The Role of Affect in Categorization: Toward a Reconsideration of the Concept of Attitude

ABSTRACT - Affect (and attitude) have been viewed as the resultant of a set of analytical operations carried out on feature-based information using one or another learned combinatorial rule. This paper develop; an alternative conceptualization of the evaluation process that builds on recent work on nonanalytic concept identification and behaviorally functional categorization and which is also consistent with schema-theoretic concepts and analog-based imagery processes.

Citation:

Joel B. Cohen (1982) ,"The Role of Affect in Categorization: Toward a Reconsideration of the Concept of Attitude", in NA - Advances in Consumer Research Volume 09, eds. Andrew Mitchell, Ann Abor, MI : Association for Consumer Research, Pages: 94-100.

Advances in Consumer Research Volume 9, 1982      Pages 94-100

THE ROLE OF AFFECT IN CATEGORIZATION: TOWARD A RECONSIDERATION OF THE CONCEPT OF ATTITUDE

Joel B. Cohen, University of Florida

ABSTRACT -

Affect (and attitude) have been viewed as the resultant of a set of analytical operations carried out on feature-based information using one or another learned combinatorial rule. This paper develop; an alternative conceptualization of the evaluation process that builds on recent work on nonanalytic concept identification and behaviorally functional categorization and which is also consistent with schema-theoretic concepts and analog-based imagery processes.

INTRODUCTION

It is traditional in consumer behavior research to assume that a consumer's overall judgment and attitude toward a product are based upon implicit or explicit use of some information processing rule. Such rules (ant their paramorphic representations: linear compensatory, weighted average, conjunctive, etc.) are frequently compared and debated, but their basic premise (i.e. the computational nature of information integration) frequently goes unchallenged. It is important to note that the application of a computational rule may often give the right answer or even outperform the supposed rule user because such rules are so robust: people, we are not surprised to learn, tend to prefer products which possess desirable qualities and not those which possess undesirable qualities. Yet such rules may not reflect how the individual actually goes about arriving at his/ her overall evaluation of the Product.

We have accumulated a fairly substantial literature which shows that many people can, in fact, process information using a number of possible feature/attribute combinatorial rules, especially if we provide some training in their use and an indication that it is appropriate to use them. In addition, it's easy to dismiss as an inarticulate boob the focus group respondent who casually remarks that he/she grabbed that particular package because "it just seemed to be what I was looking for," all the while secretly smiling to ourselves that Nisbett and Wilson (1977) really had a point in underscoring people's lack of ability to report on internal processing operations.

There are, however, some serious challenges to the prevailing view that evaluation is merely the end result of a feature/attribute based information processing rule. This paper will explore some of these perspectives and will propose an alternative conceptualization of the evaluation process.

Some of the recent work on cognitive organization and representation and the processes of concept identification and categorization may be viewed as inconsistent with earlier associational and rule-based cognitive processes affecting judgment. These are each complex and subtle research domains, and any brief attempt to integrate these must, of necessity, focus on particular aspects of this work and ignore much more. With appropriate caution, therefore, let us examine the relevance of each to an orientation we shall loosely characterize as "judgment through evaluative frameworks."

SCHEMA THEORIES

Though the schema notion goes back at least as far as Bartlet (1932), its strong reemergence in recent cognitive and social psychology is ample tribute to the attractiveness of the concept. There are a great many schema theories and definitions. For our purposes it is useful to think of a schema as a hypothetical cognitive structure that integrates existing information into a more cohesive and directive organizational unit. Since the two key elements of this definition are that a schema is cohesive and directive, several cautions are in order. First, there is little evidence to suggest that individual memory traces are simply reorganized into larger schema-like structures and that their individual identity is then lost. It is also difficult to show unequivocally that activation of particular elements of an associative network would not provide an alternative explanation for many schema-like results.

The case for schema models as an accurate structural representation is not, however, crucial to this discussion. Much more important is the directive aspect of focused and organized cognitive structures (here termed schemas) on information processing. Such effects can be seen on attentional processes, on the encoding and integration of new information and with respect to subsequent inferences and memory reconstructions. Some of the early work on stereotyping (Allport and Postman, 1947), research on the role of implicit personality theory in impression formation, (Hastorf, Schneider and Polefka, 1970; Rosenberg and Sedlak, 1972) on the superior recall of information when a schematic context was presented (Bransford and Johnson, 1973; Smith, Adams and Schorr, 1979) and on the role of processing goals and tasks on perception and memory-based judgments (Lingle and Ostrom, 1979; Hamilton and Zanna, 1974; Cohen and Ebbesen, 1979) illustrate the directive role of memory schemata. In applying the schema notion to the concept of attitude, Tesser (1978, p. 297) writes: "There is not a single attitude toward an object but, rather, any number of attitudes depending on the number of schemas available for thinking about the object."

There are two aspects of this rich literature on cognitive schemas that are central to the topic of this paper. The first emphasizes the operation of integrated rather than discrete cognitive representations in information processing. The second focuses on the purposive and goal-directed nature of schemas, indeed on their readiness to respond to certain types and patterns of information. Schemas, in this latter sense, direct conscious attention to particular targets. And how are these targets selected and identified? Once a given schema is either activated or created, relevant research suggests that it functions as an organized whole. Thus, in the literature on person perception and impression formation such schemas (e.g. extravert, bleeding heart liberal, red neck bigot) are often treated as prototypes, and so, "It is probably the degree of 'family resemblance,' not the continuous surpassing of a few critical properties tests, that determines category membership in everyday person perception" (Cantor and Mischels 1979, p. 30).

When the activated schema is simply a framework instigated by a context or meaning cue, then the schema serves mainly as a perspective for attending to and interpreting events. When, however, the activated schema is instigated by personal goals and values or the need to take some action, in the broadest sense what we will be describing is judgment based on a type of matching process in which object categorization and evaluation are implicitly linked. To keep these types of schemas distinct, we shall refer to the latter schemas as evaluative frameworks. We will return to this conceptualization after discussing some other relevant perspectives.

THE ROLE OF MENTAL IMAGERY

If evaluative frameworks are presumed to function in this wholistic manner and to participate in a type of matching process, it would be reasonable to expect their memory representations to possess qualities that are consistent with such processing requirements. While both analytic and analog representational structures can, in general, be formulated to explain almost any particular research finding, there is a growing body of evidence to suggest that, whatever the deep representation of information may look like, an imagery system is an important aspect of a surface representation (Kosslyn, 1978). Kosslyn (Kosslyn and Pomeranti, 1977; Kosslyn, 1978) uses the metaphor of images being like displays on a cathode ray tube that are generated by a computer program. There is diverging opinion as to whether such an image is simply a byproduct of an underlying propositional structure or is, in fact, a representation that is decidedly more similar to that evoked when looking at a picture than when merely describing one (e.g., Anderson and Bower, 1973; Pylyshyn, 1973). Though it may not at first seem parsimonious to think of a representational system as having two different interpretive devices for images and propositions, such a "mint's eye" mechanism would appear to be needed to receive and hold perceptual images during information acquisition. In addition, a possibly more parsimonious structural account may be very much more complex in terms of the processes required to operate on propositional representations in explaining some key findings in the imagery area. Let us just highlight some pertinent findings which support a functional view of imagery and which thereby suggest that this type of representation may be particularly appropriate for the types of evaluative frameworks we are describing.

To help choose between imagery and associational explanations for the fact that it takes more time to mentally picture properties of smaller objects (e.g., the nose of a rabbit when imaged as standing next to an elephant than when imaged as standing next to a fly), Kosslyn (1976) found that people took more time to mentally picture smaller but more highly associated (in a propositional sense) defining properties than larger but more weakly associated properties. This finding was the reverse of that found under standard verification instructions which encouraged the use of propositional information. The results of a series of memory scan investigations (e.g., in one case people learned to draw a map of a mythical island and were then asked to image it) suggest that time to locate a given place on the map increased with distance from the focal point. This was not the case when people were asked to answer without being instructed to image the map. Jonides and Baum (1978) discuss this in terms of a "mental ruler" strategy as applied to the imaged representation of a map. Such results suggest that imaged representations are not only experienced but actually play a major role in information processing and judgment. Note that none of this research argues convincingly that images are stored in a rote form and are then simply replayed, like film through a projector. A deep representation could take several other forms (e.g. storage in a matrix format), though they must allow for sequential encoding and upgrading of an image over time, image transformations (e.g., zooming in on a particular property or rotating the image), as well as the enhancement or construction of an image from purely propositional information (e.g., a Florida "gator" leaning on the Sc. Louis Arch).

At this time it is simply not possible to conclusively demonstrate that any particular underlying representation is propositional or analog. A strong version of an analog representation would hold that there is a high degree of isomorphism between a visual image and what it represents while a weak version might be that only certain spatial information is retained in an analog representation with the remainder filled in by propositional information. Palmer (1978, p. 297) terms analog representations "intrinsic" since, "whatever structure is present in an analog representation exists by virtue of the inherent constraints within the representing world itself, without reference to the represented world." Unfortunately we cannot look inside a person's head to know what inherent constraints on analog representations exist. About the strongest assertion that can be mate is that imagery appears to selectively disrupt perception in the same modality (Kosslyn and Pomerantz, 1977), thus suggesting that there appears to be at least shared surface level processing.

While these are important representational issues, they are not central to the foregoing functional analysis of imagery in schema-based operations. It is more important to review some evidence on when images appear to be utilized. Clearly we to not need to construct or call upon a mental image to know whether or not Chevas Regal is a more expensive brand of Scotch than the discount house's surprise Scotch or whether a Cadillac El Dorado is more expensive than a Chevy Citation. It is reasonable to think of price as encoded propositionally (assuming the information has been encoded at all), in the sense that whatever relationships there may be among prices in the to be represented world can be extrinsically represented through a set of propositions that retain informational correspondence (but not isomorphism) with the represented world (see Palmer, 1978). In my own case, though, I believe I would have to try to picture a Chevrolet Malibu and a Monte Carlo to answer a similar question about price, since I do not know the prices but might be able to infer them from my overall conception of each car. What I think I would do is scan both images until I discovered aspects from which I could infer price differences. I certainly believe I would have to do so before deciding which is roomier or more luxurious or, in an overall sense, which I like more. Results from a series of size-comparison tasks, in which the helpfulness of propositional information was varied, are consistent with these introspections.

One question, then, is to what degree the kinds of schemas we have discussed earlier as evaluative frameworks rely mainly on propositional information and representations. To the extent they do, such schemas could certainly operate much like information processing rules or procedures and may be consistent with feature or attribute based processing algorithms. Hastie (1981) refers to this general type of schema as a "procedural schema," and it bears some relationship to Miller, Galanter and Pribram's (1960) concept of a plan as well as to various computer-based processing analogies. To the extent that the process relies more on a type of comparison between some overall or analog representation and sensory input (particularly when the system is more conceptually-driven), it may be poorly described and understood through the use of attribute/feature based processing algorithms. Much of the conceptualization and research relevant to this issue is centered in the concept identification and categorization literatures.

ANALYTIC CONCEPT IDENTIFICATION

Traditional concept identification research is more consistent with propositional representation. Recently, however, this evidence has come under attack from Medin and Schaffer (1978) and Brooks (1978). Some parallel developments can be seen in Rosch's (1978) work on principles of categorization and, within social psychology, in the work on person perception by Cantor and Mischel (1979). Let us begin by characterizing much of the early concept identification literature as interested in the types of generalities people would extract from various learning tasks. Much of this research has been carried out on novel and unfamiliar stimuli such as geometric shapes in combination with colors (for reviews see Dominowski, 1974; Bourne, 1974) and somewhat more recently on artificial grad ar (Reber, 1967, 1969). Brooks' recent review of this work indicates that, as the regularities in the material (i.e., natural grammars) become reasonably complex, subjects intent on learning a rule to define a concept simply cannot do so, given any sort of time constraints. Furthermore, there quickly become so many possible rule-based hypotheses to be retained and tested that subjects generally stop trying. Now, that poses no problem if we are simply interested in how people learn to discriminate relatively simple concepts which obey straightforward conjunctive or disjunctive rules, though even here there may well be serious experimental demand effects which themselves lead people to search for and test such rules. So, we cannot conclude that such behavior is to be expected or preferred by people in more natural contexts. Brooks (1978, p. 180) refers to this type of learning as analytic concept identification and defines it as, "a process whose direct effect is to separate aspects of the stimulus and evaluate their ability to predict category membership." What emerges from this type of learning, then, is a rule which is "an explicit or implicit summary of the aspects of the stimulus that is used to assess category membership of any item in which these aspects occur.)

CATEGORIZATION

Let us now briefly leave this part of our story having engendered some doubt as to whether analytic concept identification is very descriptive of what people actually do either in relatively complex learning situations or when they are simply trying to judge the identity of some novel housewarming gift or mystery beast. We will return to consider Brooks' alternative proposal shortly. In our 1979 ACR conference paper (Cohen, Miniard and Dickson, 1980) we advanced an active hypothesis testing approach to overall judgment. In essence it imagined a person generating a small number of plausible hypotheses (e.g., as to the identity of a mystery beast) as soon as some minimal amount of perceptual information was processed. This process is analogous to the standard recognition paradigm used within cognitive psychology in which perceptual representations are compared to memory representations on the basis of judged similarity. Identification is easy or difficult depending on how well learned and discriminable the memory representations are, the similarity of the perceptual input to one of the representations, the accessibility of a given category at the time as well as contextual factors such as the presence of distraction. We envisioned concept identification as an overall categorization of a stimulus resulting from attention to stimulus information judged most likely to discriminate the most salient alternative concepts/categories. Salience was conceived to be a joint function of stimulus cues and memory representations activated by each hypothesized concept.

Working from a different starting point, Rosch (1978) has advanced a set of principles to explain the set of categories found in a culture to describe natural objects. In doing so, she defines certain criteria for category membership, although her work does not directly address either processing or representational issues. One important premise is that through categorization we achieve an important degree of cognitive economy: "To categorize a stimulus means to consider it, for purposes of that categorization, not only equivalent to other stimuli in the same category but also different from stimuli not in that category" (Rosch, 1978, p. 28). It is to a person's advantage to be able to predict as many properties as possible simply by knowing the category membership of an object, but the large number of resulting categories (with fine discriminations between categories) would magnify behaviorally irrelevant differences. The meaningfulness of a given category should bear a close relationship to "cue validity," which Rosch defines as the validity of a given cue X as a predictor of a given category Y. This increases as the frequency with which cue X is associated with category Y increases and decreases as the frequency with which cue X is associated with categories other than Y increases.

The most relevant aspect of Rosch's work to the present discussion pertains to the assignment of objects to categories. She argues that categories do not have clear boundaries, Rather, a better way of achieving a separation and clarity among related categories is by conceiving of each category in terms of its clear cases and not its boundaries. Rosch refers to such clear cases as prototypes, though there is no attempt to link this concept to any particular model of information processing, representation or learning. Nevertheless, ratings of prototypicality follow closely the cue validity notion: the more prototypical the category member the more attributes it has in common with other members of the category and the fewer attributes it has in common with members of contrasting categories. In addition, prototypical category members tend to represent the means of attributes that have a metric. Research carried out on categorization indicates that the greater the similarity of a target object to a category prototype the faster it was identified, the more quickly children were able to learn to categorize it, the more frequently and earlier it was produced when subjects were asked to list members of a category, and the more frequently it is equated with the category of which it is a member (e.g., Robin or sparrow but not turkey or penguin, may sometimes be substituted for the category "bird" in sentences without greatly altering the meaning or the response to the sentence).

One way of looking at concept learning, then, is in terms of the development and differentiation of a cognitive category. Concept identification, similarly,may be seen as the placement of an object in a cognitive category. One way of carrying out these operations is to develop or discover the rules that define category membership by focusing on the features/attributes of the category, considering these one by one in a feature-by-feature comparison with respect to a given target. The features thus constitute necessary and sufficient conditions for category membership. This set of operations does not, however, account for the wide differences in perceived degree of typicality among objects and, therefore, does not address the range of within-category variability. A prototype matching strategy may be somewhat different in the sense that rather than attribute-by-attribute "pass-fail" decisions, this approach considers the degree to which the target is similar to a prototypical representation of the category. The prototypical representation, however, still is likely to be defined in terms of features, though more of an overall matching process is implied (Posner and Keele, 1970), possibly through the use of weighted features (e.g., the defining/characteristic attribute distinction made by Smith, Shoben and Rips, 1974) or some multidimensional index of similarity.

A somewhat different category membership criterion may be thought of as "family resemblances." Many things that people group together have no obvious similarity to one another across a set of attributes. Wittgenstein's identification of the category "game" (discussed in Glass, Holyoak and Santa, 1979) provides a nice illustration of this, since professional football, chess, golf and solitaire really have no obvious defining set of attributes. Instead, each separate instance of the category "game" appears to be linked by some key feature to another recognized instance until they comprise a type of family in which each member has key elements in common with at least one other member.

The family resemblance notion is somewhat disturbing to both feature-by-feature comparison and prototype matching conceptualizations of categorization. This is not because categories such as "game," "rule," "friend," and the like are necessarily typical, but rather because there is the implication that categorization and more generally, concept learning may follow a very different path. Recall the earlier cited conclusion by Cantor and Mischel (1979) in reviewing the person perception area, in which they stress that family resemblance rather than comparisons on a set of defining attributes seemed to determine category membership when categorizing other people. Cantor and Mischel, however, prefer to describe the categorization process through the use of the prototype construct, stressing the "global configuration" aspects of prototype matching. In doing so they do not appear to have satisfactorily dealt with what may be an irreconcilable difference between a prototype model and one based on the family resemblance notion. That is, it does not seem to be the case that one would categorize separate game instances together by comparing each to some overall category prototype. Indeed, the point is that such a prototype may be an idealized abstraction and may not even exist in the memory structure of the individual.

NONANALYTIC CONCEPT IDENTIFICATION

We return now to an approach Cantor and Mischel term "promising" and which is the thrust of Brooks' proposal. Brooks, it will be recalled, approaches the topic a little differently, since he begins with the question of how concepts are learned in the first place. This seems a sensible place to start to appreciate category formation (and possibly some representational issues as well) and how we assign instances to categories. In distinction to the earlier described processes of analytic concept identification (with its emphasis on rule learning), nonanalytic concept identification, according to Brooks, leads to an inference of category membership based on an object's "overall similarity to a known individual or low-level cluster of individuals." Unlike prototype or defining features approaches, stimulus aspects are not weighted for their criteriality with respect to the concept being considered. Using Brooks' "mystery beast" example, a person would be using nonanalytic concept identification if he decided that the beast was a dog and not a cat because it looked a lot more like Lassie than Fluff. If, instead, the mystery beast was categorized as a dog because it possessed specific canine features and because it did not possess retractable claws, and other cat-like properties, this would be an example of analytic concept identification. Note that nonanalytic is not equated with configural, since the use of overall body shape or other configural aspects involves a weighting of such attributes in a similar fashion to other object features in the development of an analytic rule for category membership.

The essence of nonanalytic concept identification is the learning and storage of an adequate number of specific instances. Learning idiosyncratic information is itself highly rewarded, since we do not behave in relation to categories of things (e.g., togs) but to particular instances (e.g., we decide to pet this particular dog). Abstraction and generalization may be both efficient and economical, but these are overlaid on memory for specific instances, which are then available as exemplars. In addition, memory for particular exemplars is often very rich and well developed since in-depth exposure to a few individuals in a category rather than feature-based comparisons on many individuals appears to be the norm. This means that we are likely to know a fair amount about a few instances and are likely to regard that information as particularly reliable. When judgments regarding a possible instance of one or another concept are to be made, often under time pressure and given incomplete information, specific category exemplars may play a dominant role. When we encounter instances that are clear cases of a category, i judgment may be almost instantaneous, and operationally, there may be little to choose among alternative analytic and nonanalytic concept identification processes in terms of accuracy or speed of performance. As we approach the boundaries of the category, however, we are likely to achieve better discrimination by comparing the instance to exemplars of each category than by either attempting to employ an analytic rule which summarizes stimulus properties of a given category or by comparing the instance to more idealized prototypical representations of each category. Thus, while it is usually the case that a number of somewhat different category exemplars may be available in memory to help in categorizing an object (e.g., Lassie,the poodle down the street, aa Irish Setter) the utility of a central prototype in discriminating, say, a wolf from an Alaskan Malamute or Siberian Husky is not self evident. Even in those cases in which individuals have learned appropriate rules (e.g., English language pronunciation), , Brooks argues that they are more likely to pronounce a F list of novel words by analogy to specific known words.

CONVERGING PERSPECTIVES

Throughout the paper there is at least the suggestion that the perspectives we have discussed are converging. It is now appropriate to clarify the sense of this implication and to indicate more specifically how this might be the case. First, we do not mean to imply that the evidence now points to one dominant form of memory representation or method of learning concepts or categorizing objects. Instead, the evidence suggests a high degree of flexibility in cognitive processes and operations such that, for example, either analytic concept identification or nonanalytic concept identification may be employed to fit individual and environmental circumstances. Convergence is used here in an overlapping sets sense. Specifically, certain types of schemata and the use of mental imagery appear to be more likely correlates of the types of category representations used in nonanalytic object judgments.

The basic argument runs as follows. We initially learn what most things are like through exposure to individual instances of them (e.g., the small child sees his father point at a robin and utter the word "bird.") or through learning about the specific experiences of others. These concrete and vivid instances are particularly amenable to imagery. Since it is very helpful to construct a meaningful and efficient organizational structure with respect to the objects and events that concern us, we begin to construct categories for similar instances. In order to determine that a new instance fits one category better than another, the perceptual representation of the new instance can be compared with the images of category exemplars (at first especially instances around which a category may be said to have developed). This would allow us to make at least a first cut. We can then "zoom in" on particular features through the use of both analog and analytical (i.e. propositions stored in an associative network) representations if we have the time and inclination and there is sufficient need to do so. These categories, once formed, allow us to predict sets of properties, object uses, and person behavior on the basis of our assignment of such objects and people to them, without directly observing such aspects for each instance (Higgins et. al., 1977; Srull and Wyer, 1979; 1980). The set of beliefs and expectations we have about objects or people residing in a given category may be said to comprise a schema for that category. Through the use of such schemas we infer missing information about particular category members, and this is manifest both in responding to and encoding incoming information, in organizing such information in memory and in reconstructing memory representations. For many natural objects, the shared experiences of a culture (not to mention cross-cultural constancies in sensory structures and processes) and the widespread presence of correlated attributes (rather than random assignment of features and attributes to individual instances) leads to considerable uniformity in category structure and membership.

When, however, there exists much greater variance in experience, tastes and purpose, and properties of objects are not fixed, one would expect different sets of categories to emerge and there to be different object placements across the population. Under these conditions, the particular instances that become category exemplars are crucial to the person's subsequent judgments regarding the appropriateness of any newly presented instance for the particular category. This latter state of affairs may pertain farily well to the marketplace for goods and services and to consumer judgments about them. If this is so then it may be useful to explore the relationship between such notions and concepts in the consumer behavior literature such as "evoked sets" (Howard and Sheth, 1969) and particular marketing and advertising practices (e.g., developing a "unique selling proposition" for each brand). The former concept may refer specifically to category membership and the letter to an exemplar based strategy for defining a particularly narrow category.

THE ROLE OF AFFECT IN CATEGORIZATION

Before developing this position a little further, it is important to specifically address the role of affect in such a process. There is an intriguing quote from Zajonc (1980) that is relevant here.

The stimulus features that serve us so well in discriminating, recognizing, and categorizing objects and events may not be useful at all in evaluating these objects. If this is indeed the case, then there must exist a class of features that can combine more readily with affect and thereby allow us to make these evaluations, to experience attraction, repulsion, pleasure, conflict and other forms of affect, and to allow us to have these affective reactions quite early after the onset of the sensors input (p. 159).

Rather than seeking affect in a unique and separate class of features (termed "preferenda"), perhaps we should treat affect as residing in the category itself. Earlier in the paper we described a class of schemata that we characterized as instigated by personal goals and values or the need to take some action as evaluative frameworks. Such evaluative framework schemata may be the outgrowth of categories that develop in response to behavioral demands and subsequent learning. After all, a key premise underlying the hypothesized existence of categories in the first place is that they are functional: they help us to organize and cope with the environment. We don't simply organize the environment out of idle curiosity. As Jones and Gerard put it (1967, p. 185), "There is considerable uncertainty involved in predicting outcomes and the individual tries to reduce this uncertainty by information-seeking and to deny it by moving toward an unequivocal behavior orientation." Developing behaviorally functional categories (rather than merely mental dictionaries), such that objects grouped together meet a common purpose would be a useful way to organize one's environment. To achieve such a purpose, however, categories should not be neutral but inherently evaluative. (Recall that the single most important factor underlying the meaning of concepts is their evaluation -- how good or bad, valuable or worthless -- Osgood et. al., 1957).

Similarly, attitudes were conceived to be predispositions to respond in a favorable or unfavorable manner, and to exert a directive influence upon the individual's responses to objects and situations with which they are linked. To have an attitude toward an object, then, requires; first, that the object be recognized as an instance of a previously defined category, ant, second, that the category possess a positive or negative value to the individual. Learning theory models of attitude (e.g. the Fishbein model) and related multiattribute formulations, in essence, attach affect to particular object features and by use of a combinatorial rule arrive at both an identification of and an overall attitude toward the object. The parallel with analytic concept identification processes is straightforward (Rhine, 1958). We have argued, however, that the use of such rules, particularly in natural settings, under time and capacity constraints and when clear category exemplars are available in memory from which to base an overall judgment should not be assumed. The role of advertising in creating product exemplars and the role of in-store promotions and packaging in priming these would appear to make this especially the case for many consumer products. In essence, then, matching an object directly against exemplars of plausible alternative categories not only provides an efficient means of identifying it but also evaluating it. To the extent that this is a particularly important judgment or when a lack of experience or stimulus ambiguity make it difficult to confidently compare the object to an exemplar, the person may try to develop or apply a feature-based categorization rule. The prevalence of analytic and nonanalytic processes in consumer behavior would appear to be an important empirical question.

In either case it seems reasonable to hypothesize that evaluative concepts may help define membership in a particular category. Thus, instead of thinking only in terms of descriptive product categories (e.g., cars, detergent, fast foot restaurants) we might be well advised to think about categories that incorporate affect (e.g., affordable cars, detergents that clean well and fast foot restaurants I'd eat at). Such categories would be at different levels of evaluation abstraction as a function of the judgments a person is called upon to make. They should, however, evolve according to the same principles advanced earlier.

In the case of a consumer product, the number of categories and the nature of each category should bear a strong relationship to the diversity of experiences with particular instances. One consumer may categorize Ill fast foot restaurants together in guiding his/her behavior. Another may have a number of subordinate categories each developed around a particular exemplar. When a new instance of a fast foot restaurant appears, it should be matched against the exemplars of the available categories to determine how to think about it, feel about it and behave towards it. The role of category-based schemas in filling in missing information about a particular instance has already been discussed.

A related point of view has developed in the person perception area (see Hastie et. al., 1980, for a more inclusive treatment of this literature). Categorizing people and situations not only helps to organize one'< environment, it helps a person to plan his/her social interactions (Snyder and Cantor, 1979); Once we have "located" a particular instance as being a member of a certain class of situations, we can imagine and often picture how exemplars of various person categories would behave, and we can use this information to guide our own behavior.

There are two principal areas of application of the principles discussed in this paper within consumer behavior. The first would be to much more fully develop the information processing characteristics of an evaluative framework approach, and the second would be to actually focus on the nature and content of categories used by consumers in particular domains or under different conditions. For example, to what degree do differences in experiences (i.e. product knowledge and satisfaction) across brands lead to differences in the number and affective aspects of the categories used by a consumer to describe a product domain and to categorize new instances? Rosch's work on natural categories would be a reasonable place to begin the latter endeavor.

The information processing differences between the present approach and models built on the premises of analytic concept identification are far from trivial. It would be crucial to know and understand the conditions under which each was used. For example, it would be quite helpful to explore such things as the nature, development and role of category exemplars, the importance of category-based imagery, differences in both processing and outcomes when single versus contrast categories are used, the effects of "priming" particular categories or exemplars and, of course, the implications of making evaluative judgments (both the process and resulting outcomes) based on particular matching or overall similarity criteria for consumers, marketers and public policy makers.

In summary, we have, for a long time, treated affect (and of course attitude) as the resultant of a set of analytical operations performed on information using one or another fairly abstract rules. It may now be important to advance the twin propositions that nonanalytic and fairly concrete processing operations may be quite prevalent and that affect is an important defining property of the cognitive categories to which we assign personally relevant environmental stimuli.

REFERENCES

Allport, G. W.,, and Postman, L. (1947), "The Psychology of Rumor." (New York: Holt).

Anderson, J. R., and Bower, G. H. (1973), Human Associative Memory (Washington, DC: Winston).

Bartlet, F. C. (1932), Remembering: A Study in Experimental and Social Psychology (London: Cambridge University Press).

Bourne, L. E. (1974), "An Inference Model For Conceptual Rule Learning, (ed.) R. L. Solso, in Theories in Cognitive Psychology, (Potomac, MD: Erlbaum).

Bransford, J. D., and Johnson, M. R. (1973), "Considerations of Some Problems of Comprehension, (ed.) W. Chase, in Visual Information Processing (New York: Academic Press).

Brooks, L. (1978), "NonanAlytic Concept Formation and Memory For Instances, (eds.) E. Rosch, and B. B. Lloyd, in Cognition and Categorization (Erlbaum).

Cantor, N., and Mischel, W. (1979), "Categorization Processes In the Perception of People, (ed.) L. Berkowitz, in Advances in Experimental Social Psychology, Vol. 12, (New York: Academic Press).

Cohen, C. E., and Ebbesen (1979), "Observational Goals and Schema Activation: A Theoretical Framework for Behavior Perception, Journal of Experimental Social Psychology, 15, pp. 305-329.

Cohen, J. B., Miniard, P. W., Dickson, P. R. (1980), "Information Integration: An Information Processing Perspective," (ed.), J. C. Olson, in Advances in Consumer Research, Vol. 7. (Ann Arbor: Association for Consumer Research).

Dominowski, R. L. (1974), "Row do People Discover Concepts?" (ed.) Solso, R. L., in Theories in Cognitive Psychology, (Potomac, MD: Erlbaum).

Glass, A. L., Holyoak, K. J., and Santa, J. L (1979), Cognition, (Reading, MA: Addison-Wesley).

Hamilton, D. L., and Zanna, H. P. (1974), "Context Effects in Impression Formation: Changes in Connotative Meaning." Journal of Personality and Social Psychology, 29, pp. 649-654.

Hastie, R., Ostrom, T. H., Ebbesen, E. B., Wyer, R. S., Hamilton, D. L., and Carlston, D. E. (1980), Person Memory: The Cognitive Basis of Social Perception. (Hillsdale. NJ: Erlbaum).

Hastie, R. (1981), "Schematic Principles In Human Memory." (eds.) E. T. Higgins, C. P. Herman, and M. P. Zanna, Social Cognition: The Ontario Symposium, (Hillsdale, NJ: Erlbaum).

Hastorf, A., Schneider, D., and Polefka, J. (1970), Person Perception, (Menlo Park CA: Addison-Wesley).

Higgins, E. T., Rholes, W. J., and Jones, C. R. (1977), "Category Accessibility and Impression Formation," Journal of Experimental Social Psychology, 13, pp. 141-154.

Howard, J. A., and Sheth, J. N. (1966), The Theory of Buyer Behavior, (New York: Wiley).

Jones, E. E., and Gerard, H. B (1967), Foundations of Social Psychology (New York: Wiley).

Jonides, J., and Baum, D. R. (1978), "Cognitive Maps As Revealed by Distance Estimates. (Chicago: Paper Presented At The Meeting Of The Midwestern Psychological Association).

Kosslyn, S. M. (1976), "Can Imagery Be Distinguished From Other Forms Of Internal Representation? Evidence From Studies Of Information Retrieval Tine." Memory and Cognition, 4, pp. 291-297.

Kosslyn, S. M., and Pomerantz, J. R. (1977), "Imagery, Propositions, and The Form of Internal Representations." Cognitive Psychology, 9, pp. 52-76.

Kosslyn, S. M., (1978), "Imagery and Internal Representation." (eds.), E. Rosch, and B. B. Lloyd, Cognition and Categorization, (Erlbaum).

Lingle, J. R., and Ostrom, T. H. (1979), "Retrieval Selectivity In Memory-based Judgments." Journal of Personality and Social Psychology, 31, pp. 180-194.

Medin, D. L., and Schaffer, M. M. (1978), "Context Theory of Classification Learning." Psychological Review, 85, pp. 207-328.

Miller, G. A., Galanter, E., and Pribram, R. (1960), Plans and the Structure of Behavior, (New York: Holt, Rinehart, and Winston).

Nisbett, R. E., and Wilson, T. D. (1977), "Telling More Than We Know: Verbal Reports on Mental Process." Psychological Review, 84, pp. 231-259.

Osgood, C. E., Suci, G. J., and Tannenbaum, P. H. (1957), The Measurement of Meaning, (Urbana, IL: University of Illinois Press).

Palmer, S. E. (1978), "Fundamental Aspects of Cognitive Representation." (eds.), E. Rosch, and B. B. Lloyd, Cognition and Categorization, (Erlbaum).

Posner, Mc I., and Keele, S. W. (1970), "Retention of Abstract Ides s." Journal of Experiments 1 Psychology, 83, pp. 304-308.

Pylyshyn, Z. W. (1973), "What The Mind's Eye Tells The Mind's Brain: A Critique of Mental Imagery." Psychological Bulletin, 80, pp. 1-24.

Reber, A. S. (1967), "Implicit Learning of Artificial Grammars." Journal of Verbal Learning and Verbal Behavior, 6, pp. 855-863.

Reber, A. S. (1969),"Transfer of Syntactic Structure in Synthetic Languages." Journal of Experimental Psychology, 81, pp. 115-119.

Rhine, R. J. (1958), "A Concept Formation Approach to Attitude Acquisition." Psychological Review, 65, pp. 362-370.

Rosch, E. (1978), "Principles of Categorization." (eds.) E. Rosch, and B. B. Lloyd, in Cognition and Categorization, (Hillsdale, NJ: Erlbaum).

Rosenberg, S., and Sedlak, A. (1972), "Structural Representations of Perceived Personality Trait Relationships." (eds.) A. R. Romney, R. Shepard, and S. B. Nerlove, in Multidimensional Scaling, (New York: Seminar Press).

Smith, E. E., Adams, N., and Schorr, D. (1978), "Fact Retrieval and the Paradox of Interference." Cognitive Psychology, 10, pp. 438-464.

Smith, E. E., Shoben, E. J., and Rips, L. J. (1974), "Structure and Process in Semantic Memory: A Featural Model for Semantic Decisions, Psychological Review, 81, pp. 214-241.

Snyder, M., and Cantor, N. (1979), "Testing Hypotheses About Other People: The Use Of Historical Knowledge." Journal of ExPerimental Social Psychology, 15, pp. 330-342.

Srull, T. R. and Wyer, R. S. (1979), "The Role of Category Accessibility in the Interpretation of Information About Persons: Some Determinants and Implications. " Journal of Personality and Social Psychology, 37, pp. 1660-1672.

Srull, T. R. (1980), "Category Accessibility and Social Perception: Some Implications For The Study of Person Memory and Interpersonal Judgments." Journal of Personality and Social Psychology, 38, pp. 841-856.

Tesser, A. (1978), "Self-generated Attitude Change." (ed.), D. Berkowitz, in Advances in ExPerimental Social Psychology, Vol. 2, (New York: Academic Press).

Zajonc, R. B. (1980), "Feeling and Thinking: Preferences Need No Inferences." American Psychologist, 35, pp. 151-175.

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