PERSONALITY CHARACTERISTICS AND CONSUMER - PRODUCT PERCEPTIONS: METHODOLOGICAL ADVANCES AND EMPIRICAL VERIFICATION OF A COGNITIVE MODEL Thesis for the Degree of Ph. D. MICHIGAN STATE UNIVERSITY LEIGHTON ADAMS PRICE 1 9 72 LIBRARY Michigan State University This is to certify that the thesis entitled PERSONALITY CHARACTERISTICS AND CONSUMER-PRODUCT PERCEPTIONS: METHODOLOGICAL ADVANCES AND EMPIRICAL VERIFICATION OF A COGNITIVE MODEL presented by Leighton Adams Price has been accepted towards fulfillment of the requirements for Ph . D . degree in Psychology aggcw; .. Q LJ-~L\2€n—SL’ Major professor Date June 26, 1972 0-7539 I I .-I 800K BINDERY iNC.II II HUAG & we I! LIBRARY b NOE 9‘1 S'IIIB'IIRY flui- T w. I. _A H__..-—---_ .———u. ABSTRACT PERSONALITY CHARACTERISTICS AND CONSUMER-PRODUCT PERCEPTIONS: METHODOLOGICAL ADVANCES AND EMPIRICAL VERIFICATION OF A COGNITIVE MODEL BY Leighton Adams Price Objectives Psychological theories of complex perceptual-cognitive Phenomena generally suggest that broad cognitive characteristics are related to perception of rather specific stimuli; nevertheless, empirical support for such relationships has been meager. The study of consumer behavior is one such area where general characteristics IGJL. personality measurements) and specific characteristics (e.g., consumer-product perceptions) have seldom been found to be related to each other as logically expected. It remains possible, however, that the meager evidence is symptomatic of theoretical and methodo- 1Ogical deficiencies rather than weak relationships in the actual Phenomena. The present research was based on the assumption that theoretical and methodological deficiencies have contributed nflxutantly to this state of affairs. Consequently, the research Was concerned simultaneously with: Leighton Adams Price 1. Utilizing a model of perceptual-cognitive processes incorpo- rating constructs which were adequately matched with the complexities of the phenomena involved. 2. Utilizing a methodology which realistically matched data collection and analysis with the complexity of the constructs in the model. 3. Performing empirical tests of relationships between two classes of cognitive phenomena described by the model (i.e., rather general cognitive characteristics represented by personality measures and more specific cognitive character- istics represented by sterling tableware perceptions). The cognitive model and the methodology developed for the present research have their immediate origins in the typological theories and pattern-analytic methods of McQuitty, in Stephenson's Q-methodology, in Osgood's model of meaning systems and Semantic Differential, in Fishbein's model of attitude formation and methods for measuring attitude, as well as in Rokeach's theory of belief-value systems. The new model seeks to integrate and extend existing models so that the inherent complexity of perceptual-cognitive phenomena may be more adequately conceptualized. The methodology seeks to capture the detail and organization of perceptual-cognitive phenomena while simultaneously relaxing measurement and statistical constraints. Results 1. Logically expected relationships between personality and consumer-product perceptions were obtained through analysis 2. Leighton Adams Price of data collected from an appropriate sample of university women having varying degrees of interest in sterling tableware design. These results, in turn, provided indirect support for the validity of the new model despite the demanding empirical context for the research. a. Two rather different kinds of personality variables (i.e., cognitive content and cognitive structure vari- ables) were found to relate to sterling tableware perceptions in theoretically expected ways. In particular, homogenous but contrasting personality "types” differed in the content, response-style, and structural character- istics of their perceptions. Since these analyses were performed for several related inventories, a quasi multi- trait multi-method cross-validation was achieved. Indirect support for the validity of the model manifested itself on several levels. Aggregate analyses of per- sonality "types" yielded content, response-style, and structural results supporting basic constructs of the model. These analyses also indicated that within-type similarities could realistically be treated as a system. Analyses for individuals indicated that evidence of broad cognitive constructs could be captured in the detailed responses of single individuals. The strength of the hypothesized relationships also held many implications for the merits of the present data collection and analysis methods. Leighton Adams Price a. The principal technique developed to measure sterling tableware perceptions (the Object Descriptions Task) constituted a realistic yet easily constructed and versatile instrument. Unique characteristics of the task provided some of the main differentiations between content, response-style, and structural characteristics of contrasting "types." b. Methods developed to analyze the masses of perceptual detail yielded by the Object Descriptions Task were clearly sensitive to both the content and organization of cognitive phenomena. Conventional methods would have been hard pressed to handle such masses of data without imposing more severe measurement and statistical con- straints. It should be noted, however, that the present methods would not have been feasible without computers. Applications The most obvious applications of the model and methodology center on consumer behavior (e.g., analyses of product image, relation— ships among competitive products, segmentation by personality chanacteristics, changes in product perception over time). Neverthe- less, these are merely examples of the many ways that the model and nmnhodology could be employed. The research strategy is a comprehensive one applicable to many problems in the social sciences, and it may even be directed toward several problems simultaneously. As a result, the methodology can help to achieve research efficiency while the model helps to achieve parsimony in explanation. Leighton Adams Price Even though models and methods are never really developed as much as they might be, the strength of the results indicates that the present model and methodology are formidable competitors of existing approaches and that they are sufficiently well developed to be used in applied research. PERSONALITY CHARACTERISTICS AND CONSUMERrPRODUCT PERCEPTIONS: METHODOLOGICAL ADVANCES AND EMPIRICAL VERIFICATION OF A COGNITIVE MODEL BY Leighton Adams Price A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Psychology 1972 @ Copyright by LEIGHTON ADAMS PRICE 1972 ii AC KNOWLE DGMENT S First and foremost, I should like to express my gratitude to the chairman of my committee, Dr. Frederic R. Wickert. Through much of the developmental work underlying the present research he made many suggestions and useful criticisms. More importantly, he has generously given thorough and constructive criticism throughout the effort to present a rather complex research endeavor as a single integrated package of theory, methodology, and empirical tests. I should also like to extend my gratitude to Dr. Eugene H. Jacobson, Dr. William D. Crano, and Dr. William J. E. Crissy. Numerous aspects of the present research strategy took shape in the course of discussions with them. In addition, these discussions were directly responsible for focusing analyses upon the content, response- style, and structural characteristics of object perceptions manifested kw contrasting personality types. While the research was not explicitly concerned with non- verbal communication, this area of investigation was clearly relevant tX>the development of the present model and methodology. In this regard, Dr. Randall P. Harrison of the Department of Communication contributed importantly to the evolution of thinking which culminated in the present research. iii 'I-H. n . . ”o... ’0... Wu'. . :9“. -A._ _;I ‘I. I K. -I ‘ \ . . , 3.. ‘u . ‘- o .o ,. s ' a . o u u an I" Special debts of gratitude are also owed to Dr. Louis L. McQuitty and Dr. James A. Clark. Their ideas on typological analysis and data collection in the social sciences constituted the major stimuli for pilot research from which the present model and methods evolved. I also found their reactions to my own ideas to be both perceptive and constructive. I should also like to thank Dr. Charles F. Wrigley of the Computer Institute for Social Science Research for supporting the extensive computer time needed to develop analysis methods used in the research and to analyze the data collected, for funding a number of qualified undergraduates apprentices to help program several of the methods for computers and to assist with some of the analyses. Financial support for persons hired to help prepare materials, to assist with data collection, and to code data for analysis was also provided by a Biomedical Sciences Support Grant (FRO 7049-03). The sterling silver tableware used in the research was generously loaned to me by the Gorham Corporation of Providence, Rhode Island. Several hundred forks covering a broad range of styles were provided. These designs were used in both pilot research and the research reported here. Finally, I should like to dedicate this dissertation to my vfife, Dorothy, who has provided moral support for carrying out this .research and has contributed to it through valuable critical discussion rand extensive amounts of typing, to our children, Diana and David, who luave endured the demands of carrying out the numerous phases of this research, and to my father, Mr. J. Russell Price, whose own work in iv the area of consumer preferences stimulated several of the data collection and analysis methods developed as part of the present research. TABLE OF CONTENTS Page LIST OF TABLES o o o o o o o o o o o o o o o o x LIST OF FIGURES. o o O o o o o o o o o o o o o Xiii Chapter I. THE PROBLEM OF STUDYING COMPLEX PERCEPTUAL-COGNITIVE PmCESSESo O O O O O O O I O I O O I O l IntrOduCtion O O O O O O O O O O O O O O 1 Origins of the Model and Research Strategy. . . . 4 An Overview of the Objectives of the Research. . . 7 Theoretical Objectives. . . . . . . . . . 7 Methodological Objectives. . . . . . . . . 9 Practical Objectives . . . . . . . . . . 11 II. A MODEL OF COMPLEX PERCEPTUAL-COGNITIVE PROCESSES AND A RESEARCH STRATEGY FOR TESTING THE MODEL . . . . 14 A Model of Complex Perceptual-Cognitive Processes . 14 Overview of the Model . . . . . . . . . . 14 Definitions . . . . . . . . . . . . . 1? Extensions of the Basic Model . . . . . . . 24 Comparisons With Other Models . . . . . . . 27 A Strategy for Studying Complex Perceptual-Cognitive Systems and Comparisons With Other Strategies . . 31 Overview of the Research Strategy . . . . . . 32 Data Collection . . . . . . . . . . . . 34 Data Analyses. . . . . . . . . . . . . 39 III. RESEARCH METHODS . . . . . . . . . . . . . 62 SWjects O O O O O O O O O O O O O O O 62 Materials. . . . . . . . . . . . . . . 64 vi Chapter Preliminary Instructions . . . . . . . . Personality Inventories . . . . . . . . Consumer-Products Selected for the Research. . Object Evaluations Task . . . . . . . . Object Descriptions Task . . . . . . . . Word Associations Task. . . . . . . . . Background Information Questionnaire . . . . Data Collection Design . . . . . . . . . Data Collection Controls. . . . . . . . . Between Groups . . . . . . . . . . . Between Tasks. . . . . . . . . . . . Within TaSks O O O O O O O O O O O 0 Data Collection Procedures . . . . . . . . Research Setting. . . . . . . . . . . Assistants. . . . . . . . . . . . . Group-Administration Procedures. . . . . . Preparing Data for Analysis. . . . . . . . Machine-Readable Forms. . . . . . . . . Pre-Coding. . . . . . . . . . . . . Post-Coding . . . . . . . . . . . . Data Editing . . . . . . . . . . . . Reordering Randomized Data . . . . . . . Classifactory Analyses of Personality Data. . . Cognitive Complexity Index . . . . . . . Dogmatism Scale . . . . . . . . . . . Orientation Inventory . . . . . . . . . Value Survey . . . . . . . . . . . . Methods for Mapping Consumer-Product Perceptions. Distance Matrices for Relationships Among Elementary Cognitive Subsystems . . . . . Content Analyses. . . . . . . . . . . Response-Style Analyses . . . . . . . . Structural Analyses. . . . . . . . . . Methods for Differentiating Between Cognitive Maps for Different Personality Types. . . . . . vii Page 64 65 68 71 72 82 82 83 85 85 85 85 86 87 87 88 92 92 92 92 93 93 94 94 94 95 96 97 97 105 108 111 115 Chapter Iv. Content Differentiations . . . . . . . . Response-Style Differentiations. . . . . . Structural Differentiations . . . . . . . Summary of Analyses . . . . . . . . . . RESULTS AND DISCUSSION . . . . . . . . . . Review of the Research Objectives and Strategy . Content Differentiations. . . . . . . . . Review of Content Differentiation Procedures . Results of Content Differentiation Analyses. . Response-Style Differentiations . . . . . . Review of Response-Style Differentiation Procedures . . . . . . . . . . . . Results of Response-Style Differentiation Analyses. . . . . . . . . . . . . Summary of Content and Response-Style Results. . Personality Inventories Concerned With Structure 0 O O O O O I O O I I 0 Personality Inventories Concerned With Content. Structural Differentiations. . . . . . . . Review of Structural Differentiation Procedures Results of Structural Differentiation Analyses. Summary of Structural Differentiation Results . CONCLUSIONS, SUPPORTING RESULTS, AND IMPLICATIONS FOR FURTHER RESEARCH . . . . . . . . . . Theoretical Conclusions . . . . . . . . . Personality Characteristics and Consumer-Product Perceptions. . . . . . . . . . . . Indirect Support for the Model of Complex Perceptual-Cognitive Processes . . . . . Methodological Conclusions . . . . . . . . Data Collection . . . . . . . . . . . Data Analysis. . . . . . . . . . . . viii Page 116 116 117 118 121 121 125 126 129 156 157 161 179 180 186 196 197 202 217 223 223 224 228 233 234 240 '..o ,_.,.a . uu.-.. n a o .0.0 ~. o., Chapter Page Practical Conclusions . . . . . . . . . . . 249 Implications for Further Research . . . . . . . 260 BIBLIOGRAPHY . . . . . . . . . . . . . . . . . 268 APPENDICES Appendix A. Preliminary Instructions. . . . . . . . . . . 276 B. Personality Inventories . . . . . . . . . . . 280 C. Objects Used in the Study . . . . . . . . . . 285 D. Object Evaluations Task . . . . . . . . . . . 291 .8. Object Descriptions Task. . . . . . . . . . . 294 1?. WOrd Associations Task . . . . . . . . . . . 308 (3.. Background Information Questionnaire. . . . . . . 313 ix . u.- . Table l. 11) 11 LIST OF TABLES Constraints and Capabilities of Several Methods for Differentiating Between Groups . . . . . . . . Summary of Differentiation Analyses Performed on the Consumer-Product Perceptions of Contrasting Personality Subgroups . . . . . . . . . . . Salient Content Differentiating Between the Consumer- Product Perceptions of Cognitive Complexity Subgroups............... Salient Content Differentiating Between the Consumer- Product Perceptions of Dogmatism Subgroups . . . . Salient Content Differentiating Between the Consumer- Product Perceptions of Self-Oriented and Interaction- Oriented Subgroups . . . . . . . . . . . . Salient Content Differentiating Between the Consumer- Product Perceptions of Task-Oriented and Interaction- Oriented Subgroups . . . . . . . . . . . . Salient Content Differentiating Between the Consumer— Product Perceptions of Self-Oriented and Task- Oriented Subgroups . . . . . . . . . . . . Value Configurations for Terminal Value Subgroups . . Salient Content Differentiating Between the Consumer- Product Perceptions of Terminal Value Subgroups . . Value Configurations for Instrumental Value S‘lbgroups I O O O O O O O O O O O O O O Salient Content Differentiating Between the Consumer- Product Perceptions of Instrumental Value S‘Jbgroups O O O O O O O O I O O O O O O Page 60 119 131 135 139 141 144 147 150 153 156 '0: Table 12. 13. 14. 15. 16. 17. 18. 19. 20. 214 122 323 3L4 Response-Style Differences Between the Consumer- Product Perceptions of Cognitive Complexity Subgroups.............. Response-Style Differences Between the Consumer- Product Perceptions of Dogmatism Subgroups . . . Response-Style Differences Between the Consumer- Product Perceptions of Self-Oriented and Interaction-Oriented Subgroups . . . . . . . Response-Style Differences Between the Consumer- Product Perceptions of Task-Oriented and Interaction-Oriented Subgroups . . . . . . . Response-Style Differences Between the Consumer- Product Perceptions of Self-Oriented and Task— Oriented Subgroups . . . . . . . . . . . Response-Style Differences Between the Consumer- Product Perceptions of Terminal Value Subgroups . Response-Style Differences Between the Consumer— Product Perceptions of Instrumental Value Subgroups . . . . . . . . . . . . . . Summary of Content and Response-Style Results for Subgroups Identified Through Measures of Personality Structure . . . . . . . . . . Summary of Content and Response-Style Results for Orientation Inventory Subgroups. . . . . . . Summary of Content and Response-Style Results for Value Survey Subgroups. . . . . . . . . . Dimensional Usage Differences Between the Consumer- Product Perceptions of Cognitive Complexity and Dogmatism Subgroups. . . . . . . . . . . Dimensional Usage Differences Between the Perception of Products Liked and Products Disliked by Dogmatism Subgroups. . . . . . . . . . . Other Dimensional Usage Differences Between the Perception of Products Liked and Products Disliked by Dogmatism Subgroups. . . . . . . . . . xi Page 163 167 171 173 175 177 178 181 188 192 206 210 211 so Table Page 25. Other Indices of Structural Differentiation Between the Perception of Products Liked and Products Disliked by Dogmatism Subgroups . . . . . . . . 213 26. Structural Similarity of Systems Underlying the Perception of Products Liked and Products Disliked. I I I I I I I I I I I I I I I 216 C-1. List of Forks Used in the Object Evaluations Task. . . 286 C-2. List of Forks Used in the Object Descriptions Task . . 288 E-l. List of Attributes Presented. . . . . . . . . . 301 E-2. Random Orderings of the Forks . . . . . . . . . 302 E-3. A Classification of Recurring Content Groupings . . . 307 xii lCL 11” LIST OF FIGURES Model of Complex Perceptual-Cognitive Processes . . Spatial Representation of the Components of an Elementary Cognitive Subsystem. . . . . . . . Spatial Representation of the Relationship Between Two Meanings That are Associated With an Object . . Hierarchically Organized System Representing a Cognitive Object . . . . . . . . . . . . Spatial Representation of a Cognitive System Including Several Cognitive Objects . . . . . . Comparison of an Elementary Cognitive Subsystem With the Cognitive Units of Other Models . . . . Schematic Summary of the Data Collection Design . . Cube Representing Object Descriptions Task Data . . Single "Chip" from the Object Descriptions Task Data Cube. . . . . . . . . . . . . . . Flowchart for Content Differentiations Between the Consumer-Product Perceptions of Contrasting Subgroups. . . . . . . . . . . . . . . Flowchart for Response-Style Differentiations Between the Consumer-Product Perceptions of Contrasting smroups I I I I I I I I I I I I I I I xiii Page 15 19 22 23 26 29 84 99 100 128 161 Figure Page 12. A Flowchart for Group-Composite Structural Differ- entiations Between the Consumer-Product Perceptions of Contrasting Subgroups . . . . . . . . . . . 201 13. A Flowchart for Individual Structural Differentiations Between the Consumer-Product Perceptions of Contrasting Subgroups . . . . . . . . . . . . 203 xiv CHAPTER I THE PROBLEM OF STUDYING COMPLEX PERCEPTUAL-COGNITIVE PROCESSES Introduction Psychologists concerned with cognitive organization generally agree that the perceptual-cognitive representations of many stimuli are quite complex. For example, the mental processes associated with abstract objects (e.g., social roles, self-image, or ideologies) and With many everyday physical objects (e.g., buildings, consumer products, or other people) are generally viewed as being inherently complex. Despite this consensus, many of the models and research techniques which psychologists devise seem to fall short of capturing the essential complexities of the underlying perceptual-cognitive phenomena. The SOrts of problems encountered with existing models and methods are that: (1) models tend to oversimplify the phenomena involved, (2) techniques for collecting perceptual-cognitive data often impose unnecessary constraints to achieve quantification, and (3) analysis te<21'1niques tend to impose statistical constraints which may further prOhibit discovering important characteristics of the processes ‘1 n"<3 lved . .11 The thesis here is that models of complex cognitive organi- zation as well as techniques for studying these systems can be improved appreciably by treating model building and methodological developments as interdependent aspects of the research process. The model and the methods developed for the present investigation are offered as efforts toward integrating and extending existing knowledge of perceptual-cognitive phenomena by more fully and more realistically accounting for the complexity of such phenomena. The present research has focused on the perceptual-cognitive processes (or systems of meanings) representing two rather different sorts of objects: (1) the abstract object called self-image or personality, and (2) material objects belonging to a class of consumer products (sterling silver tableware). With respect to self-image, the underlying systems of meanings were studied through the use of personality inventories. For the consumer products selected, the underlying meaning-systems were measured and analyzed with methods developed especially for this research. It is expected, however, that the model and the methodology employed should apply equally well to a wide variety of other abstract and material objects. Personality characteristics and consumer products were originally selected for the purpose of examining theoretical and pmactical questions related to consumer behavior. However, in the tunader context of evaluating a model of complex perceptual-cognitive representations, they mainly provided the conditions for testing the idea that apparently dissimilar objects may be psychologically related. Personality traits emerge from an individual's life experiences and pertain to rather general behavior patterns. While sterling tableware is often an integral part of a social setting and hence may gain symbolic significance, many of the responses to these products are likely to be specific to the immediate object and situation. In other words, the systems of symbolic meanings representing these objects are unlikely to overlap unless there are substantial relationships between these symbolic domains. The significance of seeking relationships between represen- tations for different sorts of objects derives from the fact that psychologists have frequently been unsuccessful in finding relation- ships between general characteristics, such as personality traits, and the more specific representations for everyday objects in the real world. If stable relationships are revealed by the present approach and these relationships also make psychological sense, then it may be tentatively concluded that consumer-product perceptions are, in part, reflections of an individual's personality traits. More importantly, such results would provide some indirect support for both the present model of complex meaning-systems and the research strategy employed. Most importantly, progress would be achieved toward better operationalization, mapping and understanding of complex perceptual-cognitive systems. In order to test the model developed for this research, it was necessary to develop both data collection and analysis techniques. Specifically, the methodological contributions of the research lie in the development of: (l) a technique for collecting more realistically complex perceptual-cognitive data, (2) operationalizations of a model which make it possible to quantify relationships among meanings, (3) methods for analyzing the content and organization of meaning- systems, (4) methods for comparing the content and organization of different meaning-systems, and (5) computer programs for performing many of the complex data management and analysis tasks of mapping and comparing perceptual-cognitive systems. Extensive developmental work has gone into these methods, and the research reported here is a first effort to explore their potential and to obtain ideas for further development. The methodology of the present research is clearly dependent upon the use of large-scale computers, and the computer programs developed to help make sense of complex perceptual-cognitive data may very well constitute some of the more enduring contributions of the research. In the absence of large-scale computer facilities, neither the developmental work nor the empirical efforts to unravel the complexities of meaning-systems would have been feasible. Origins of the Model and Research Strategy The concern of this investigation with the development of a model and its accompanying methods has its origins with McQuitty's interest in the mutual development of theory and method for hier- archical pattern-analysis (e.g., McQuitty, 1959; McQuitty, 1966b; McQuitty , 1967) . For some years, McQuitty has stressed the importance of regarding the development of typological theory and typological analysis methods as interdependent aspects of his research. The writer has extended this viewpoint to include operationalizations ‘9 '- making data collection compatible with the model and with the data analysis methods. While the general form of the present investigation was originally inspired by a typological and pattern-analytic study of cognitive systems for individuals (McQuitty, Abeles, and Clark, 1970), both the model and methods used in this study have evolved to the point where they are now rather remote from their origins. For example, in the course of conducting pilot research using pattern-analytic techniques to analyze a single individual's perceptions of several consumer products (McQuitty, Price, and Clark, 1967), it was found that data collection techniques devised for the study were related to certain definitions of attitude (e.g., Rokeach, 1968). As a result, relationships of these measurement techniques to existing attitude measures were explored and some aspects of the present model began to take shape. In subsequent pilot research (Price, 1968), when McQuitty's methods were again used to study relationships between personality characteristics and consumer-product perceptions, evidence showed that hierarchical pattern-analysis revealed connections between belief-value systems (as defined by Rokeach, 1960, 1968) and certain perceptual- cognitive representations of consumer-products. However, while attempting to analyze these data, the writer discovered a number of deficiencies in McQuitty's hierarchical clustering techniques and developed an alternative clustering method which attempted to avoid some of these deficiencies (Price, 1969). While developing data collection methods for the present investigation, it occurred to the writer that many of the ideas which underlie the model and its operationalizations have much in common with Osgood's (Osgood, g£;21,, 1957; Osgood, 1962, 1965) notions of semantic space, with consistency models of cognitive organization (e.g., Abelson, 1959; Abelson and Rosenberg, 1958; Cartwright and Harary, 1956; Heider, 1946; Rosenberg, 1956, 1960), and with be- havioral models of attitude organization (Fishbein, 1967a, 1967b, 1967c; Rhine, 1958). As a result, the present technique for collecting perceptual-cognitive data has some of the features of the Semantic Differential (Osgoodq SELiEL'I 1957), as well as features of an attitude measurement technique based on beliefs about objects and the evaluative aspects of those beliefs (Fishbein and Raven, 1962; Anderson and Fishbein, 1965). It should also be noted that, in addition to McQuitty's influence upon the choice of analytic techniques, the techniques developed especially for this research were influenced by other researchers as well. For example, techniques for identifying personality types and comparing the mappings of different types were influenced by the Q-methodology of Stephenson (1953). Similarly, techniques for analyzing the organization of meaning-systems for different types were influenced by an exact-pattern clustering method developed by Clark (1968). An Overview of the Objectives of This Research The objectives of this research fall into three categories: theoretical, methodological, and practical. Theoretical Objectives The theoretical objectives of this thesis were: (1) to develop a model of complex perceptual-cognitive processes which attempts to integrate a number of existing models while simultaneously going beyond the scope of these models, and (2) to test the resulting model by applying it to the study of relationships between personality characteristics and consumer-product perceptions. Although the model is not radically different from existing models of perceptual-cognitive systems, it appears to have several advantages over them. First, it provides a theoretical framework from which reasonably precise operationalizations may be derived. In the second place, several theoretical viewpoints may be seen as but different aspects of the same general model. Third, the model takes factors into account which are generally either ignored or not as well operationalized as they might be. In other words, the model may provide a closer approximation than existing models to the way perceptual-cognitive processes actually work. The approach to testing the model was based on the assumption that an individual's life experiences can simultaneously affect numerous cognitive domains. Specifically, it was assumed that: (1) personality characteristics could be taken as examples of the built- in effects of long-term life experiences, (2) the symbolic I-o- I .- 0.... representations for certain consumer products could be taken as examples of a rather specific cognitive domain, and (3) these two rather different domains could become interrelated through common or related experiences. While the testing of the model has many implications for the psychology of consumer behavior, it might also help to provide insights into the dynamics of complex meaning-systems. Personality variables were selected for use in the present research for two main reasons: (1) they provide broad and well- researched, though often somewhat unreliable, measures of cognitive behavior patterns, and (2) self-image (as measured through the self- report items of personality measures) may be expected to be funda- mentally different from the highly specific sorts of responses given for consumer products. In brief, it should be difficult to find relationships between these two rather different classes of responses unless: (1) the objects are linked through related experiences in physical-social settings, (2) similar systems of symbolic meanings are associated with both sorts of objects, and (3) the research methods employed are realistically sensitive to the content and organization of the perceptual-cognitive systems under consideration. The tests of relationships between personality variables and consumer-product perceptions were severe in other respects as well. For one, personality measures certainly oversimplify the actual situation to achieve quantification. Furthermore, personality measures all too seldom have been found to correlate with other variables. Because of such considerations, it should be doubly difficult to find relationships between general personality measures and specific responses to consumer products. On the other hand, should relationships be found, these relationships would provide support for the model as well as inferential validity for the personality measures employed. Methodological Objectives To adequately test the model of complex perceptual-cognitive representations for objects, it was, as mentioned earlier, necessary to devise data collection and data analysis techniques that would operationalize the theoretically generated components of the model. As a result, efforts to test the model were simultaneously efforts to test the utility of data collection and analysis techniques which operationalize the model. Collecting Complex Perceptual-Cognitive Data.--The problem of measuring the meanings associated with consumer products and other real-life stimuli is the familiar one of determining how physical reality is perceived and what psychological representations are achieved for these stimuli. While many techniques for collecting complex data are restricted to somewhat global responses, the thrust of this research has been to work toward obtaining detailed and specific reactions to objects. The data collection techniques developed for the present research facilitate obtaining a maximum amount of associational information from respondents instead of asking them to give responses summarizing their reactions. Respondents were asked to rapidly record a very large number of highly specific reactions; they were 10 not asked to summarize their reactions or give overall impressions. As a result, the burden was put on the quantitative methods used to summarize these masses of perceptual-cognitive responses. Analyzing Complex Perceptual-Cognitive Processes.--Two assumptions important in the development of the analysis methods were that: (l) the meanings associated with any object take the form of complex response syndromes or structures, and (2) the character- istics of these complex responses differ from one personality "type" to another. These assumptions pose an enormous challenge for analysis methods. The methods should be capable of handling both linear and nonlinear relationships and, at the same time, be able to differ- entiate between the various parts of meaning systems or between personality "types." In other words, it is important to use analytic techniques capable of handling highly flexible and varying content. As mentioned earlier, the analysis techniques used in this research were influenced both by type-identifying methods developed by Stephenson (1953) and by pattern-analytic methods developed by MoQuitty (1967). Stephenson's influence manifests itself mainly in the use of a Q-methodology approach to identifying personality types and to comparing the results of analyses for different types. lkwmwer, with respect to the means by which types were identified and perceptual-cognitive processes were analyzed, the methodology was influenced more by McQuitty. Although McQuitty was himself strongly influenced by Stephen- son '8 work, he argued that potentially nonlinear systems should be analyzed with pattern-analytic techniques rather than factor analyses, 11 thereby avoiding the assumption of linearity and other constraints. McQuitty's impact on the present research is reflected in the fact that personality types were identified with cluster analysis and that object associations were also cluster-analyzed. Despite these similarities with other methods, there are several ways in which the present methodology differs from its antecedents. Most importantly, the methodology is distinguished by the fact that two psychologically very different classes of perceptual-cognitive phenomena are being related to one another: personality characteristics and consumer-product representations. Traditionally, cluster analysis has been applied in studies of single content areas. In the present research, however, a typological approach has been combined with distribution and cluster analyses of meaning-systems for the purpose of studying relationships between the meaning-systems underlying responses to personality inventories, on the one hand, and responses to consumer products, on the other hand. Practical Objectives While applications have not been given direct attention in this thesis, the model and the methodology developed for the research have many practical implications. The practical objectives of the research were, therefore, to consider some of the ways in which the model and methodology might be employed in the real world. Initially, some consideration was given to the implications of this research for similar research in the area of consumer behavior, its applicability to a broad range of marketing problems, and its 12 still more general applicability to other research problems in the social sciences. Second, since the data collection techniques developed for this research were intended to be highly versatile, a number of specific applications in the area of consumer behavior were con- sidered. For example, in addition to the present use of the techniques to test theoretical questions concerning complex perceptual-cognitive phenomena, the techniques might be used to study people's images of particular products, relationships among products in a line of goods, desired characteristics of products, similarities of product perceptions among the members of demographi- cally or psychologically defined market segments, intensive analysis of people's perceptions of a particular product, and many others. The data collection techniques sought to provide a common, psycho- logically based, unit of measurement which could be used in studies involving different content, objects, and persons. That is, the techniques hopefully could bring some measurement efficiency to areas of research which have proliferated different measurement techniques for each content area needing measurement instead of developing a single technique which may be adapted to different pueblems. For example, techniques for studying likes and dislikes for products traditionally have been quite different from those used to determine why people react as they do. Finally, the practical merits of employing a typological research strategy and the set of methods developed for analyzing complex perceptual-cognitive systems are discussed. For intensive 13 analytic methods to be applicable to the improved understanding of consumer behavior as well as to realistic marketing problems, it is important that the amount of information be maximized relative to the cost of obtaining it. The analysis methods developed for the present research Should be suited to analyzing masses of data that may be viewed from many perspectives; moreover, these methods should be well suited to analyzing configural differences which may characterize a typological research strategy. Since the methods should be applicable in small sample research, the objective of maximizing utility may also be achieved. CHAPTER II A MODEL OF COMPLEX PERCEPTUAL-COGNITIVE PROCESSES AND A RESEARCH STRATEGY FOR TESTING THE MODEL A Model of Complex Perceptual-Cognitive Processes The model developed for this research is described mainly in terms of a single individual's representations for a single object. First, an overview of the model is presented. Second, the various components of the model are defined. Third, extensions of the model to situations involving more than one object and/or more than one person are defined. And, finally, comparisons of the model with the basic features of related models are considered. Overview of the Model Basically, this model is designed to describe what happens psychologically when a person becomes aware of an object. The general form of the model is one which is familiar in cognitive psychology. Figure 1 illustrates only the main characteristics of the model. Feedback loops and other complicating factors related to the origins of symbolic meanings are omitted for the sake of simplicity. At the left of the figure, the two major sorts of input to the system are represented--the perception of physical 14 . .. u . ..-.u I vs .fus. unus- ~ :I--- ~ 15 mmcwcmwe Hoeuo mcoem mQHeOCOwumawu Bonn po>wuop mcflcmoe m Ammmcommwm Hmouo>v Domnno em cues owumfloomm< I nonwcmmz \. \ mo uuommm O>wuowmmouucH c< \: .mmmmoooum O>wuflcmoonamoumwonmm PMI ‘ \M — uoohno nonhuman uo Hmowmzem mamcwm m newswmop mmcwcmoz owaonshm mo Emummm e \\ soumhmnzm one 4/ madame OHHOOESm Omum>wuom c4 xOHmEOO «0 proz 4 .H magmas Domhno uomuumna no mo cowumwocoo .HO uomflno Hmoflmsea m mo cowumoouom 16 objects and the conception of abstract objects. The center portion represents the broad system of symbolic meanings which includes one's cognitive definitions of various objects in the real world as well as one's self-definition, and the center of this region represents one's definition of the object perceived or conceived. The rightmost part of the figure represents verbal output from the system, verbal expression of the symbolic meanings defining the object that was perceived or conceived. Essentially, Figure 1 illustrates that awareness of some physical or abstract object activates meanings within various portions of a broad system of symbolic meanings. These meanings are assumed to be activated because of previously established connections with characteristics of the object in question. Together, these meanings contribute to a person's definition of the object. Once formed, however, the definition should not be regarded as static. The meanings intitially activated are likely to undergo reorganization from time to time. Such reorganization may also lead to the synthesis of new symbolic meanings constituting an emergent by—product of relationships amOng other meanings. While both verbal and nonverbal symbolic meanings are assumed to be among the meanings that define an object, it is also assumed that an introspective verbal report of meanings is reasonably representative of the total meaning-system. Although this procedure is admittedly incomplete, the expectation is that appropriate analyses of verbal meanings can uncover the major organizational characteristics of the underlying system. 17 Definitions The definitions presented in the following sections develop the conceptual framework of the model. The first definition pertains to the basic units or "elements" (symbolic meanings) which define a single individual's perceptions of a single object. Since, in the model, each "element" is regarded as a subsystem in its own right, the second definition specifies the major characteristics of each symbolic meaning. Once the components of a single symbolic meaning have been described, it is possible to define relationships between symbolic meanings. Fourth and finally, the organization of meanings comprising a person's definition of an object is defined. Elementary Cognitive Subsystems.-—In the present model, each symbolic meaning is called an elementary cognitive subsystem, and these subsystems constitute the building blocks of the model. Definition 1: An elementary cognitive subsystem is a denotative or conno- tative meaning associated with either a physical or abstract object. While an elementary cognitive subsystem constitutes the basic unit of analysis, this subsystem should not be regarded as irreducible. Instead, as the next definition points out, each elementary cognitive subsystem is regarded as having two major dimensions. The definition given above is very similar to Fishbein's (1967a, 1967b, 1967c) definition of "beliefs about" an object. The Fishbein definition states that "beliefs about" an object are beliefs 18 in the existence of a relationship between an object and some attribute, goal, concept, or other object. A major difference between the definition given above and the Fishbein definition is that in the present approach inter-object relationships are excluded from the definition of an elementary cognitive subsystem. Since objects are assumed to be defined by a system of cognitive "elements," relationships between objects are regarded as complex matters to be handled through data analysis. Components of Elementary Cognitive Subsystems.--Each elementary cognitive subsystem is regarded as having two major dimensions or components. Definition 2: An elementary cognitive subsystem has two major dimensions: (1) strength of association with an object, and (2) valence toward that association. These two dimensions (or components) of an elementary cognitive subsystem may be represented by the axes of a two-dimensional subspace. Any particular symbolic meaning may be represented as a point in the subspace as shown in Figure 2. Although the two components are con- sidered to be orthogonal, strength of association and evaluation of the association are viewed as interdependent psychological processes. Furthermore, while later research may reveal more appropriate units for the strength of association and valence dimensions, equal units are used for both scales as a first approximation. 19 A point representing a meaning associated with an object High ' I I I I l *""'1 I Strength I I of : I Association . | I . l ' l l I I I ' I I LOW 1 4 - 0 + High Negative High Positive Valence Figure 2. A Spatial Representation of the Components of an Elementary Cognitive Subsystem. 20 For many years now, cognitive theorists have regarded beliefs and attitudes as rather different psychological phenomena. The above definition attempts to reverse this situation by viewing attitude as a derivative of the evaluative (or valence) components of all the beliefs linked to some object (association strength). The view expressed above is consistent with recent efforts to explore relationships between evaluative and non-evaluative responses to objects. For example, beliefs and attitudes have been described as having both affective and cognitive components (Fishbein, 1967b, 1967c; Krech, Crutchfield, and Ballachey, 1962; Rokeach, 1968) and Rokeach has theorized that these elements are interdependent and that they reinforce one another. Similarly, Osgood, EE_2£° (1957) have obtained factorial evidence that every point in a semantic space has an evaluative (valence) component. Relationships Among Elementary Cognitive Subsystems.--The preceding definitions establish the conditions which make it possible to calculate relationships among elementary cognitive subsystems. Definition 3: The relationship between any two elementary cognitive sub- systems is the Euclidean distance between the points representing the symbolic meanings under consideration. To visualize how the relationship between two symbolic meanings may be measured, think of using the two-dimensional subspace described 53*7Ve to plot the points representing two different symbolic meanings. 21 The relationship between two meanings is simply the distance between the two points. This situation is illustrated in Figure 3. Since the axes of the two-dimensional representation are identical for all symbolic meanings, the distance between any two meanings may be calculated in exactly the same manner. Moreover, matrices of relationships may be calculated for a large number of symbolic meanings. Cognitive Object.--Given that the meanings associated with an object constitute an individual's definition of that object, the content and structure of these meanings may also be considered. Definition 4: A cognitive object is a complex constellation of elementary cognitive subsystems, where the relationships among elementary cognitive subsystems determine the organization of this constellation. Whereas elementary cognitive subsystems specify the cognitive details of a perceived object, a cognitive object is the organization of these details and this organization may be quite complex. Although an elementary cognitive subsystem may be represented as a point in a simple two-dimensional subspace, the organization of a large number of points may be complex and nonlinear in form. For example, the system may be hierarchically organized as shown in Figure 4. This conceptualization of the cognitive structure underlying a person's representations for an object is similar to the idea that an object may be viewed as a semantic space (Osgood, et a1., 1957). 22 A point representing a meaning associated with an object The distance A///”between two High meanings associated with an object Strength of Association r—A point representing a second meaning associated with an object Low - 0 + High Negative High Positive Valence Figure 3. A Spatial Representation of the Relationship Between Two Meanings that are Associated with an Object. 23 A point representing an elementary cognitive subsystem A pair of closely related elementary cognitive subsystems Figure 4. A Hierarchically Organized System Representing a Cognitive Object. 24 The main difference is that here it is unnecessary to restrict the space to a system of linear bipolar dimensions. Instead, the space may have nonlinear characteristics of various sorts. The definition of a cognitive object is also related to the notion of a habit-family hierarchy as adapted from Hull by Fishbein (1967a). In this case, however, the model differs in that the "elements" of the hieararchy are elementary cognitive subsystems rather than beliefs. Finally, the definition of a cognitive object has implications for the operationalization of several concepts used by Rokeach (1960, 1968). Specifically, Rokeach describes a belief-value system as a hierarchically organized structure wherein the cognitive elements (i.e., elementary cognitive subsystems as defined here) vary in centrality and connectedness, and the system as a whole varies in integration, differentiation, and other organizational properties. Assuming that the operationalizations of the present model are satisfactory, it may in turn be possible to operationalize methods for examining higher-order system characteristics such as integration, differentiation, and the like. To date, Rokeach has not devised operationalizations at the systems level. Extensions of the Basic Model Thus far the model has been described with respect to a single individual's reactions to a single object; however, the model may be readily extended to situations involving more than one object and/or more than one person. 25 A Cognitive System.--The first logical extension of the model concerns the representation of one person's responses to more than one object. Definition 5: A cognitive system is a constellation of cognitive objects. Since the cognitive objects comprising a cognitive system are made up of elementary cognitive subsystems, the object systems should overlap to the extent that they have their origins in common or related experiences. A spatial representation of such a cognitive system is shown in Figure 5. This definition of a cognitive system is crucial to the present research. It provides much of the rationale for seeking relationships between personality traits and consumer-product perceptions, since the cognitive processes associated with these two rather different classes of objects are viewed as possible parts of the same general system. It should also be noted that relationships between objects are not specified by simple statements concerning some overall relation- ship between two objects as assumed in Fishbein's (1967a, 1967b, 1967c) definition of "beliefs about" an object. Rather, relationships between objects take into account the many specific relationships among the elementary cognitive subsystems defining each object. Multiple Cognitive Systems.--The model may also be extended to inelude the meaning-systems (or cognitive systems) of more than one Person . A cognitive object A cognitive object having elements in common with several other objects ‘ A relatively isolated cognitive object Figure 5. A Spatial Representation of a Cognitive System Including Several Cognitive Objects. 27 Definition 6: A multiple cognitive system is a super-constellation of the cognitive systems for different individuals. When several persons have responded to several objects in terms of a common set of verbal meanings, it is possible conceptually to "superimpose" the mappings of the several general cognitive systems involved. When this combining is done, the overlapping portions of different cognitive systems may be regarded as constituting inter- person commonalities} in perceptual—cognitive representations. If data for persons with similar viewpoints are analyzed on a composite basis, the commonalities should be substantial. Moreover, Simi- larities in one perceptual-cognitive domain (e.g., personality traits) may be related to commonalities in other domains (e.g., consumer-product perceptions). Comparisons With Other Models Although the present model is related to existing psychological models at several levels, Similarities and differences are most 1The word "commonality" rather than the word "communality" has been used since it comes closer to reflecting the characteristics of definitions and operationalizations employed in the present research. Commonality may be defined as possession with another of certain attributes while communality may be defined as concordance or agreement in opinion throughout a group (see Webster's Third New International Dictionary, Merriam Company, Springfield, Massachusetts, 1963). Commonality was selected so there would be no confusion with factor analytic literature where communality implies something quite different from possession of common attributes. On the other hand, one should be careful to avoid confusion with another definition of commonality which refers to a body corporate or to common people. 28 easily examined in terms of the basic characteristics of each related model. The following paragraphs describe the essential character- istics of elementary cognitive subsystems and compare these with what the writer views as the essential characteristics of consistency and balance models of cognitive structure and attitude organization. According to the present model, elementary cognitive sub- systems are symbolic meanings associated with objects. These subsystems may be described in terms of four characteristics: (1) content (symbolic meaning), (2) an object (physical or abstract), (3) a strength of association relationship, and (4) a valence attached to the perceived meaning-object relationship. While strength of association and valence are viewed as separate components, they are assumed to be interdependent psychological processes. Furthermore, the object being perceived or conceived helps to establish the context for particular meanings. To facilitate the comparison of elementary cognitive subsystems with the characteristics of other models, the subsystem has been represented as shown in Section g_of Figure 6 rather than as the two-dimensional subspace described earlier. A consistency model of cognitive structure may be described rather similarly. For example, the theories of Cartwright and Harary (1956), Heider (1946), and Newcomb (1953) may be described in terms of four characteristics: (1) symbolic meanings, (2) objects, (3) unit (U) relationships which are essentially strength of association relationships for particular meanings, and (4) liking (L) relationships 29 a. Elementary Cognitive Subsystem (Present Model) [Strength of Association Meaning T l [ Object J IValence of AssociationJ b. Heider (and other balance theorists) [Unit (U) Relationship ( meaning relationship) I Object [ Object l lLiking (L) Relationship ( valence relationship) c. Osgood [ Meaning J/Erength of Association _ - l Object I Evaluation — - (a variety Strength of Association 7 of meaning) d. Fishbein ‘ Strength 0 _ vaaluation Associatio I [ Meaning 1 l I Strength of [ Object J lAssoc1ation| I 7 Figure 6. A Comparison of an Elementary Cognitive Subsystem with the Cognitive Units of Other Models. 30 which are essentially valence relationships. However, despite these similarities, there are some fundamental differences between such models and the present model. In the first place, the theories focus on object-object relationships rather than meaning-object relation- ships. Second, unit and liking relationships between objects are viewed as independent rather than interdependent psychological processes. Third, the liking relationship is concerned with the objects as a whole rather than with the affective significance of a particular meaning-object relationship. This situation is represented in Section p_of Figure 6. Behavioral models of cognitive structure and attitude formation, such as those developed by Osgood, 35‘s}, (1957) and Fishbein (1967a, 1967b), may also be described in terms of four characteristics: (1) symbolic meanings, (2) objects, (3) strength of association relation- ships, and (4) evaluation (or valence). However, the differences between such models and the present model are considerable. While Osgood has suggested that every meaning-object relationship has an evaluative component, both his model and operationalizations treat evaluation as a separate dimension of meaning (see Section g_of Figure 6) rather than as a component of each dimension. Whereas Fishbein has recognized the importance of examining the evaluative aspect of an association with an object, he considers it possible to measure the evaluation of a belief (or meaning) independently of the association between a meaning and an object (this situation is represented in Section g'of Figure 6). In other words, while Fishbein assumes a multiplicative relationship between a belief and its 31 evaluative significance, evaluative information is gathered outside the context of a meaning-object relationship. That is, the evaluative response is toward the meaning in abstract rather than toward a particular meaning-object relationship. In summary, the advantages of the present conceptualization over the models described above are that: (1) only meaning-object relationships are considered, thereby avoiding oversimplification of inter-object relationships, (2) evaluation is viewed as a component of every meaning-object relationship, and (3) association strength and valence (evaluation) are viewed as interdependent processes which should be measured in the same meaning object context. Furthermore, since the present model possesses some of the characteristics of both balance and behavioral models, it may be regarded as an effort toward unifying these approaches. A Strategy for Studying Complex Perceptual-Cognitive Systems and Comparison With Other Strategies In order to test the model described in the preceding section, it was necessary to utilize some sort of psychological content. Two rather different sorts of content were purposely chosen for the research: (1) responses to several personality inventories, and (2) responses to a set of consumer products. Personality inventories were selected as points of reference for examining relations between general cognitive characteristics and more specific perceptual characteristics. That is, traditional personality measures were assumed to be reasonably capable of mapping broad psychological characteristics. 32 Operationalizations of the model's constructs included the development of both data collection techniques and methods for mapping and comparing consumer-product perceptions. The methodology was believed to be freer of artificialities and methodological constraints than was the case with earlier methods. The data collection technique used to gather object perception data was an operationalization of an elementary cognitive subsystem. Furthermore, analysis methods constituted operationalizations of relationships among elementary cognitive subsystems, of the organization of cognitive systems, and of the organization of multiple cognitive systems. The remainder of this chapter is divided into three main sections. The first section gives an overview of the present research strategy. The second describes general characteristics of the data collection techniques and compares them with existing techniques. The third section describes the general form of the analysis methods and compares them with existing methods. Overview of the Research Strategy For purposes of the present research, personality data were used in forming groups of persons. Each group was selected so as to be homogenous yet widely separated from every other group of persons. Each group was formed under the assumption that persons with similar personality characteristics represented a personality "type." For each separate personality "type," the consumer-product perceptions of its members were then mapped. Some of these mappings dealt with typal commonalities in the content and organizational characteristics of consumer-product perceptions; others mapped 33 organizational characteristics of perceptions for the individual members of personality "types." Once the mapping of consumer-product perceptions had been performed for separate "types," the results for contrasting "types" were compared. The content and organizational characteristics of these mappings were differentiated using a variety of methods. In each case, however, the purpose was to test hypotheses concerning relationships between personality characteristics and consumer- product perceptions. It was expected that these two rather differ- ent classes of cognitive phenomena would be related with respect to both content and organization. In summary, much of the research strategy may be viewed as a mixture of what are described in factor-analytic literature as Q, R, and P techniques (Guilford, 1954). The main features of Q technique were represented in typological analyses of personality data. R technique was represented by cluster analyses of product- perception data for separate personality types, and the more intensive analyses of product-perception data for small groups and individuals corresponded to P technique. While the methods for differentiating between personality types fall outside this classification scheme, they have characteristics in common with such methods as discriminant analysis, automatic interaction detection, and several of Tryon's (Tryon and Bailey, 1970) typal differentiation methods. While this investigation was limited to studying relationships between personality traits and consumer-product perceptions, the methodology may certainly be used with many other sorts of variables. 34 For example, relationships between sociological characteristics and product perceptions or relationships between physical contexts and person perception might be investigated in a similar manner. The possibilities are endless. Data Collection The techniques described below were used to collect data with which personality characteristics could be mapped, consumer-product perceptions could be mapped, and the perceptions of different types could be differentiated. Technigues for Collecting Personality Data.--Several existing personality inventories were selected for the study. These inventories served as points of reference for testing the present model. 1. Assumptions.--A person's personality characteristics (or traits) were assumed to constitute abstract cognitive objects, where these objects were defined by sets of closely related elementary cognitive subsystems (see Definition 1, page 17). This.assumption appeared reasonable in light of the fact that personality inventories are often constructed by means of item analysis which attempts to locate clusters of closely related items. 2. varieties of personality measures used.--Both univariate and multivariate personality measures were used. The univariate measures focused on structural characteristics of one's personality (i.e., cognitive complexity and dogmatism). The multivariate measures focused on content features of personality (i.e., orientations to various social situations and personal values). 35 Techniques for Collecting Product-Perception Data.--Two techniques for collecting perceptual-cognitive data were used in this research. One pertained to overall evaluations of objects and the other pertained to detailed associations with objects. Since most of the analyses were based on data gathered with the latter technique, only this technique will be discussed here. The technique for gathering detailed associations with consumer products was an effort to operationalize the concept of an elementary cognitive subsystem, and it was devised since existing techniques failed to provide an adequate operationalization of the concept. Data pertaining to the symbolic meanings that people associate with consumer-products (i.e., sterling silver tableware) were obtained by presenting respondents with a large number of attributes and several tableware patterns. For each pattern, SS were asked to consider each attribute in turn and indicate: (1) the extent to which each attribute seemed to apply to the pattern (strength of association), and (2) the degree to which they liked or disliked this perceived association with the object (valence). gs were also asked to think of attributes which contrasted with the ones presented and to react to these using the applicability and liking scales. 1. Assumptions.--A sterling silver tableware pattern was assumed to constitute a physical object which would be defined by a system of elementary cognitive subsystems. By gathering data on the elementary cognitive subsystems, it was thought to be possible to analyze the characteristics of the meaning-systems underlying a person's perceptions of this sort of consumer product. 36 2. Other characteristics of the main data collection technigpe.--The characteristics listed below helped to make the principal data collection technique rather unlike other techniques for collecting product-perception data: Large amounts of perceptual-cognitive detail were collected. Respondents were not asked to analyze or summarize their reactions but just to give first impression responses to the attributes presented. The procedures were simple enough for gs to react quickly. The measurement procedures did not oversimplify perceptual- cognitive processes (i.e., while the task was the same for all attributes, gs had considerable latitude regarding the attributes to which they responded and the form of the response). Unnecessary linearity constraints were avoided. The technique provided sufficient depth of information for investigating within-person phenomena. Comparisons With Other Techniqges for GatheripgyPerceptual— Cognitive Data.--While the personality inventories represented traditional approaches to collecting data on general characteristics of individuals, the technique for collecting detailed associations represented a considerable departure from other approaches to gathering masses of perceptual detail. The fundamental differentiating characteristic of the present technique for collecting potentially complex perceptual-cognitive data was that associations with objects were viewed as joint functions of 37 association strength and valence responses obtained in the same meaning-object context. While a number of theorists have recognized the importance of measuring the evaluative significance of each association with an object, the present technique is unique in its operationalization of the idea (see Comparisons With Other Models, pp. 27-31). Moreover, measuring the evaluative significance of associations appears to facilitate the identification of complex, nonlinear differences among individuals and types. Several commonly used approaches to collecting complex perceptual-cognitive data are discussed below. 1. Multidimensional scaling methods.--When psychometric methods are extended for use with perceptually complex stimuli, only rather vague global responses are obtained (Torgerson, 1958). As a result, sensitivity to potentially important individual differences may be prevented by the gross form of the responses given, and understanding of the behavior in question may be hindered by the absence of cognitive detail. In other words, such techniques impose severe constraints on the variety of information gathered and on the form of the responses. 2. Semantic Differential.--While a technique such as the Semantic Differential (Osgood, g£_2i,, 1957) allows respondents to express a variety of associated meanings, it requires them to make ratings along bipolar dimensions that are not necessarily perceived as opposites. Again, sensitivity to potentially important individual differences in an underlying cognitive system may be reduced. 38 While sensitivity to individual differences is increased in the Fishbein and Raven (1962) adaptation of the Semantic Differential for purposes of attitude measurement, a number of constraints remain. On the positive side, both belief strength and belief evaluation are measured. On the negative side, respondents are constrained to using bipolar response scales which may not allow them to express what they think. 3. McQuitty's approach to collecting detailed perceptual— cognitive data.--McQuitty, Abeles, and Clark (1970) have developed a technique which avoids bipolarity constraints as well as linearity constraints. Respondents were asked to express dichtomous responses for a rather large number of attributes, and the data were handled as nominal level information. However, what their technique may gain by relaxing the constraints of the Semantic Differential, it may lose in the fact that response scales are reduced to two choices. 4. Summary of other data collection techniques.--The three data collection techniques mentioned above have a number of dis- advantages. These disadvantages include: (1) obtaining oversimplified (or global) responses, (2) forcing respondents to react to attributes in terms of bipolar dimensions, (3) failing to regard evaluation as an aspect of each association, and (4) severely restricting the range of responses. The data collection technique developed for the present study as an operationalization of elementary cognitive subsystems sought to overcome these disadvantages and simultaneously offer greater measurement flexibility. 39 Data Analyses The analysis methods employed in this research were of three main types: (1) classificatory analyses of personality data, (2) mappings of the perceptual-cognitive systems underlying consumer- product perceptions, and (3) analyses of relationships between these two rather different domains of cognitive functioning. The first mode of analysis was concerned with establishing the conditions for typological analyses of differences among groups of persons. The second constituted the variety of ways that consumer—product per- ceptions were analyzed to reveal the content and organization of these perceptions. The third constituted the various ways that the impact of personality upon consumer-product perceptions was examined. Classificatory Analyses of Personality Data.--The first step in testing the present model was to establish conditions favorable to revealing complex relationships between personality and consumer- product perceptions: that is, to perform the first type of analysis in order that the other two types might be performed. To establish such conditions, relatively small homogeneous groups of SS were formed. Each group represented an empirically defined personality "type" which was maximally different from other "types" identified through use of the same personality measure. The rationale for using small groups was as follows. While analyses for individuals may reveal configural response characteristics, results are not easily generalized. On the other hand, if large groups are studied, differences in kind are likely to be obscured and important information may thereby be lost. The present research seeks 40 to achieve a realistic balance between the specific and the general. By using small groups that are relatively homogeneous, the present approach attempts to remain sensitive to the perceptual-commonalities among the members of particular personality "types" while simultane- ously obtaining results that may be generalized to these same types in a broader population. For univariate personality measures, subgroups were formed by selecting the persons who scored highest and the persons who scored lowest on a particular measure. The subgroups were kept small so that within-group similarity would be high relative to between-group similarity, and group size was restricted to ten (a major consideration here was the expense of processing larger amounts of data). For multivariate personality measures, the situation was more complex. In this situation, homogeneous subgroups were identified by cluster-analyzing interrelationships among the response profiles of the individuals in the total sample. The purpose of such analyses was to identify subgroups for which within-group similarity was high relative to between-group similarity. l. Assumptions.--The ways in which the personality data were handled derive from two assumptions. First, people were assumed to differ in kind as well as in degree, and the meaning-system defining an object may, therefore, vary in many ways from one person or group to another. Second, some individuals‘were assumed to have enough in . common that they may reasonably be classified as representatives of a particular viewpoint (or type). 41 2. A "type" as a statistical concgpt.--As employed here, the concept of a type is empirical and is tolerant of variation, rather than simply being a template which must be fitted perfectly. A definition of a pure "type" which may be extended to meet these conditions has been offered by McQuitty (1967). "In a statistical sense, a type is acategory of persons wherein everyone is more like every other person in the category than he is like any person in any other category" (McQuitty, 1967, p. 23). The above definition indicates that a type should be defined relative to the characteristics of the population under consideration. As long as subgroups are relatively homogeneous and are isolated from one another, the definition will be satisfied. The definition does not require that the members of types be virtually identical, provided that the conditions of the definition are satisfied. What the definition does not take into account is the fact that all behavior has an error component. To the extent that inter- person distances include error components, it is inappropriate to employ an absolute definition of a "type." In other words, it would be desirable to use the definition as the ideal and develop indices reflecting the degree to which an absolute definition has been satisfied. While the methodology of the present research does not include any formal tests of degree of fit, procedures used to identify personality "types" assumed that error components were a part of the data. The classificatory analyses Simply sought the most homogeneous yet distinctly different groups of gs. Furthermore, given this error 42 variance, it is likely that relationships found between personality and product perceptions will underestimate actual relationships. Methods for Mapping Consumer-Product Perceptions.--Several methods for analyzing consumer-product perceptions were developed especially for this research because existing methods were not consistent with the characteristics of the present model. These new methods were concerned with the featured content of meaning-systems, with the response-style characteristics of perceptions, and with the structure of meaning-systems. These three aspects of cognitive functioning were examined for two reasons: (1) much theory in the social sciences focuses on one or another of these levels of abstraction as a means of describing human behavior, and (2) most measurement techniques used in the social sciences focus on the content of responses, the manner in which responses are given, or the relationships among responses. The content and response-style analyses were performed on group-composite data. Some of the structural analyses were also performed with group-composite data; others were performed separately for each individual. While the formation of contrasting personality subgroups (or "types") may establish the conditions for studying perceptual- cognitive commonalities, the methods for analyzing consumer-product perceptions were entirely independent of the particular persons included in any given analysis. On the other hand, it should be pointed out that the same sorts of analyses were performed for each separate personality subgroup. 43 l. Assumptions.--The analyses of consumer-product perceptions were based on three main assumptions. First, it was assumed that people's responses have the characteristics of syndromes. Second, it was assumed that the meaning-system defining an object may be complex and nonlinear in form. Third, perceptual—cognitive commonalities among people were assumed to manifest themselves in the content and organizational characteristics of meaning-systems. 2. Content anaIySeS.--The procedures for analyzing the content characteristics of perceptual-cognitive commonalities involved three steps. First, distance matrices of relationships among elementary cognitive subsystems (or symbolic meanings) were calculated for the system of symbolic meanings defining a multiple cognitive system (i.e., the perceptual responses of the members of a personality subgroup). The calculation of each relationship in a matrix was based on a summation across inter-attribute distances for all the objects to which each subgroup member responded. Second, the distance matrix was cluster-analyzed with a hierarchical clustering method developed by the writer (Price, 1969). This clustering method was developed in an effort to avoid the several methodological difficulties inherent in McQuitty's pattern- analytic methods (refer to Cluster analyses, pp. 51—53), and it was used here to reveal the hierarchical organization of a multiple cognitive system. Third, the featured content of a system was identified. In other words, content clusters were identified for which within-cluster distances were low relative to the distances between these clusters and other parts of the meaning-system. The 44 approach was closely related to procedures used in pilot research exploring the use of other hierarchical methods for analyzing consumer-product perceptions (Price, 1968). 3. Response-style analyses.--Behavior at the level of separate elementary cognitive subsystems was studied by forming bivariate frequency-distributions for symbolic meanings. The data in the distribution for each symbolic meaning constituted the strength of association and valence responses that respondents within a particular personality subgroup expressed for each of several objects. The distributions were summarized in two main ways: (1) by tallying the frequency with which content was associated with objects, and (2) by calculating a weighted average that summarized the joint distribution of association strength and valence responses. Precedence for examining response characteristics derives from test development research and personality research. In the former, response sets or styles constitute troublesome biases. In personality research they are often measured as a means of differentiating among persons. The present typological approach also allows response sets to be measured as possible means of differentiation. Provided that within-group commonalities are reasonably high, the present approach may reveal response sets that differentiate between personality I: types . n 4. Structural analyses.--The organization of meaning—systems was analyzed in two main ways. The first approach entailed the use of factor analysis, and the second involved cluster analysis and a new 45 method for calculating the overall similarity of hierarchical systems (Price, 1970). Several sorts of matrices were analyzed in both instances. Attribute by attribute matrices were formed for each of the subgroups yielded by the classificatory analyses of personality structure data. These matrices were formed as described in the section on content analyses. Additional attribute by attribute matrices were formed by separating responses to objects liked from responses to objects disliked and then calculating matrices for each subset of the data. In other words, two matrices were calculated for each personality subgroup (e.g., the two subgroups defined by the dogmatism variable). similarity transformations of these matrices were then factor analyzed. The clustering and structure-similarity analyses were likewise performed with some matrices based on responses to liked and disliked objects. In this case, however, these two sorts of matrices were formed for each individual in each subgroup yielded by a classi- ficatory analysis of dogmatism data. After these matrices were cluster analyzed, the overall similarity of the two hierarchies was measured with the new method for evaluating structure similarity. Methods of Differentiating Between CognitivegMappings for Different Personality Types.--While analyses of the consumer-product perceptions for each separate personality subgroup may in fact reveal complex nonlinear characteristics, there remains the problem of determining the manner in which subgroup characteristics differ from one another. This problem of differentiating between subgroups 46 was handled by developing several methods for further analyzing and comparing the results of the previously described analyses. These methods are closely tied in with both the model of perceptual- cognitive processes and the data collection techniques developed for this research. 1. Assumptions.--The methods for differentiating between the consumer-product perceptions of contrasting personality subgroups were based on three assumptions. First, these two very different classes of phenomena were assumed to be related through meaning- systems underlying each class. Second, it was assumed that the relationships would be complex and nonlinear. Third, it was assumed that personality inventories concerned with content as well as those concerned with personality structure would have implications for both content and organization of consumer-product perceptions. 2. Content differentiations.--The featured content groupings yielded by each cluster analysis of consumer—product perception data were summarized as a content vector. The vector was formed by assigning a weighted average of association strength and valence responses to each featured attribute and then ordering the content from highest positive to highest negative weighted average. The content vectors for contrasting personality subgroups were then compared in order to identify content that differentiated between these personality "types." To make this differentiation, a second pair of vectors was formed. These vectors represented the oOntent that appeared in the first vector for one personality "type" 'vv 1*: ’U ‘- ..~ fn 47 but did not appear with the same sign or by implication (a contrasting idea with opposite sign) in the first vector for the contrasting personality "type." To the extent that these differentiators revealed a unified viewpoint for each personality "type," they constituted a basis for interpretation. 3. Response-style differentiations.--Contrasting personality subgroups were also differentiated in terms of: (l) the frequency with which attributes and contrasting ideas were associated with sterling silver tableware patterns, and (2) the weighted averages for association strength and valence responses. The frequency data were compared by simply counting the number of times that usage of an attribute or contrasting idea was greater for one of the personality subgroups. The weighted averages were used by counting the number of times: (1) a more extreme weighted 'average (absolute) was obtained by one subgroup, (2) a more positive (signed) weighted average was obtained, and (3) more extreme reactions were positive and extreme reactions were negative for one subgroup. 4. Structural differentiations.--The structural character- istics of consumer-product perceptions for contrasting personality "types" were differentiated by comparing the factor analysis results for these "types" and by comparing structure—similarity results for the individual members of these types. Factor analysis results were compared by counting the number of times that an attribute and its contrasting idea had highest 48 loadings on the same factor. This approach was used for both the analyses of data based on summations across all objects and the separate analyses of responses to objects liked most versus objects liked least. Factor analysis results were also compared in terms of the proportion of variance accounted for and other indices. The structure-similarity results were compared by examining the distributions of similarity indices for contrasting personality subgroups. Each value in the distributions represented the simi- larity of the meaning-system underlying the objects liked versus the system underlying the objects disliked. Copparisons With Commonly Used Methods of Analyzing Compigx Perceptual-Cognitive Data.--In this section, a number of other analysis methods are compared with methods developed for the present research. First, several classificatory methods are considered. Then, several methods for mapping relationships among variables are examined. Finally, a number of methods for differentiating between groups are discussed. 1. Other classificatory methods.--The methods used here to identify personality "types" bore a strong resemblance to some existing methods. Despite the general similarities, however, the present methods differed in some important ways. a. Known-groups method.--The handling of univariate personality data bore some resemblance to the "known-groups" approach to test validation. As with the known-groups method, contrasting groups, or scission types (Stephenson, 1953, pp. 158-164), were identified from 49 a criterion variable and compared in terms of some other behavior (consumer-product perceptions). The main difference between the usual application of a known-groups method and the present analyses was that psychological information (personality data) rather than demographic, sociological, or political information was used to identify groups. Another difference, and a potential weakness, was that the criterion data was obtained from the persons being classified. b. Typological analyses.-—While the present handling of multivariate personality data was inspired by the Q-methodology of Stephenson (1953) and hierarchical clustering methods developed by McQuitty (e.g., 1959, 1960, 1961, 1963, 1966a, 1966b, 1966c, 1967) the intent of these analyses was quite different. Typically these other methods have been used to analyze a single body of data. That is, the data have been used to identify "types," and the "types" have then been compared in terms of the data configurations which defined them. In the present research, cluster-analyses of inter- person relationships were performed to identify ”types" and the "types" were subsequently compared with respect to other behaviors. The sorts of multivariate methods appropriate to identifying personality "types" are also appropriate for mapping cognitive systems. The relative merits of such methods will be examined in the next section. At this point, let it suffice to say that the McQuitty variety of clustering was used since it relaxes a variety of constraints imposed by other methods. 50 2. Other cogpitive mapping methods.-—Clinical psychologists, personality theorists, and others concerned with cognitive organi- zation have given considerable attention to the concept of a syndrome (McQuitty, 1959) and to mappings of systems having nonlinear character- istics (e.g., the graph theoretic concepts of Cartwright and Harary, 1956). Some of the other analysis methods discussed here are suited to analyzing syndromes and systems; others are not well suited because of linearity, normality, and other constraints. a. Correlation and agreement matrices.--Whereas the present analyses used mainly dissimilarity matrices, other research concerned with cognitive mapping typically employs similarity matrices of various sorts (e.g., correlation or agreement matrices). When research is based on a diverse set of variables having different measurement scales, correlation is particularly useful. A correlation matrix reflects the patterns of differences between Observations and controls for differences in magnitude. For the present research, however, measurement scales were constant for all object perception data, and the conceptualization of an elementary cognitive subsystems suggests that both pattern and magnitude should be taken into account (each meaning is a point in a two-dimensional subspace). The measurement of distance relationships can take these factors into account, and the two-dimensional subspaces are suited to such calculations. Agreement matrices, such as those typically calculated for {McQuitty's pattern-analytic methods, reflect pattern and magnitude under the constraint that only identical values in corresponding 51 observations are considered. AS a result, agreement matrices are most appropriately calculated from nominal level data. If the data are continuous, pattern and magnitude are more sensitively measured with distance calculations. In Factor analyses.--While R- and P-type factor analyses have been employed for mapping complex systems, the linear model of factor analysis is incompatible with many aspects of the present model of complex perceptual-cognitive processes. For example, an elementary cognitive subsystem is defined as a two—dimensional subspace and a meaning is represented by the combination of intensity of association and valence components. Whereas distances between points in the subspace can be calculated, the form of the subspace is such that these distances cannot be translated into correlations. In addition, there is little reason to expect that a cognitive system can be parsimoniously described in terms of either an orthogonal or oblique axis system. To the extent that complex configurations define groupings of subspaces, factor analysis may obscure all but the most clearcut groupings and yield factors that account for rather little variance. Nevertheless, it remains possible that the mathematical power of factor analysis may have to be carefully balanced against losses in the ability to map certain varieties of relationships and against the necessity of imposing linearity, normality, and absence of interaction constraints. c. Cluster analyses.--While a considerable variety of clustering methods has been developed in recent years, only a few 52 methods are sufficiently similar to those used in the present research to be discussed here. Some of these methods, e.g., key- cluster analysis and cluster structure analysis (Tryon and Bailey, 1970), have much the same intent as the present analyses yet they utilize factor analytic methods for reducing a space and therefore impose linearity and other constraints. On the other hand, methods which relax some of the constraints of factor analysis are generally less concerned with the problems of examining cluster structure. Included here are a matrix ordering method developed by Hunter (1968), an iterative method of Euclidean distance clustering (Tryon and Bailey, 1970), hierarchical clustering methods developed by Johnson (1967), and an entire array of hierarchical pattern-analysis methods developed by McQuitty and colleagues (e.g., McQuitty, 1959, 1960, 1961, 1963, 1966a, 1966b, 1966c, 1967; McQuitty, Price, and Clark, 1967; McQuitty and Clark, 1968; McQuitty, 1971; McQuitty and Frary, 1971). While some of these methods relax parametric constraints, each has its drawbacks. For example, the Hunter method tends to yield only gross groupings of variables. The Tryon method of distance clustering tends to make complex relationships appear overly simple since overlapping groupings cannot be formed. The Johnson methods use matrix reduction procedures which can be shown to seriously distort many kinds of relationships in a matrix. Finally, many of the McQuitty methods employ matrix reduction procedures which are very similar to Johnson's techniques as well as procedures that emphasize hand calculation--considerations that are both unnecessary and unfortunate in this era of computers. , i I. fly! '0... .7 I! a... 1,... :- 'Ih-u . ‘4': "o -. 'I‘ -. “on. d ‘5. ‘¢-- . ( _‘ 1,, -_A I" 53 While the McQuitty variety of clustering seemed to seek objectives which were compatible with the complexity of the present model, the writer found it necessary to develop an alternative method of hierarchical clustering (Price, 1969). The advantages of the method developed for this research are threefold: (1) the method is oriented toward the use of computers and avoids restrictions which merely facilitate hand calculation, (2) the method works with the original matrix of relationships throughout an analysis and, thereby, avoids matrix reduction procedures which distort relationships, compound decision errors, or fail to take chance variation into account, and (3) by allowing for overlapping clusters, the method makes it possible to map relationships which cannot realistically be handled in a simple manner. 3. Other methods of differentiating_between group_.--While there is a considerable variety of methods for differentiating between groups, rather few of these seem well suited to differentiating between groups which are defined by configurations, syndromes, or system characteristics. The methods described in Section A through G below represent some of the principal methods available for differ- entiating between groups. They are presented roughly in order of their ability to cope with complex systems. a. Item analysis.--One of the simpler methods of differ— entiating between groups is item analysis (Gulliksen, 1950). The method determines the extent to which individual items predict the values of a criterion variable (in this case, a dichotomous variable 54 representing membership in contrasting personality subgroups). The predictive power of each item is examined separately from any other item's predictive power. The method assumes that item responses are linearly related to the criterion variable and that response distributions are normal. Prediction is best when items account for different portions of the variance in the criterion variable. In contrast to item analysis, the present methods assume that personality types may differ in both degree and kind, that interactive combinations of items may have greater predictive power than linear combinations, and that linear constraints may obscure differences ‘between groups. b. Multiple regression.--When multiple regression methods are used to predict a dependent variable representing two different groups, the objectives are quite similar to those of item analysis. In both instances, the variables which best predict the dependent variable are identified (Walker and Lev, 1953). The main advantages over item analysis are that relationships among predictors are taken into account by determining the effect of each predictor with the caffects of others partialed out and by determining weights for Predictors which maximize prediction from a linear combination of Predictor variables. That is, the method reveals the relative jJREKDrtance of different variables to the extent that linearity, additivity, normality, and absence of interaction constraints satisfactorily model the behavior in question. Although multiple regression comes a step closer to the Objective of examining the structure of a system, the constraints 55 are still quite severe. The underlying systems must not have configural characteristics, and groups must not differ in kind. c. Discriminant analysis.--The present methods for differ- entiating between personality "types" also bear some resemblance to discriminant analysis (Cooley and Lohnes, 1962). In discriminant analysis, groups are treated as independent categories of a nominal level variable rather than as points along a continuum. The method facilitates group comparisons by determining weights for predictor variables such that each group's mean score is maximally different from every other group's mean score. The objective is accomplished by forming a pooled within-groups cross-products matrix of deviations of scores from group means and a between-groups cross-products matrix of deviations of group means from the total sample mean. Discriminant functions are computed as vectors associated with the latent roots of an equation for maximizing the ratio of between-group to within-group sums of squares. Although the method makes it possible to differentiate among Criterion categories (groups), both linearity and normality constraints aare imposed by the factor analytic methods employed, and the dis- <1rindnant function requires that group differences be satisfactorily modeled by a linear combination of several continuous variables. Again, the assumptions of linearity, normality, addivitivy, and absence of interaction restrict the method's usefulness with social Science data . 56 d. Interaction detection.~-Sonquist and Morgan (1964) have developed a method for identifying interactive combinations of variables that predict a criterion variable. The method may be used for differentiating between groups provided that the groups can be represented by a dichotomous dependent variable or as the values of an interval scale. The method employs a nonsymmetric branching process to sequentially subdivide a sample into subgroups which maximize prediction of the values of the dependent variable. Analysis of variance techniques, rather than regression techniques, are used to identify predictors which provide the largest reduction in unexplained variance when used to subdivide a sample. Predictors may meet the conditions of either nominal or ordinal level measurement, and the final groups will consist of persons characterized by interactive combinations of values for the variables used in predicting subgroup membership. While this method of detecting interaction effects avoids the assumptions of linearity, normality, and absence of interaction found -in many other multivariate methods, the method has several drawbacks. Iiirst, since interactive combinations of predictors are built up 8eQuentially, the proportion of error in the residual variance may 11N3rease with each split. Second, decisions to split a sample will be dominated by chance factors when different variables account for Similar proportions of variance. Third, since a sample is split into a number of subsamples, the original sample must be quite large. 1‘5 it is not large, only a few splits can be made. As a result, the a u. ”I D" n. i. In. a» I 57 method is inappropriate for analyzing data from small groups or individuals. e. Exact-pattern methods.--Exact-pattern methods for differ- entiating between groups have been developed by McQuitty and others. Unfortunately, McQuitty's methods have not constituted an improvement over linear methods (McQuitty, 1957), and other methods designed to handle configurations or syndromes have been shown to capitalize on chance occurrences of response patterns (Clark, 1968). In contrast to exact-pattern methods, the present methods differentiate groups in terms of "imperfect” patterns. f. Criterion pattern-analysis.--A method which is related to the strategy of the present research was developed by Clark (1968). The method is called criterion pattern-analysis, and it bears a strong resemblance to discriminant analysis. The method involves searching for response configurations which characterize one group but not others. The objective of the technique is to search for the laugest response configurations (or patterns) that differentiate aunong the groups. The acceptability of patterns is decided upon in terms of the frequency with which a pattern is more characteristic of one group than of others. The advantages of criterion pattern-analysis are that: (l) the linear constraints of item analysis, multiple regression, dis- <=ziilninant analysis, and factor analysis are avoided, (2) interactive <=cnmm>inations of variables may be found to predict group membership, sinus, (3) multiple or overlapping patterns of variables may differ- entiate among the various grOUPS- 58 The disadvantages of criterion pattern-analysis are that: (l) the method tends to identify a large number of rather small patterns, (2) only discrete data with few categories may be used, and (3) the patterns may be difficult to replicate since they are absolute rather than probabilistic. In contrast to criterion pattern-analysis, the present methods for differentiating among types: (1) may be applied with discrete or continuous data, (2) do not require absolute configurations, and (3) make it possible to examine the structure of relationships among patterns of meanings and thereby describe behavior in terms of system characteristics instead of simple response profiles. 9. Comparative dimensional analysis and comparative typological analysis.--Two methods, whose objectives are quite similar to the objectives of methods developed for the present research, have been developed by Tryon (Tryon and Bailey, 1970). These methods are called comparative dimensional analysis and comparative typological .analysis. Although the methods are not entirely appropriate to the IRroblems studied here, the basic approach is in keeping with the Irresent methods for differentiating between groups. Comparative dimensional analysis is of two sorts--subjective and objective. In the subjective form, inter-cluster correlations are calculated from the raw data defining the clusters of each group, and groups are then compared subjectively in terms of their patterns 0f intercorrelations. In the objective form, both within-group and between-group similarity indices are calculated from the factor I. I II PI v ‘A h i . 59 patterns that define clusters. The resulting similarity matrix is then submitted to key-cluster analysis. In comparative typological analysis, the objective is to discover the degree to which object—types within different groups have the same structure. This is accomplished in two ways: (1) by examining the frequency with which the pattern of standard scores defining different subtypes of a group happen to occur, and (2) by direct comparison of z-score profiles for subtypes generated for each group. While Tryon's methods are certainly in the spirit of methods used in the present research, they use different sorts of data from that gathered for the present research. Specifically, in Tryon's methods, scores rather than response vectors are the basic data, the methods are in some ways concerned with more detailed analysis of within-group organization, the results of separate analyses are described primarily in dimensional terms, and a number of parametric assumptions are required. Nevertheless, these methods are concerned tnith "types" and "subtypes," and the potentially configural character <>f groups characteristics is clearly recognized. 4. Summary of other differentiation methods.--While differ- entiation methods have been described roughly in order of their ability to handle complex systems, the full implications of this discussion can only be seen by summarizing the methods. Table l Presents some of the major constraints imposed by these various I“ethods as well as major capabilities of these methods. The methods 60 TABLE 1 Constraints and Capabilities of Several Methods for Differentiating Between Groups Constraints Capabilities to (D .H 'H m E H O: o u o 5 u s u o m o c H >. 'o H O . .4 m H AnalySis Methods 5 g P u m 5 :3 -a o m m o .x m o u A m m -H m E s o s u a m -a In 0 "O w-l U) 4-3 :4 «4 a) H «4 r: .12 0'1 O u .4 c u o p O u -o o m -H -H m a c A c z u u 3.4 H m o o m o u.» m >. m -H o a 0 >1 'H o c .4 u u .u-o c >1 >1 u o o m.4 m p m m-H a u u -a .4 a o o .c > ca > o r! -:-l :> 0 CG (D H -H U 4-) H «4 or! n .4 -H o a u .3.» o c and is m c u c 9 s o. O a o s -H o E -a o o m -a O s c o u H c seweessssea.s g g a 'fi H:z > m 0 Dim a o m Item Analysis Y Y N.A Y Y N.A. N Y N Y Multiple Regression Y Y Y Y Y N N Y N Y Disciminant Analysis Y Y Y Y Y N N Y N Y Comparative Dimen- Sional AnalySIs Y Y Y Y N Y Y Y Y Comparative Typo- logical Analysis Interaction Detection N N Y N N Y Y Y N Y Exact Pattern Meth Criterion Pattern N N N N N Y Y N Y Y Analysis Content Differ- entiations N N N N N Y Y Y Y Y Res.ponse-Style Differentiations N N N N N N.A. Y Y Y Y 531113actural Differentiations N N N N N Y Y Y Y Y \ N.A. = not appropriate. 61 are arranged in order of the number of constraints that they impose. The methods developed for the present research are also included. Furthermore, it appears that the methods developed for the present research impose very few constraints yet have a very broad range of capabilities. Essentially, these methods seek to avoid constraints which distort relationships among variables or oversimplify the structure of a matrix and thereby yield unrealistic results. From the table it may easily be seen that the constraints and capabilities of various methods differ rather sharply. It should be noted, however, that the listed capabilities of different methods tend to increase as constraints decrease. The most severely constrained methods have very limited capabilities and vice versa. Since the data collection techniques developed for the present research appear useful for studying a wide variety of behavior, the apparent strength of the present analytic methods suggests that this combination of data collection and analysis methods can be extremely .powerful tools for social scientists. CHAPTER I I I RESEARCH METHODS While Chapter II describes the general strategy of this research, the present chapter describes the more technical aspects of how the strategy was implemented. Included in this chapter are discussions of subjects selected for the research, personality inventories and object perception instruments employed, the research design, randomization and other controls, data collection procedures, data preparation procedures, and data analysis methods. Instruments and analysis methods developed especially for this research are treated in detail. In Chapter IV, prior to discussing the results for each method of differentiating between contrasting personality subgroups, the steps of the analyses performed will be briefly reviewed. For this reason, the reader who is interested in no more than an overview of the methodology may skip sections dealing with product selection, controls, preparing data for analysis, and details concerning the formation and analysis of attribute interassociation matrices. Subjects Undergraduates enrolled in several courses at Michigan State University during the spring of 1969 participated in this study. 62 63 The project was publicized as an Esthetic Preference Study, and students were asked to volunteer approximately three hours of their time. Participation was restricted to females, and usable data were obtained from 128 of the 129 §§ completing all parts of the study. About 72 per cent of those who initially volunteered for the study were enrolled in Introduction Psychology classes offering some course credits for participation in research projects. Other students, who participated on a purely voluntary basis, were enrolled in sopho- more level Psychology courses, a freshman level Home Economics course, and a freshman level Communications course. §s were scheduled for one session a week for three weeks. Of the 161 students who initially signed up for the study, 88 per cent (141) attended first-week sessions. Ninety-five per cent of the first-week gs attended second-week sessions, and 96 per cent of these returned for the third part. Calculated in relation to the number of first-week participants, the loss of SS from the first to the third week was just 9 per cent. Considerable effort was devoted to reminding SS of the times for which they were scheduled and to rescheduling SS who failed to come to a session. In evaluating the participation figures, however, it should be noted that several gs (about six) were dropped from the study because of rescheduling complications. Since different procedures were followed on different days and at different sessions during the same day, it was sometimes impossible to find a satisfactory rescheduling time. 64 With minor exceptions, only the data obtained from the 128 ES completing all parts of the study were used in the major analyses. Furthermore, only one of the SS who attended all three sessions had to be dropped from the study (a major mistake in the second-week procedures which was not noticed by the subject had made much of her data unusable). For gs who participated all three weeks, nearly all data were complete. The potential problem of incomplete data was controlled by checking whether gs had filled out everything and by keeping group tasks carefully coordinated. Materials All of the instructions, instruments, and apparatus used in the research are described in this section. Materials developed 'specifically for the study are discussed in detail and are included or pictured in Appendices A through G. Materials which are com- mercially available are described more briefly. Preliminaryrlnstructions Three sorts of preliminary instructions were used. The three included a description of participation requirements, an introduction to the purposes of the study, and some general instructions which were appropriate to all parts of the study. These materials are found in Appendix A. At the first-week sessions, all three preliminary instruction forms were used. At sessions during the second and third weeks, only the description of participation requirements and the general instructions were used. 65 Personality Inventories The personality inventories used in this research were selected to represent several areas of cognitive and social functioning: the perception of role persons, the nature of belief- disbelief systems, orientations to social situations, and the content of value systems. The first two personality inventories are concerned mainly with the structure of cognitive systems. The latter two focus on the specific content of one's cognitions. The inventories not available through commercial sources are included in Appendix B. Cogpitive Copplexity Index (Bieri).--The index of cognitive complexity used in this research was basically a form of the Role Construct Repertory Test (Kelly, 1955) with provided constructs. The original form of this scale was developed by Tripodi and Bieri (1964) and is described in detail by Bieri (1966). Bieri (1969) subsequently changed the format of the scale to simplify the task and control for order effects. The form used here was a minor modifi— cation of Bieri's revision designed to minimize other possible sources of ambiguity in the instructions and the arrangement of scales. To score the scale, responses were put into a logical rather than a random order, and a "role persons" by "construct dimensions" matrix was formed. A §fs score was calculated by comparing all pairs of ratings made by each "role person" and then counting the number of exact matches. This count was next divided by the number of com- parisons for which neither response was missing (a modification of Bieri's method designed to handle missing data). Scores were inversely related to cognitive complexity. '3’Y‘ u-~ .I/ .- ..._ ., . a. _ I v. a ‘- o'- A II> i: . 66 A computer program was written to carry out the response reordering and score calculations. Dogmatism Scale (Rokeach).--The dogmatism scale used in this research was a 20-item short—form developed by Trodahl and Powell (1965). With one exception, the items were not modified from their wording in Rokeach's 40-item version of the scale (Rokeach, 1960). The exception was an item appearing to have a strong sex bias (item 25 in the Rokeach scale). The wording of the item was changed from " . . . become a great man, like . . ." to " . . . become a great person, like . . ." Since all items on this scale were stated so that agreement reflected dogmatic thinking, a §fs score was the sum of the codes for positions along a 6-position Likert-type rating scale. The codes used ranged from 2 ("disagree very much") to 7 ("agree very much"). A neutral score for the 20-item scale was 20 x 4.5 or 90. Adjustments for missing data were unnecessary since these data were complete, and scores were calculated by a computer program written for the purpose. Orientation Inventory (Bass).--The Orientation Inventory (Ori) was developed by Bass (1967) to measure the impact of personality factors upon one's orientation to a variety of social situations. In particular, the instrument is based on the idea that there are three basic orientations to group situations-—self—orientation, interaction-orientation, and task-orientation. That is, in any social situation, different individuals will be motivated by 67 different concerns and will attend to different content. These perceptual differences should, in turn, affect their social behavior. The instrument has its origins in the thinking of McClelland and Schacter, and in Bass's own efforts to develop an instrument which would be useful for predicting job-satisfaction, style of leadership, and job performance. The instrument was scored as described in the manual for the scale (Bass, 1962). The scores for the three subscales (self, interaction, and task orientation) were then converted to T-scores using norms published in the manual. There were norms for female college students, primarily freshmen and sophomores, from various parts of the country. Value Survey (Rokeach).--Form D of Rokeach's Value Survey (Rokeach, 1967) was used in this research. This version contains Terminal and Instrumental value subscales with 18 values each. Each value is printed on a gummed label and respondents rank order the values by moving the labels to the desired rank positions. Respondents are instructed to rank the values of each subscale from the one which is most important to the one which is least important to them. As with the Orientation Inventory, the Value Survey was used because it focuses on content features of cognitive systems-~content presumed to be central to one's belief system. While Rokeach (1968) has argued that matters of taste fall at the level of inconsequential beliefs, the present research seeks evidence of connections between values and the meaning-systems underlying consumer-product perceptions and hence can serve to check on Rokeach's assertion. 'aqoe unlov- 4". q‘..‘ a.‘ \ 'u‘. '. h s.‘ 68 The Value Survey was not scored in a conventional manner. Instead, the configurations of ranks for different values were used as the data for other analyses. Consumer-Products Selected for the Research As mentioned in the preceding chapter, sterling silver table- ware was the consumer-product selected for use in this research. Forks were borrowed from a single manufacturer. They included designs marketed by the particular company and its subsidiaries during the past 50 or 60 years. Such a broad range of years was chosen in order to represent the wide variety of motifs found on the market today, not simply the most popular styles of the present. Rationale for Selecting Sterling Silver Tableware.—-To test the present model of complex perceptual-cognitive processes, it was important to select a suitable class of products. High quality home furnishings appear to have characteristics which lead to a high probability of finding relationships between personality character- istics and the meaning-systems underlying product perceptions. Sterling silver tableware was selected as a representative of this class. The characteristics of high quality home furnishing which appear to increase the probability of relationships with personality are listed below: 1. These products are sufficiently expensive that purchase decisions are likely to be carefully considered ("financial risk" is involved). 69 2. Quality furnishings are commonly used in relatively formal settings established for social interactions ("social risk" is involved). 3. Alternative products of the same type and quality differ mainly in physical appearance (styling differences are important). 4. Style characteristics of such products are commonly imbued with symbolic meaning by consumers, advertisers and designers. As a result, the symbolic associations with products may come to play an important part in development and maintenance of the consumer's self-image and in the presentation of a consumer to others. 5. There may be large individual differences in the perceptual- cognitive categories employed, depending upon differences in psychological and sociological histories. That is, for different sorts of persons there may be different sorts of relationships between personality and object perceptions. Sterling silverware seemed a good choice for several additional reasons : 1. Silverware is sufficiently small that a rather large number of pieces may be conveniently used in a study. 2. There are a wide variety of styles having pieces of approxi- mately the same size. 7O 3. There are a wide variety of styles which do not utilize color. In other words, the absence of color controls for a potentially complicating factor. Forks Selected for Object Evaluations.-—Sixty forks were selected for a task in which gs indicated how well they liked each design. The writer selected sixty forks from a field of about 200 by first sorting the forks according to their apparent elaborateness (see Appendix C for a list of the forks selected). An effort was made to keep in mind such criteria as object outline, bulk of decoration, and density of decoration, but the major criterion was simply the impression of overall degree of elaborateness. The sorted objects were then divided into six levels of elaborateness. Again, this categorizing was done on the basis of overall impression. Finally, 10 forks were selected from the variety of designs within each level. Forks Selected for Object Descriptions.--Eighteen of the sixty forks used in the evaluations task were selected for a task in which SS indicated what attributes they associated with each fork. The intent of this selection was to identify a wide variety of designs for which inter-person and intra-person evaluations of different designs could be expected to range from quite favorable to quite unfavorable (see Appendix C for a list of the forks selected). The eighteen forks were selected on the basis of evaluative ratings obtained during the first week of the study. Frequency distribution analyses of these ratings were performed, and forks were 71 considered if their evaluative ratings covered a broad range of favorability and the distribution was fairly flat. The field of designs was narrowed to 30 on the basis of distribution statistics, and 18 forks which seemed to the writer to be reasonably representative of the full range of designs were selected for the Object Descriptions Task. Object Evaluations Task The sixty designs selected for this task were displayed in a random sequence on two large tables (see Appendix D). The sequence Was determined with a random number generator on a computer, and the same sequence was employed throughout the study. The forks were displayed on six 11 by 27 inch sheets of flat-finish black cardboard. There were 10 forks per sheet of cardboard, and they were numbered sequentially. The forks were placed on the background cardboards With their tines toward the _S_ and with the identifying numbers immediately ahead of the tines. Instructions.-—§s were required to express their overall degree of like or dislike for each of the sixty forks using a ten- point scale that ranged from "dislike very much" to "like very much" ( See Appendix D) . Rating procedures.--§_s marked their evaluative responses on spe<=ially printed machine-readable answer sheets. The numbers 1 u‘rough 60 were printed adjacent to rows of 10 mark-sense positions. clipped to the top of each answer sheet was a printed template l‘lentifying the answer sheet columns corresponding to positions of In ‘1‘ ‘ Htt 72 the rating-scale. On each S's answer sheet a starting number was circled. gs were to begin their ratings with the fork of the same number and proceed in sequential order. gs were assigned to one of six starting positions (1, ll, . . ., 51). After reaching object 60, SS rated objects 1 to their starting number. Clipboards were provided to make the rating task more convenient. Object Descriptions Task The Object Descriptions Task was based on an operationalization 0f the fundamental cognitive unit of the present model (i.e., the Concept of an elementary cognitive subsystem). The technique was Originally inspired by research of McQuitty, Abeles, and Clark (1970) and an early form of the technique was used in research concerned with a few of the problems to which the present research was directed (McQuitty, Price, and Clark, 1967; Price, 1968). . The following discussion of different aspects of the Object Descriptions Task describes the task required of a single _S_. Minor Variations necessitated by group-administration procedures are described in a later section. The instructions, response forms, apparatus, task layout, and ron arrangement for the Object Descriptions Task may be found in Appendix E . Overview of the Object Descriptions Task.--For each of the 18 ft>3ii‘ks, a §_'s task was the same. A list of 36 attributes was presented to the S. For each attribute, she was asked to indicate: (1) the e“tent to which it seemed to apply to the object, and (2) the degree 73 to which the perceived association was liked or disliked. The § was then asked to think of an attribute or idea which contrasted with the one listed and then react to this content in terms of applicability and valence. gs rated the applicability of a characteristic along a five- POint scale from "doesn't apply" to "extremely applicable." Degree Of like or dislike for an observed characteristic of an object was .rated along a five-point scale from "Dislike Considerably" to "Like Considerably." There were two degrees of positive and two degrees of negative affect plus a neutral position. gs were instructed to respond on the basis of first impressions, axui they were free to respond to as many or to as few characteristics as seemed to apply. Booklets of Attribute Lists.--§s used booklets containing 18 liAStS of 36 attributes each. The same 36 attributes appeared on each Page but in 18 different random orderings. For each new object, it was necessary to turn to the next list of attributes. The randomi zed liésts of attributes were printed as computer output, cut to 8 1/2 x 11 .inch size, and stapled together to form a booklet. 1. Identifying appropriate attributes.--The procedures used hi :ldentifying appropriate attributes for describing sterling silver takileware are discussed in Appendix E. In general, the attributes can‘e from a variety of sources including: (1) research into the meaning structure of paintings (Osgood, g£_2l3, 1957, pp. 68-70), (2) a.thesaurus search for likely adjectives, (3) vocabulary used in 74 books on the design of silverware and other home furnishings, (4) attributes suggested by designers and marketing researchers, (5) the writer's pilot research on words associated with silverware, and (6) a variety of other research concerned with the meaning-systems underlying the perception of physical or social objects. Both denotative and connotative attributes were included among those selected, but a few constraints were imposed on the selection of attributes. In the first place, attributes which were :seldom associated with silverware were screened out (some of the anniter's own research findings were used to quide this part of the selection). Second, an effort was made to select attributes which tiiki not seem highly evaluative. This was done in order to minimize tine confounding of description and evaluation. And finally, the hnriter attempted to select attributes which could be interpreted as desirable traits by some §_s. Object Sequences.--The 18 objects selected for this task were Prwesented to gs in three different randomized sequences. The sequence was changed every second session, and each sequence was used about the Saune amount during the study. gpparatus and Response Forms.--Several materials were prepared “3 .standardize the Object Descriptions Task, minimize procedural errors, and facilitate rapid response. 1. Object display board.--Each §_was provided with a 6 3/4 by ll.inch piece of flat-finish black cardboard. This cardboard 9'; "a u.‘ ‘v‘ u 0“ “. 1‘71 1... )1 75 provided a standard background on which gs could place a fork while responding to it . 2. Response apparatus.--The main piece of apparatus con- structed for the Object Descriptions Task was designed to facilitate rapid response rates and reduce procedural errors. The apparatus was Constructed of cardboard and measured 14 inches by 14 inches. Machine-readable answer sheets were placed in the apparatus and were held steady by it. A summary of the questions to be answered was Printed across the top of the apparatus. The various response categories were aligned with appropriate columns of the answer sheet. The side of the apparatus was numbered from 1 to 36. These numbers were aligned with rows of an answer sheet and represented the 36 attributes on a page of the booklet of such lists. A sliding row- guide was also provided as a means of reducing errors in marking responses. 3. Answer sheets.--There was one answer sheet for each of the 18 forks. The answer sheets were IBM machine-readable multiple- Ch<>ice forms which had been printed without number identifications for any of the mark-sense positions. To make sure that _S_s were using the correct answer sheet for each object, a number from 1 to 18 was written in the center of each answer sheet. Materials were arranged so that answer sheet number, attribute list number, and ordinal position of an object in the presentation sequence all corresponded. v.‘.' n... ...- CU‘. 4 ~9‘.‘ ‘Uid - a. .0- v- «No In. ‘ II . 'Vcb 3'9. 5.... '7 ‘ i. ‘« “ 76 A Comparison of the Object Descriptions Task and the Semantic Differential.--When the present research was in its initial planning stages, it was hoped that a Semantic Differential could be used for studying cognitive systems. However, after closely examining the Semantic Differential, the writer concluded that it had several drawbacks which could be avoided with a somewhat different instrument. The Object Description Task was developed as a first effort toward snalving a variety of the problems inherent in the Semantic Differ- ential . The sorts of problems solved by the Object Description Task axre explained by comparing the two approaches on several points. 1. The respondent task.--While both techniques ask gs to use ii rather large number of verbal symbols in expressing reactions to .Ffluysical, social, or abstract objects, the rating tasks are rather di ff erent . a. Semantic Differential.--Respondents are asked to express each reaction to an object by marking a position along a bipolar rafting scale. The poles are constructed to be "semantic opposites" and the positions of the scale are mutually exclusive. In using a Semantic Differential, the researcher must assume tlhain (1) all respondents view the polar adjectives as opposites, and (:3) that polar adjectives may be realistically treated as the end- Points of a single dimension with mutually exclusive positions. 'N‘EIt is, the technique cannot measure certain varieties of individual differences in cognitive habits. 77 b. Object Descriptions Task.-—Respondents are asked to express each reaction to an object by rating the applicability of an attribute and the degree of like or dislike for a perceived association between attribute and object. Respondents are also asked to think of a word or idea which contrasts with the attribute given and make the same sorts of ratings. These latter ratings for Contrasting ideas are made under the assumptions that: (1) there Imay be large individual differences in what is thought to contrast Vflith a given attribute, and (2) contrasting ideas may be thought to apply even when the attribute paired with it applies (they may not be nuitually exclusive). Since §s react to each attribute separately and are free to define contrasting ideas in accordance with the characteristics of tflmeir own perceptual systems, both of the assumptions made for the Semantic Differential are avoided. Furthermore, if _S_s view given airtributes as opposities or view contrasting ideas as opposites to Pinbvided attributes, the psychological distance between the attributes Wiflll be maximized and they will have responded as if ratings had been made on bipolar dimensions. This situation holds true regardless of Whether other _S_s respond in a similar fashion. In sum, the Object Descriptions Task appears to be more flexible and more sensitive to complex cognitive phenomena and individual differences than is the Semantic Differential. At the sauflue time, additional information is obtained in the form of like- dLESlike responses for each meaning-object association. 78 2. Construction and interpretation.--The two techniques are also quite different with regard to problems of constructing an instrument and interpreting results. a. Semantic Differential.--The opposites used for bipolar response scales are defined empirically by obtaining large numbers of judgments. If judges are not in sufficient agreement about the Opmnsites for given attributes, these attributes are eliminated from Ifurther consideration. To the extent that requiring low variability 5J1 the judgment of opposites tends to eliminate a non~random subsets <3f relevant content, the resulting instrument will not be repre— sentative of the meaning-systems it purports to measure. The situation is complicated further by the fact that érttributes perceived as semantic opposites in one context are not necessarily perceived that way in a different context. New instruments may need to be developed for different contexts. An additional problem occurs because the Semantic Differ— errtial was designed to measure mainly the connotative meanings associated with objects. On the one hand, attributes most likely to YiJeld high agreement in judgment of opposites are denotative rather truan connotative. On the other hand, it is difficult for researchers tc> interpret responses unless the scales are highly connotative and t1"lese connotations are uniformly understood by most people. In other words, the researcher is on the horns of a dilemma. For example, if at; (object is described as "rough," it is impossible to know whether rOughness is regarded as pleasant or unpleasant in this instance. If an object is described as "pleasant," it is impossible to know why 79 it is regarded as pleasant. In other words, the construction of the Semantic Differential conflicts with its objective of identifying relationships between sign, object, and person. To the extent that content failing to meet the conditions described above is eliminated, the instrument becomes less repre- sentative of the relevant universe of content. b. Object Description Task.--A list of attributes for use in anLObject Descriptions Task may be assembled rather easily. Since gs