSCAL (Green and Rao, 1972, p. 53 and 236ff). The routine produces output of where individuals are located, given their MDS perceptions, and provides magnitude estimates for the "salience" of each dimension. This may be the same as determinance, in the sense of the variation that is observed in ratings of objects along some revealed (by MDS) dimension is a determinant of the overall evaluations of these objects, provided the MDS input has been expressed in terms of perceived similarities between objects and an "ideal" object, thereby yielding a preference map. However, since this is the farthest thing from "top-of-mind" verbalization of attribute weights, it can hardly be called "salience," in the psychological sense used by Kretch & Crutchfield. "Hidden" dimensions should not at the same time be termed "salient." Further, most MDS routines work with similarity maps, rather than preference maps; hence revealed dimensions are neither "salient" nor necessarily related to preference distributions, in most MDS applications. The dimensions revealed as "salient" by MDS might certainly be reasonable candidates for "determinants" of purchases, given that they can be adequately interpreted as dimensions and then operationalized in products, at least as readily as those pre-specified attributes used in alternative approaches. To these authors' knowledge, there has not yet been a comparative test of the predictive validity of MDS's "salient" dimensions vs. those identified as "important" or "determinant" or "salient" by other methods. The next section will summarize the comparative validity research that has been done.

TABLE 1

STATED VS. CORRELATED IMPORTANCE: AIRLINE TRAVEL

Comparative Research with Determinance vs. Importance

Since the term was introduced, a number of studies have been published which measure attribute determinance by taking account of perceived differences in attributes also viewed as important to purchase decisions (e.g., Alpert, 1972; Peterson & Alpert, 1972; Anderson, Cox & Fulcher, 1976). Three in particular have attempted to test for differences in the predictive validity of "importance-measures" that rely upon differing numbers of aspects of the attributes rated (such as "importance," "differences," "saliences"). Alpert (1971) found that attributes chosen as determinant using scales for both importance and differences were better able to model overall preference ratings for inexpensive pens than were those selected using importance-ratings alone.

Wilkie and Weinreich (1972) found no differences in the patterns of predictive validity obtained with models formed by choosing supermarket attributes via either importance scores or by determinism scores. However, their measure of determinism involved multiplying standardized importance scores (I_i^*) for an attribute, times the standard deviation (s_i, over all stores) of the i^th attribute (D_i = I_i^* x s_i). Given that the first scale is standardized within the respondent across attributes and the second across stimuli only, they may have been unequally weighted in producing determinism scores. In fact, the importance component may have -dominated, thereby leading to artificially similar results between the two methods.

A recent study by Berkowitz, Ginter, and Talarzyk (1976) provides both confusion and clarification of the issues involved. The confusion is that they consider stated importance and determinism to be indicators of attribute "saliency." We would prefer that what they term "salience" be called determinance (or perhaps importance), and that the issue would then become one of how best to measure determinance of various attributes upon some relevant dependent variable.

These authors used the Wilkie and Weinrich measure of determinism and found it to have about the same predictive power in selecting attributes for predicting automobile rank order in a "Fishbein/Rosenberg" multiattribute model as did importance scores alone. This finding was similar to the earlier Wilkie/Weinrich study. However, Berkowitz, et al., noted that this measure yielded larger variation in standardized importance scores than the standard deviations of attribute ratings. When they corrected this by standardizing the standard deviations prior to multiplying them by the standardized importance ratings, this new determinism measure produced a very different ranking of attributes for inclusion in the model, along with substantially more accurate predictions.

The net impression from these studies is that differently operationalized measures of attribute determinance have differing predictive validities, but that equally-weighted consideration of perceived variations along attributes, along with importance scores, can significantly improve the identification of those attributes that are strongly related to overall preferences.

UNANSWERED QUESTIONS

There are a number of issues which these comparative validity studies do not resolve concerning the "salience-importance-determinance" question. Additional research is needed to resolve many issues, and its character is dependent on the practical and theoretical questions involved.

For example, one issue concerns how one might identify the relative determinance of attributes in shaping brand choices, taking the attribute as a whole, rather than specific levels of the attributes. Assuming that current brands represent the relevant range for the attributes researched, one might proceed to compare a variety of methods for choosing determinant attributes (Alpert, 1971). In addition to those discussed above, one might consider refinements of dual-question methods, for example. A third question might be asked to ascertain the perceived desirability of existing levels of attributes among current alternatives. One potential weakness of the dual question method is that an attribute may be important, but considered as non-determinant, since all brands are seen as equal in that attribute. Without examining the absolute level of satisfaction, a manager might neglect the improvement and promotion of this attribute. This makes sense if all target customers are essentially satisfied (e.g., safety in banks), but not if they are unhappy with all (e.g., impersonal treatment by banks). In the latter case a favorable niche might be obtained by focus on an attribute that is high in stated importance, low in differences, yet low in satisfaction with available alternatives.

It is possible that refinements, such as adding a 3rd dimension to dual questioning, can improve the identification of determinant buying attitudes. More extensive comparative validity studies might provide better evidence for choosing a particular approach.

One advantage of obtaining evidence on the relative ability of dual question, stated importance, salience, and other methods for estimating attribute weights is that accurate alternatives to obtaining a complete matrix of attribute ratings are desirable. There is currently reason to believe that inferring determinance from covariation among attributes and overall evaluations (or product choice when discriminant analyses are used) is perhaps more accurate than these short-cut methods. However, additional data requirements increase survey costs and respondent fatigue. Further, if one seeks to use the individual as the unit-of-analysis (Wilkie and Pessemier, 1973, give good arguments for this), it would be desirable to avoid the degrees-of-freedom problems associated with numerous attributes and limited brands, when correlation is employed to estimate weights. [Gensch, Golob, and Recker (1976) provide a rationale for employing cross-sectional analysis on some relevant group of customers, arguing that aggregation individuals' response functions may not be managerially meaningful. They propose multinomial logit analysis to estimate relative determinances (they call them importances) without being subjected to some distortions resulting from regression analysis of multi-attribute data.]

Another issue involves determining thresholds needed to establish minimum acceptable levels of attributes (Bettman, 1974), as well as the relative accuracy of alternative cognitive models when these levels are violated (Wright, 1975). While determinance may be measurable in the typically relevant range for many decisions, in some instances this may be inaccurate for particular types of customers. Further, in times of scarcity, it is worth determining what minimum acceptance levels for costly features are, and what happens to overall utility as these are approached.

Recent work with conjoint measurement (Johnson, 1974; Graen, Wind, and Jain, 1972) shows potential promise to reveal not only relative determinance of attributes in general, but also the utility of specified levels of each attribute. As the number of attributes and levels within attributes increase, conjoint methods become increasingly difficult. Research is needed to indicate the circumstances under which these methods are superior, as well as those for which more traditional methods are as good or better. It may be, for example, that one can combine dual questions to pre-screen for determinant attributes, with conjoint measurement of the key attributes and trade-offs that are obtained.

Major problems remain in attitude research. Identifying which attitudes towards which attributes, lead to what behavioral consequences, in what contexts, for what types of people, are but a few of the areas where additional inquiry is taking place. Clarity in definitions and use of such key terms as "importance," "salience," and "determinance" will not solve all the problems and confusions among differently drawn conclusions of various studies in attitude-modeling. However, examination of the studies reported here indicates that more rigorous definition and use of these terms may be necessary and useful.

REFERENCES

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

Mark I. Alpert, "Personality and the Determinants of Product Choice," Journal of Marketing Research, 9 (February, 1972), 89-92.

W. Thomas Anderson, Jr., Eli P. Cox III, and David G. Fulcher, "Bank Selection Decisions and Market Segmentation," Journal of Marketing, 40 (January, 1976), 40-45.

Nell E. Beckwith and Donald R. Lehmann, "The Importance of Halo Effects in Multi-Attribute Attitude Models," Journal of Marketing Research, 12 (August, 1975), 265-75.

Eric N. Berkowitz, James L. Ginter, and W. Wayne Talarzyk, "An Investigation of Alternative Indicators of Attribute Saliency and Their Effects on the Size of the Attribute Set in the Multi-Attribute Model," American Marketing Association Proceedings, 1976, 611-15.

James R. Bettman, "A Threshold Model of Attribute Satisfaction Decisions," Journal of Consumer Research, 1 (September, 1974), 30-35.

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

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.

Dennis H. Gensch, Thomas F. Golob, and Wilfred W. Recker, "Regression is Inappropriate for Analyzing Cross-Sectional Multiattribute Data," American Marketing Association Proceedings, 1976, 120-24.

Paul E. Green and Vithala R. Rao, Applied Multidimensional Scaling (New York: Holt, Rinehart and Winston, 1972).

Paul E. Green, Yoram Wind, and Arum K. Jain, "Preference Measurement of Item Collections," Journal of Marketing Research, 9 (November, 1972), 371-77.

Richard M. Johnson, "Trade-Off Analysis of Consumer Values," Journal of Marketing Research, 11 (May, 1974), 121-27.

David Krech and Richard S. Crutchfield, Theory and Problems of Social Psychology (New York: McGraw-Hill Book Co., 1948).

James H. Myers and Mark I. Alpert, "Determinant Buying Attitudes: Meaning and Measurement," Journal of Marketing, 32 (October, 1968), 13-20.

James H. Myers and Jonathan Gutman, "Validating Multi-Attribute Attitude Models," American Marketing Association Proceedings, 1974, 95-99.

Robert A. Peterson and Mark I. Alpert, "Consumer Choice Patterns and Psychographics: A Test of Predictability," American Institute for Decision Sciences Proceedings, 4 (November, 1972), 173-80.

Milton J. Rosenberg, "Cognitive Structure and Attitudinal Affect," Journal of Abnormal and Social Psychology, 53 (1956), 367-72.

Michael J. Ryan and Michael J. Etzel, "The Nature of Salient Outcomes and Referents in the Extended Model," Association for Consumer Research Proceedings, (1975).

Webster's New Twentieth Century Dictionary, 1961.

William L. Wilkie and Edgar A. Pessemier, "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, 10 (November, 1973), 428-41.

William L. Wilkie and Rolf P. Weinrich, "Effects of the Number and Types of Attributes Included in an Attitude Model: More is Not Better," Association for Consumer Research Proceedings, (1972), 325-40.

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