Hierarchical Cognitive Content: Towards a Measurement Methodology

Citation:

Elizabeth C. Hirschman and Susan P. Douglas (1981) ,"Hierarchical Cognitive Content: Towards a Measurement Methodology", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 100-105.

Advances in Consumer Research Volume 8, 1981      Pages 100-105

HIERARCHICAL COGNITIVE CONTENT: TOWARDS A MEASUREMENT METHODOLOGY

Elizabeth C. Hirschman, New York University

Susan P. Douglas, New York University

INTRODUCTION

In recent years, increasing interest has arisen in developing methodologies for measuring cognitive phenomena. In particular, growing use has been made of information processing approaches to understand consumer decision making (Bettman 1980), product perception (Olson 1975) and attitude structure (Capon and Lutz 1979). Attention has thus been drawn to identifying what information relating to Products is used by consumers, and how that information is organized in memory.

Research has focused primarily on examining product attribute information, and on assessing how this is used in evaluating brands and products. Use has been made of protocol techniques (for a summary see, Douglas, Craig and Faivre 1980) or less commonly free recall prompted by an external cue or cover story (Russo and Johnson 1980, Olson and Muderrisoglu 1977). This research has, however, largely ignored the broader semantic network surrounding the product, namely product-variant linkages to form categories or product classes and brand-product hierarchical linkages to form general to specific orderings of content (Lachman, Lachman, Butterfield l979).

One exception to this is the work of Gutman (1980) which has explored the broadness/narrowness of categories created by consumers from given product stimuli. The hierarchical character of product-brand-attribute categories has, however been largely ignored.

For example, whether the consumer uses the product-concept primarily as a node (storage area) for accumulating and organizing other types of information, or as an item of knowledge moderating (i.e. qualifying, or describing) category. Take, for example, the concepts "Levi's" and "jeans". For some consumers, the primary cognitive category is "jeans", for which Levi's is a moderator (i.e. qualifier). For other consumers "Levi's" is the node (category), and "jeans" is a moderator.

Such information is, nonetheless, of interest to assess how consumers categorize and classify, products, product classes, brands and domains, and how this information is maintained in memory. It might also be useful in developing advertising copy by indicating the scope of the semantic image, i.e. the combinations of products, and brands, which are stored together by consumers.

An issue which arises in this context is the appropriate methodology to examine the hierarchical content of information in memory. This paper reports an attempt to apply free-association techniques to examine cognitive content and, given the limitations of this methodology, to suggest an alternative approach.

FREE ASSOCIATION METHODOLOGIES

Free response and word association techniques are widely used in cognitive psychology (Rosch l978) and in cross-cultural research (Szalay and Deese 1978, Rosenzweig 1961) to identify domains, examine product categorization, and the meanings associated with products and objects and to compare are these across cultures. Their use in marketing has been somewhat more limited, largely due to difficulties and tediousness of analysis (Green, Wind and Jain 1973) The advantage of using consumers' free associations to a product concept, as opposed to the multi-attribute approach, is that all types of semantic associations regarding a product can be measured, not just the attributes. Hence, the measurement of consumers' product cognitions is extended beyond attribute structure to the total semantic associational structure.

The free association technique can, for example, provide valuable information about brands perceived as competing. For example, in response to the stimulus brand "McDonalds", responses of "Burger King" and "Wendy's", may be obtained, indicating that these are perceived as similar and hence potential competitors, similarly, in response to "jeans" the responses "Levi's", "Gloria Vanderbilt", "Sasson" and "Jordache" may be obtained. These, therefore, represent the consumer's evoked set for this product category. Finally, free associations may provide information on the uses to which a product/brand could he put by the consumer. For example, in response to the stimulus product "shoes", associations such as "running", "soccer", "basketball", and "jogging" may be obtained. These show both the degree of differentiation within the product category perceived by the consumer and also the possible activities for which s/he might utilize the product (i.e. shoes).

EXTENSION TO A COGNITIVE HIERARCHY

Free association techniques have been widely used in psychology to examine the hierarchical structure of information content. This perhaps best exemplified by the work of psychologist Eleanor Rosch. Rosch (1973, 1971, 1975, 1978) maintains that concepts may be arranged along a continuum from abstraction/generality to concreteness/specificity. Three levels along this continuum are identified: (1) a superordinate level of the most general or abstract concepts (e.g. furniture), (2) a basic level of more concrete/specific concepts (such as chair) and (3) a subordinate level of most concrete/specific concepts (such as dining room chair).

Rosch has focused primarily or studying horizontal categorization at each level of the category by asking respondents to indicate attributes associated with the level. Here, a key finding has been that the number of attributes associated with a concept decreases as a function of generality. Thus, fewer attributes are cited at higher levels of abstraction in the hierarchy. This does not, however, imply that the total number of cognitive associations is lower for more general concepts. Rather the character of these associations changes, with concrete nouns instead of attributes being associated with more general concepts.

This was demonstrated in consumer research by Hirschman (1980). Free association techniques revealed no relation between the absolute number of word associations to a product and its level in the hierarchy. Four levels were examined: domain-level, (entertainment), type-level (movie), generic-level, (adventure movie) and brand-level (James Bond movie). The associations made at the general domain level (entertainment) tended to be predominantly concrete nouns (e.g. a movie); whereas the associations made at the specific generic and brand levels were more likely to be descriptive qualifiers and attributes (e.g. soft, delicious). This is shown in Table 1. Thus, use of a free response work association approach not only confirms Rosch's finding that more attributes were associated with more specific conceits, but also demonstrated that different types of associational content (e.g. nouns were associated with different levels).

TABLE 1

MOST FREQUENTLY OCCURRING RESPONSES: DOMAIN OF FOOD

The use of a free association approach in examining hierarchical cognitive content has, however, some disadvantages. One inherent disadvantage of free association data is that they provide no information on the nature of the association, and, in particular, on the propositional relationship of an evoked response to the product stimulus. For example, does the association of "sneaker" and "running" mean that a sneaker is seen as a running shoe, or that sneakers are used in running?

A further disadvantage is that a stimulus list (i.e. set of product/brand names) has to be developed a priori and arranged in a specific hierarchy, For example, in one experiment, the following hierarchy was selected: Apparel, Pants, Jeans, and Levi's Jeans. This set was then presented (along with several other sets in randomized order) to the respondent and words associated with each level were recorded. In analyzing the data, it was assumed that this provided an appropriate general-to-specific ordering.

This assumption, although it appears logical and possesses "face validity" may, however, not be appropriate for some or all consumers or for some or all product classifications. A final, and perhaps most critical limitation of the free association technique in studying cognitive hierarchies, was that the associations tended to be predominantly horizontal rather than vertical. This was particularly the case in relation to the more specific concepts where modifiers were commonly provided. Consequently, while it has initially been hoped to capture more of the vertical semantic network surrounding the concepts, this was only partially achieved, with the stimulus set used.

THE HIRSCHMAN-DOUGLAS CARD SORTING TASK

Given the limitations of the word association task, an alternative procedure was developed for examining hierarchical cognitive structure. This procedure, termed the Hirschman-Douglas Card Sorting Task (or H-D CST), consists of listing on cards the names of products/trends hypothesized to represent different levels of generality-specificity in a given consumption domain. These cards are then given to subjects who are asked to sort them into piles or groups based on perceived similarity.

In a trial experiment of this method, subjects were provided with six product concepts from two of three different consumption domains (e.g. food, clothing and entertainment). The six product concepts within each dot, in were selected a priori from three hierarchical levels: Type (television, meat), Generic (talk show, hamburger), and Brand (Johnny Carson, McDonalds). Five pairs of adjectives or modifiers were also included. Three of these were the pairs found by Osgood to be universal qualifiers of meanings (Osgood, May and Miron, 1975). Two more product specific adjectives, i.e. cheap/expensive, and hard/soft, were added. Thus, the subject was given a total of 32 cards representing products/brands from two different consumption domains and five pairs of adjectives. A complete listing of all stimuli is given in Table 2.

In performing the task, subjects were instructed first to look at the cards and then arrange them into "groups that make sense to you" or "piles of things that are closest." Subjects were told that they could make as many or as few groups as they wanted and that there was no time limit and no right or wrong answer to their response. Half of the subjects were required to perform the task twice, being presented first, with one combination of two domains, e.g. food and clothing and then another, e.g. food and entertainment. These were systematically rotated to test for order or domain effects.

A convenience sample (n = 40) was used for the trial experiment and subjects were drawn from a variety of socioeconomic backgrounds: students, secretaries, guards, professors, businessmen and so forth. The sample is, therefore, not representative of any population. It does, however, provide sufficient variety to make some comments on the methodology.

The first general observation is that the task elicited a high level of involvement and interest on the part of the subjects. In contrast to the disinterest and fatigue often displayed by subjects responding to a questionnaire, the card sorting task generated enthusiasm. The majority of subjects stated that they found the task to be "fun" and "interesting" and completed it in five to fifteen minutes. Thus the recruitment and involvement of subjects does not appear to be a problem with this particular data-gathering technique.

RESULTS

These data were then analyzed by three judges to identify the patterns used by subjects in categorizing the stimulus set. Two major dimensions were identified: 1) whether subjects organized the cards vertically or horizontally, and 2) the degree of tightness, complexity or order subjects used to form the groupings. A high degree of reliability was obtained in the analysis, disagreement arising in only 5 out of 60 cases.

A key finding is that many people do appear to use a hierarchical framework in organizing their knowledge within consumption, domains, though not the tidy and neat structure assumed by the Rosch model. The majority, of subjects (80%) recovered the hierarchies in the stimulus set. They differed, however, in the extent that to which they added adjectives or pairs of adjectives to qualify hierarchies (Figure 1).

TABLE 2

PRODUCT CLASS STIMULI AND QUALIFYING PAIRS OF ADJECTIVES

FIGURE 1

A VERTICAL BRAND-ORDERED PATTERN

FIGURE 2

A HORIZONTAL ORDERED PATTERN

FIGURE 3

A VERTICAL TYPE ORDERED PATTERN

Some subjects (19%), particularly those who expressed low involvement with the stimulus set, grouped the cards into three piles - adjectives, brand names and products (Figure 2).

Another aspect was how the hierarchies were organized. Some subjects adopted a predominantly brand-organized all headed by a brand-level concept. Others adopted a type-level structure. This is shown in Figure 3. Within each grouping the Rosch-like hierarchical structure is maintained (i.e., Type, Generic, Brand) with only one interesting exception: drink, Maxwell House, coffee. Some subjects declared themselves unable to order within piles. This occurred primarily when somewhat complex structures were developed initially.

The other dimension - tightness vs. looseness - refers to how highly organized the structure created by the subject vas. A very "tight" structure was one in which all the cards were grouped into a maximum of three piles (10%) (Figure 4). This occurred predominantly where subjects appeared to have a strong need for order. 16% had moderately tight structure i.e. four to five piles. The majority (73%), adopted a loose structure in which several cards were grouped into pairs (representing a two-level, or binary hierarchy) and/or were left unrelated to one another (that is, were "grouped" as individual units).

FIGURE 4

A TIGHT STRUCTURE

A critical assumption underlying this procedure is that the groupings created by the subject with the cards correspond to the way information is structured in his/her memory. Whether this assumption is valid or not remains open to some question. However, subjects did not appear to have any difficulty with the task, or find it artificial, and the differences in the resulting structures suggest that the assumption may be appropriate.

It should be noted that the card sorting task methodology is still in the early stages of development. Further work in devising appropriate stimulus sets is required. The stimulus hierarchies were arbitrarily defined by the researchers and imposed on the subject. It is therefore difficult to assess how much of the structure obtained is a result simply of the specific stimulus set. One alternative which might partially eliminate this bias is to provide the subject with a more ambiguous and complex set of words from varied domains of consumption. If the task is repeated for several domains the consistency of organizational patterns can be examined. Furthermore, the same set of words can be given at various time intervals to measure the stability of response.

The card sorting cask may not be appropriate for all consumers. Tests conducted with a number of children indicated that children do not respond to the stimuli in the same way as adults. Children, for instance, tended to respond to the words not as representing products, but rather as semantic entities. For example, a typical child's approach to sorting the cards is to alphabetize the words, or to group them into piles according to word length, number of words on the card, and so forth.

This is consistent with studies of cognitive development (Gelman 1978) which indicate that younger children are unable to reason abstractly using semantic stimuli or handle hierarchical classification, because they have not arrived at the level of what Pisser terns symbolic operations (Piaget 1976). Hence, they do not perceive the words as representing products, but rather as semantic entities. [The card sorting cask may be similarly inappropriate for psychometricians, consumer researchers and marketing model builders. For example, one well-known marketing model-builder who, while complaining that it vas trivial and meaningless, proceeded to construct a geometric configuration atop his desk with precise euclidean distances separating the cards. Unfortunately, the author-administrator neglected to copy down the resulting multi-dimensional cognitive map. On the other hand, perhaps we are better off left in ignorance.]

ADDITIONAL APPLICATIONS OF THE FREE ASSOCIATION AND CARD SORTING TECHNIQUES

Free Association

The type of free association techniques discussed here also have a number of other potential applications. Perhaps one of the most valuable applications is in the area of subcultural and cross-cultural consumer research (Wind and Douglas 1980). The free association approach is essentially an unstructured and hence predominantly emic method for data gathering and analysis (e.g. Szalay and Deese 1978, Pike 1966, Ember 1977, Triandis 1972). [The "emic" and the "etic" distinction is an important one in cross-national and cross-cultural research (Ember 1977, Pike 1954, Triandis 1972). The "emit" school holds that attitudinal or behavioral phenomena are expressed in a unique way in each culture, and are best understood in their own terms. Following this point of view, measures specifically adapted to each cultural context will be required. The "etic" school, on the other hand, is primarily concerned with identifying universal or "culture-free" measures. This facilitates comparisons but potentially leads to loss of some precision or accuracy of measurement in each national context (Elder 1976, Przeworski and Teune 1966/7, 1970).] It does not imply the imposition of a construct identified in one culture on another culture, where it may not be relevant. Rather it allows consumers in various cultures or subcultures to respond freely in their own terms with whatever verbal or written signs they feel most appropriate.

Some cultural self-referrent bias (Lee 1966) my however be introduced by the researcher given his specific cultural referrent in the choice of the stimuli and in interpretation of the data. Hence, careful attention to "decentering" the stimulus let will be required (Warner and Campbell 1973). In addition, since the onus of interpretation is placed on the researcher in determining where there are similarities and differences between countries, use of multiple judges from different cultures will be desirable.

One potential use of free association data in this context is to compute "shared meaning" (i.e. similarity) ratios between cultures or subcultures, or to identify to which culture or subculture a consumer belongs (Szalay and Deese 1978). If the same or similar responses are associated with the same product, the cultures may be classified as similar (Hirschman 1980). Additionally, where consumers belong to mere than one subculture, for example, a Jewish American teenager, the associations made in relation to a particular product will indicate which subculture dominates in relation to that particular product.

The methodology can also be extended to examine the manning attached to different colors, shapes or to composite objects in different cultures and countries (Deregowski 1980, Pick 1980). Consumers could, for example, be given a sec of colors or shapes and asked to provide associations to them. Similarly more complex stimuli sets such as actual products and brands could be used.

Such procedures might also be appropriate to study children's responses to products, brands and advertisements. Since small children have difficulty grasping abstract concepts, use of visual stimuli may aid considerably in increasing comprehension.

Card Sorting

The card sorting task would also appear to have potential applications in cross-cultural and subcultural research. Again, the same caveats apply as for the free association task in that care must be taken to avoid a pseudo-etic bias in the design of the product set and in its interpretation. In many cases it may be desirable to include in the set, products specific to a particular culture, as well at those common across cultures.

If a set of products common across cultures is identified, the card-sorting task may be used to assess whether certain cultures are characterized by predominantly product-type, brand-type or other cognitive organization structures. The level of advertising may, for example, affect information organization. The relative looseness or tightness of the organizational pattern and complexity of organization also may vary across cultures depending upon the level of education or societal development.

As in the case of free association techniques, the Hirschman-Douglas sort task can be extended by using more complex stimuli. These might, for example, include use of black and white or colored pictures and photographs of products and brands or actual objects. Consumers would then be requested to sort these into piles based on similarity, and differences and similarities in these patterns between cultures and countries could be examined. Again, the technique might be highly appropriate with children.

Finally, both the free association and card-sorting tasks may be used to examine the relation between cognitive structure and respondent characteristics. The number of responses given to a stimulus product may be related to high levels of cognitive complexity and intelligence. The tightness and completely of the organizational pattern may also reflect intelligence and cognitive complexity, and may be related to such personality traits as desire for closure and tolerance of ambiguity. Other factors such as media usage patterns and relative level of information seeking may also relate to cognitive structure and, hence, to a consumer's performance on both the free response and card-sorting tasks.

CONCLUSIONS

The purpose of this paper is to suggest the use of two approaches - free association and card sorting - in investigating hierarchical cognitive content. These approaches provide a useful extension to the type of information obtained from traditional approaches. In particular, they may aid in extending knowledge of the content and structure of consumer product cognitions and information storage and have potential application in cross-cultural and cross-national research.

The results of the present experiment indicate that the majority of subjects organize the stimulus set into hierarchies, generally of more than five groups. The test re-test reliability of this finding was also high, indicating stability in results, and specific that the patterns obtained were not specific to a particular product class. The patterns may however be in part due to the nature of the stimulus set, although alternative groupings were feasible.

It should, however, be noted that both approaches, and in particular, the card sort task, are still in the early stages of development. Considerable further work in refinement of the method, particularly with regard to the development of the stimulus set, notably a more ambiguous set and appropriate analytic procedures will clearly be required. Furthermore, adaptation relative to specific research objectives is also needed.

REFERENCES

Bettman, James R. (1979), "Memory Factors in Consumer Choice: A Review," Journal of Marketing, (Spring), 37-53.

Capon, Noel and Lutz, Richard J. (1979), "A Model and Methodology for the Development of Consumer Information Programs," Journal of Marketing, (January), 58-67.

Deregowski, Jan B. (1980), "Perception," in Harry C. Triandis and Walter Lonnes (eds.). A Handbook of Cross-Cultural Psychology Vol. 3, Basic Processes (Boston: Allyn and Bacon, Inc.) 21-117.

Douglas, Susan P., Craig, Samuel and Faivre, Jean-Philippe (1981), "Protocols in Consumer Research: Problems, Methods and Uses," to appear in Jagdish Sheth (ed.), Research in Marketing, Vol. V.

Ember, Carol R. (1977), "Cross-Cultural Cognitive Studies," Annual Review of Anthropology, 6, 33-56.

Elder, Joseph W. (1976), "Comparative Cross-National Methodology," Annual Review in Sociology, Palo Alto, California.

Gelman, Rachel (1978) "Cognitive Development," in Annual Review of Psychology, 29, 297-332.

Green, Paul E., Wind, Yoram and Jain, Arunk (1973), "Analyzing Free-Response Data in Marketing Research," Journal of Marketing Research, X (February), 45-52.

Gutman, Jonathan (1980), "Equivalence Range in Categorization," in Jerry Olsen (ed.) Advances in Consumer Research, Vol. VII, (Ann Arbor: Association for Consumer Research), 411-416.

Eirschman, Elizabeth C. (1980), "Commonality and Idiosyncracy in Popular Culture," in Consumer Aesthetics and Symbolic Consumption, E.C. Hirschman and Morris Holbrook` (eds.), Association for Consumer Research, forthcoming.

Lachman, Roy, Lachman, Janet and Butterfield, Earl C. (1979.), Cognitive Psychology and Information Processing, Hillsdale, NJ: Lawrence Erlbaum and Assoc., 1979.

Lee, James A. (1966), "Cultural Analysis of Overseas Operations," Harvard Business Review, 44 (March-April), 106-114.

Olson, Jerry C. and Muderrisoglu, Aydin (1977), "The Stability of Responses Obtained by Free Elicitation" in Advances in Consumer Research, Vol. 6, W.C. Wilkie (ed.), Ann Arbor, MI, Association for Consumer Research, 269-275.

Olson, Jerry C. (1978), "Theories of Information Encoding and Storage: Implications for Consumer Research," in Effects of Information on Consumer and Market Behavior, A.A. Mitchell (ed.) Chicago: American Marketing Association.

Osgood, Charles E., May, William H. and Miron, Murray S. (1975), Cross-Cultural Universals of Affective Meaning Urbana: University of Illinois Press.

Piaget, Jean (1970), Genetic Epistemology, New York: Norton.

Pick, Anne D. (1980), "Cognition: Psychological Perspectives," in Harry C. Triandis and Walter Lonner (eds.), A Handbook of Cross-Cultural Psychology, Vol. 3, Basic Processes, (Boston: Allyn and Bacon, Inc.) 117-154.

Pike, Kenneth (1966), Language in Relation to a Unified Theory of the Structure of Human Behavior, the Hague: Mouton.

Preworski, Adam and Tenue, Henry (1970), The Logic of Comparative Social Intuits, New York: Wiley Interscience.

Rosch, Eleanor (1973), "Natural Categories," Cognitive Psychology, 4, 328-350.

Rosch, Eleanor (1974), "Universals and Cultural Specifics in Human Categorization," in Cross-Cultural Perspectives On Learning, R. Breslin, W. Conner and S. Bachner (eds.) London: Sage Press.

Rosch, Eleanor (1975), "Cognitive F. Representations of Semantic Categories," Journal of Experimental Psychology, 104, 192-233.

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

Rosenzweig, Mark R. (1961), "Comparisons of Word Association Responses in English, French, German and Italian," American Journal of Psychology, 74, 347-360.

Russo, Edward J. and Johnson, Eric J. (1980), "What Do Consumers Know About Familiar Products?" in Jerry Olson (ed.) Advances in Consumer Research, Vol. VII (Ant Arbor: Association for Consumer Research), 417-423.

Szalay, Lorand B. and Deese, James (1978), Subjective Meaning and Culture, Lawrence Erlbaum Associates: Hillsdale, NJ

Triandis, Harry C. (1972), The Analysis of Subjective Culture, New York: John Wiley.

Warner, Oscar and Campbell, Donald T {1973),"Translating, Working Through Interpreters' and the Problems of Decentering", in Paoul Naroll and Ronald Cohen (eds.) A Handbook of Method in Cultural Anthropology, New York: Colombia University Press, 398-420.

Wind, Yoram and Douglas, Susan (1980), "Comparative Consumer Research: The Next Frontier," Wharton School Working Paper, University of Pennsylvania, Philadelphia, PA, (February).

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