WWW/WWW WW \ 2%: 25¢ per day per itan RETURNING LIBRARY MATERIALS: Place in book return tor charge from circulation records JANO G lsfm r: 3 U815‘:’2 “Mama? THE USE OF OPERANT MULTIVARIATE TYPING WITH THE FISHBEIN ATTITUDE MODEL IN MASS MEDIA ATTITUDE RESEARCH By John Collins Sutherland A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Ph.D. Program in the Mass Media College of Communication Arts and Sciences 1980 ABSTRACT THE USE OF OPERANT MULTIVARIATE TYPING WITH THE FISHBEIN ATTITUDE MODEL IN MASS MEDIA ATTITUDE RESEARCH By John Collins Sutherland Attitude studies are common in mass media research. Attitudes are studied in measuring media effects, in public policy research, in planning media content, and in advertising decision making. While there is disagreement over the operationalization of the concept, most agree that attitude is a learned presdisposition to respond in a con- sistently favorable or unfavorable manner to a given object. However, because there is little agreement on operationalization, many mass media attitude studies are not directly comparable or cumulative. The Fishbein attitude model offers a solution to these problems. Fishbein's model hypothesizes that an individual's attitude toward an object is a function of his beliefs about the object and the evaluative aspects of those beliefs: n A = z b.e. 0 i=1 1 1 where Ao_= attitude toward object "0" U” II the strength of belief "i" about the attitude object "0," that is, the Ii :- 1;; 7m .v‘;4 probability or improbability that "o“ is related to some other object "xi" ei = the evaluative aspect of "bi’" that is, the evaluation of "xi"--its goodness or badness n = number of beliefs While the model has performed well in empirical tests, there is a major problem in using the model in survey research. Most researchers agree that if the sample is not homogenous, the results of the model would be misleading. A plausible solution to this problem is grouping persons on the basis of their like and dislike scores before analyzing results. This requires multivariate typing if all like-dislike scores are to be used. If multivariate typing results in more homogenous groups, the model should perform better and there should be less variance about the responses of persons grouped together. To test these ideas, the Fishbein procedure for developing and administering a questionnaire was followed. Students at Michigan State University were asked to complete a questionnaire designed to measure their attitude toward the student newspaper. The students' "e" scores were Q-factor analyzed and three different groups were discovered. The performance of the model, using the Pearson product-moment correlation coefficient as a measure of performance, for the three attitude segments was compared to the performace of the model for a random sample of students, and three groups formed on the basis of reading behavior, heavy, medium and light readers. There was no significant difference in the performance of the model among the seven groups. An analysis of variance found significant difference in the variance about the "e" scores, but not about the "b" scores. The differences in like-dislike scores were attributed to the multivariate typing. It was concluded that multivariate typing using like-dislike scores tends to produce more homogenous groups than other methods. Another problem with the Fishbein model is the summation of scores to achieve an overall measure of affect. While such a measure is useful, summation precludes the study of attitude content. A method using a person's factor loading was developed to overcome this problem. The basis for this procedure is that a person who is more like a factor should have his scores weighted more than one who is not as heavily loaded on the factor. Raw scores are weighted to produce a stereotypical response pattern which retains the individual- ity of "b" and "e" items and permits the study of attitude content. The results of this study provide direction for using multi- variage typing in mass media attitude research. This typing should result in more homogenous groups and results less contaminated by the variability among persons under study. © Cepyright by JOHN COLLINS SUTHERLAND l980 To Rita and Joe ACKNOWLEDGMENTS The author wishes to express his thanks to all those who helped along this academic journey, especially the members of his committee: Dr. Tom Muth, Dr. Gordon Miracle, and Dr. Ray Cullen. I would like to express special thanks to Dr. William Crano whose unbridled guidance helped me develop an intellectual independence. I also express my gratitude to Dr. Charles Mauldin whose enthusiasm for teaching and vigor for learning taught me that professionals could and must be good humans. The author wishes to express his gratitude to his parents whose love and teaching provided the foundation for this vocation. My love and thanks to Rita for being such a good friend while sharing this experience with me. iv TABLE OF CONTENTS LIST OF TABLES .......................... ATTITUDE THEORY AND MEASUREMENT .................. What Is an Attitude? .......... . ..... . ..... The Components of an Attitude ............... The Functional Nature of Attitudes ............ Attitude Measurement ATTITUDES AND MASS MEDIA RESEARCH ................. Attitudes Reflect Mass Media Effects ........... Attitudes Can Be Used for Public Policy Decisions ..... Attitudes Can Be Used to Make Media Content Decisions . . . Attitude Research Reveals Public Opinion 9 ; ....... Attitudes Help in Making Marketingsand Advertising Decisions ........................ Attitude and Mass Media Research: Conclusions ....... An Approach to These Research Needs ............ FISHBEIN ATTITUDE MODEL ...................... Fishbein's Conceptualization of Attitudes ......... Salient Beliefs ...................... The Identification of Salient Beliefs ........... Differences Between Salience and Importance ........ Summary of Fishbein's Model ................ Evidence in Support of Fishbein's Model .......... ISSUES IN THE USE OF THE FISHBEIN MODEL ............ . . Beliefs-Only Model .................... Belief Measurement .................... Independent and Dependent Variables ............ Normalization . . . . . . . . . . . . . . . . ...... Halo Effects ....................... The Use of Multiple Regression .............. Independence of Items ................... Level of Measurement ................... Interaction Rules .............. . ...... STATEMENT OF THE PROBLEM ..................... HYPOTHESES ............................ Operant Multivariate Typing ................ Model Performance ..................... Hypothesis 1 .................... Hypothesis 2 .................... Hypothesis 3 .................... METHODOLOGY ............................ RESULTS Sample .......................... Attitude Object ...................... Questionnaire ....................... Data Collection ...................... Level of Measurement ................... Q-factor Analysis ..................... Q-factor Solution ..................... Reader Segments and the Random Sample ........... Hypothesis 1: Significance of Pearson's Product- Moment Correlation Coefficients ............. Hypothesis 2: Tests for Differences between Correlation Coefficients ................ Hypothesis 3: Differences between Average Variances , , , , Scheffé Tests ....................... Tukey Tests ........................ Limitations ........................ Hypothesis 1 ....................... Hypothesis 2 ....................... Hypothesis 3 ....................... Hypotheses 3A--38--3C ................... Hypothesis BD ....................... Discussion ........................ FACTOR ARRAYS ........................... Factor Interpretations .................. Consensus Items ...................... Factor I: Sports Enthusiasts ............... Factor II: Hard News Readers ............... Factor III: Ludenic Readers ................ Discussion ........................ IMPLICATIONS AND CONCLUSION .................... Theoretical Implications ................. Methodological Implications ................ Practical Implications .................. Conclusion ........................ 94 97 98 103 106 106 121 127 148 APPENDIX A: FOCUS INTERVIEW GUIDE ................. 154 APPENDIX B: NEWSPAPER STUDY .................... 156 APPENDIX C: VARIMAX ROTATED FACTOR MATRIX AFTER ROTATION WITH KAISER NORMALIZATION ............... 166 BIBLIOGRAPHY ........................... 169 vii Table l. Table Table Table 01 b (A) N o o o a Table Table 6. Table 7. Table 8. Table 9. Table 10. Tablell. TablelZ. Tablel3. Tablel4. Tablel5. Table16. Tablel7. Tablel8. TablelQ. LIST OF TABLES Belief Statements Used in Study of Attitudes Toward Presidential Candidates in the 1964 Election ........................ Hypothetical Scores for Persons A and B ......... Groups with Matched Subjects .............. Repeated Measures Design ................ ANOVA Table for the Two-Way Mixed Effects ANOVA with n=l ..................... Contrasts and Coefficients ............... Pearson Product-Moment Correlations ........... Comparison of Correlation Coefficients ......... Average Variances for e Items .............. Average Variances for b Items .............. Repeated Measures ANOVA for e Items ........... Repeated Measures ANOVA for b Items ........... Scheffé Contrasts—-b Items ............... Scheffé C o.ntrasts--e Items ............... Tukey Contrasts--b Items ................ Tukey Contrasts--e Items ................ Significant Differences between Attitude Segments and Reader Segments: e Items .............. Respondent Weighting Procedure ............. Consensus and Discriminating Items . . . . . ...... viii 48 108 110 110 111 111 113 Table Table Table Table Table Table Table 20. 21. 22. 23. 24. 25. 26. Demographic and Behavioral Profiles .......... 128 Consensus e Items ................... 129 Consensus b Items ................... 130 Consensus e and b Items ................ l33 Factor I Discriminating Items ............. 139 Factor II Discriminating Items ............. 142 Factor III Discriminating Items ............ 144 ix ATTITUDE THEORY AND MEASUREMENT In beginning an article on attitude theory and research, one is likely to sit for long moments pondering the "lead" to his story. Those long moments can lead to longer moments because the Concept of attitude is so old and so widely used. As Gordon Allport (1935) pointed out some forty years ago, ". . . attitude is probably the most distinctive and indispensable concept in contemporary American social psychology. No other term appears more frequently in experimental and theoretical literature" (p. 798). At times, attitude has even been considered synonymous with social psychology. Today, attitudes can be found in other theory and research areas including marketing, adver- tising, journalism, broadcasting, and public policy. As researchers have become more sophisticated, the study of attitudes has become more "scientific." But, Unlike many other concepts which seem to become the exclusive property of the research community, attitudes continue to be of interest to the common man as well as the scientist. As humans, we continue to enjoy knowing about ourselves. "In our daily news commentary, in our literature, and even in our music, we continually hold ourselves up to the mirror, trying to define how we think, how we live and who we are" (Baroody, 1978, p. 2). What is an Attitude? There are about as many definitions of attitudes as there are writers and theorists on the subject. Some of these definitions are: 1 2 Attitudes refer to the stands the individual upholds and cherishes about objects, issues, persons, groups, or insti- tutions. (Sherif, Sherif, and Nebergall, 1965, p. 4) An attitude is a mental and neural state of readiness, organized through experience, exerting a directive or dynamic influence upon the individual's response to all objects and situations with which it is related. (Allport, 1935, p. 881) An attitude is a relatively enduring organization of be- liefs around an object or situation predisposing one to respond in some preferential manner. (Rokeach, 1968, p. 112) While considerable dialogue has continued for several decades on the precise definition of attitude (McGuire, 1969), there are several characteristics of attitudes that appear to be generally accepted. Attitudes always suggest a relationship between a person and objects or situations. These attitude objects are not necessarily physical objects but could be another person or group of people or subjective objects, such as ideas, issues, or concepts. An object of an attitude may be anything that exists for an individual. Rokeach's definition points out that attitudes are learned; they are not innate. Attitudes are not brief, transient states; they are relatively enduring predispositions. Attitudes do change but usually not very rapidly. Attitudes are inferred constructs. We cannot "see" attitudes, but rather, researchers infer attitudes from what people say, their stated evaluations, preferences, intentions, or behavior. Attitude is admittedly a subjective and personal matter. Finally, attitudes are not neutral. They dispose us to like or dislike, to seek or forego, to evaluate as desirable or undesirable. 3 As Thurstone and Chave (1929) wrote, attitude denotes " . . . the sum total of a man's inclinations and feelings, prejudice or bias, pre- conceived notions, ideas, fears, threats, and convictions about any specified topic... ." (p. 6). Thus attitudes tell us what people want and what people expect (Markin, 1974). The Components of an Attitude The traditional view is that there are three underlying attitude components: (1) cognitive, (2) affective, and (3) conative. Attention to these three different aspects of attitude goes back at least to McDougall (1908) and persists in other work (e.g., Brown, 1965; Katz and Stotland, 1959; Newcomb, Turner, and Converse, 1964). The cognitive component refers to how the attitude object is perceived (McGuire, 1969). A typical measure of this component is a checklist of attributes (Gilbert, 1951; Katz and Braly, 1933). Katz and Stotland (1959) suggested analyzing the cognitive component as follows: (1) the number of its elements, (2) the degree of structure or hierarchy among these elements, and (3) the range of objects to which the cognitive component applies. Rokeach's (1960) concepts of centrality, permeability, and gradient of belief offer other ways of analyzing the cognitive component. Multidimensional, factor-analytic approaches are, in general, used to measure this component. The affective component refers to the person's feelings of liking or disliking about the object of the attitude (McGuire, 1969). As the evaluative component, some theorists consider it the core of attitude. Fishbein and Ajzen (1975) restrict the concept of an attitude 4 to this component. Classical scaling procedures (Guttman, 1944; Likert, 1932; Osgood, Suci, and Tannenbaum, 1957) and unidimensional scales tend to be used to measure the affective component. The conative (action, behavioral) component of attitude refers to the person's behavioral tendencies regarding the object. "The conative component could be measured less ambiguously by asking a person to report how he would behave in (or by observing how he actually behaves in) a situation involving the object? (McGuire, 1969, p. 156). It would appear that this component is the most directly measurable and hence the most useful. However, as McGuire (1969) stated, ". . . it tends to be measured, as frequently as to the cognitive and affective components, by a paper-and—pencil inventory which indicates how the person says he would behave in the presence of the object, rather than by observation of how he actually behaves" (p. 156). Attitude research has long indicated (Festinger, 1964; LaPiere, 1934; Mann, 1959) that the person's verbal report of his overall attitude has a rather low correlation with his behavior. Other researchers (Fendrich, 1967; Udel, 1965; Axelrod, 1968; and Achenbaum, 1966) maintain that there is a relationship between attitude and behavior. Krugman's (1977) position summarizes this paradox. He believes that it is absolutely certain that in some cases attitude precedes behavior and in other cases behavior precedes attitude. The Functional Nature of Attitudes There are a number of researchers (Katz, 1960; Smith, Bruner, and White, 1956; Sarnoff, 1960) who prefer to look at attitudes from 5 a functional rather than structural point of view. The functional approach asks the question, "Why do people hold the attitudes they do?" In 1960, Daniel Katz described four functions that well sum- marize the functional approach to attitudes. These functions are listed and defined below: 1. The instrumental, adjustive, or utilitarian function: This function recognizes the adaptive nature of the individual and his willingness to be shaped by group processes. Thus, a person adopts or maintains an attitude because of the favorable response he receives from his associates and friends when he manifests the attitude. The ego-defensive function: Attitudes are formed and shaped by the desire of the individual to protect himself from acknowledging deficiencies. The value-expressive function: Through this function the individual attempts and achieves self-expression in terms of those values that he most cherishes. Value-expressive attitudes generate satisfaction to the individual from the expression of opinions that reflect his beliefs and self-concept. The knowledge function or object appraisal: According to Katz, people seek a degree or predictability, consistency, and stability in their interaction with 6 the world. People seek knowledge to give meaning to what would otherwise be an unorganized and chaotic universe. These four general functions are not meant to be mutually exclusive or exhaustive. A given attitude may serve several functions. Attitude Measurement The simplest way for determining attitudes in a certain popula- tion is tabulating answers to a questionnaire. This is perhaps the most common approach ‘ used to study attitudes as they relate to the mass media. Using survey research, one can measure the range and distribution of public attitude, often called public opinion. Allport conducted such a study in the 19305. He surveyed over 20,000 college students in seventy colleges to determine their attitudes toward fighting for the U.S. in war. Thirty-nine percent of these students declared they would participate in no war whatsoever, 33 percent would take part only if the U.S. were invaded, and 28 percent were ready to fight for any cause that might lead the nation to declare war (Allport, 1935). While such results can be obtained by simply asking someone an open—ended question like, "How do you feel about going to war for the U.S.?" the analysis of over 20,000 open-ended responses would ‘ indeed be cumbersome. To overcome this difficulty, a priori attitude scales were develOped to improve efficiency of data collection and analysis. These scales are widely used and easy to apply, but they have been criticized. 7 The traditional scaling techniques include Thurstone scales, Likert scales, Guttman scales, the semantic differential scale, and Q-sort. The Thurstone scale is constructed by judges who place attitude statements along a scale of favorable to unfavorable (Thurstone and Chave, 1929). Each statement is assigned a scale value according to its position along the scale. Respondents are asked to indicate which statement(s) they agree with. The scale values for each statement agreed with are summed to yield an attitude score for each respondent. Difficulties arise with this method because of the time and effort required in developing the scale; because scale values of items are determined by the judges rather than by the respondents themselves; and, because different attitudinal response patterns may be expressed by the same score. People with essentially different attitudes could receive the same score. Likert (1932) developed the well-known and, perhaps, most used Likert scale in an attempt to overcome some of the limitations of the Thurstone technique. Likert suggested using a set of attitude statements, as Thurstone did, but recommended having respondents react to each statement by indicating their agreement-disagreement with the statement along a five-point scale ranging from strongly agree to strongly disagree. Since all of the items are either definitely favorable or definitely unfavorable (there should be no neutral statements), the favorable statements are assigned positive values and the unfavorable statements are assigned negative values. As a result, a respondent's attitude is determined by the algebraic 8 sum of his responses to all the statements. This type scaling has proven to be reliable (Tittle and Hull, 1967) and easy to construct. It permits use of statements not manifestly related to the attitude being measured (Selltiz, Wrightsman and Cook, 1976). However, the scale is only ordinal (although usually treated as interval) and like the Thurstone scale, the total score may have little clear meaning since the pattern of responses to various items may produce the same summed score. Many researchers use a Likert-type scale but do not sum across items in order to preserve the individuality of scale items. This permits the researcher to more closely examine the content of a respondent's attitude. Thus, the Likert and Thurstone scales permit the use of several judgments which are considered alternative measures of the same attitude. Analyses are sometimes performed on each individual item. Likewise, an index of a person's attitude may be computed on the basis of different measures. Items determined to measure the same attitude are combined to construct an overall index of attitude. With such an index individuals are usually typed as having a favorable, unfavorable or neutral position. Guttman (1944) offered a unidimensional technique for measuring attitude. In this method, attitude statements are rank-ordered by favorableness such that if a person agrees to one item, he will agree to all that follow. The higher the rank of the item and the more items the respondent agrees with, the more favorable his attitude. The result is a unidimensional measure of a person's favorable- unfavorableness toward the subject of interest. 9 Osgood, Tannenbaum, and Suci's (1957) semantic differential has also been used to measure attitudes. With this method, a set of bipolar adjectives (e.g., good-bad, rich-poor, happy-sad) are used as a series of attitude scales. Respondents are asked to indicate their reactions to the object of interest by placing an X in one of the seven spaces provided between each adjective pair. Values can be assigned to each space and an attitude score computed by summing across all the pairs. Likewise, an attitude can be represented without summing by examining the pattern of responses of all the subjects. Often semantic differential scales are subjected to R— factor analysis to examine the dimensions of people's feelings about the object of interest. Factor analysis is also the foundation of another technique. Stephenson's (1953) Q-methodology, however, utilizes Q-factor analysis, which factors persons rather than items. With Stephenson's method- ology the content of a person's attitude is of primary interest rather than a single score of favorableness—unfavorableness. In 0, respondents are asked to sort attitude statements in a forced or free choice distribution along a continuum of agree-disagree.cnr favorable- unfavorable,much like the sorting done by the judges in the Thurstone technique. The difference is that subjects scale the items rather than judges. Rather than summing across all items, persons' scores are analyzed using Q-factor analysis which yields a natural, or operant, grouping of individuals with similar sorts, or item scores. Thus, 10 Q-factor analysis allows the researcher to group people on the basis of all their scores considered individually. Stephenson refers to these groups as types, "a class of persons having a common character- istic or characteristics" (1953). Thus in research using attitude items, the groups can be considered attitudinally homogenous. In addition, the 0 analysis yields a rank-ordered array of attitude items for each type. These arrays permit the research to analyze the content of each type's attitude. The importance of knowing the content of an attitude has been pointed out by Stephenson (cited in Brenner, 1972): The status of [surveys and polling] is suspect in theoreti- cal respects . . . they ask questions which require single answers pro or con, yea or nay, neutral, no opinion, undecided, or the like. . . . These responses . . . represent broad standpoints or viewpoints without reference to the probably innumerable opinions . . . that enter into them. (p. 348) These techniques for measuring attitudes are by no means exhaustive, nor are they mutually exclusive. Many variations and com- binations of these techniques have been utilized. For example, Mauldin, Sutherland and Hofmeister (1978) and Adams (1972) utilized Likert-type scales in conjunction with Q-factor analysis. In a review of attitude research published between 1968 and 1972, Fishbein and" Ajzen (1972) found more than 500 different operations designed to measure attitude. ATTITUDES AND MASS MEDIA RESEARCH Since studies of mass media borrow theories, constructs, and approaches from the behavioral sciences, it is obvious that the concept of attitude will be found in studies of mass media. As Liebert and Schwartzberg (1977) wrote, “. . . the effects of media seem neces- sarily to depend on patterns of audience use, the nature of the material to which the audience is exposed, the degree to which such exposure transmits information and cultivates beliefs, and finally the extent to which media-cultivated information and beliefs influence the overt expression of social attitudes and behavior [emphasis added]" (p. 141). The ability of mass media to transmit information and culti- vate beliefs is the central link between mass media content and behavioral and social effects. However, the transmission of informa- tion will have no behavioral effects unless the information is comprehended, accepted, and absorbed by the message recipient so as to influence his or her attitudes (Libert and Schwartzberg, 1977). In this review of mass media attitude research, we will examine various kinds of attitude research. The review will provide examples of attitude research concerned with representing a person's attitude by a single score measured either by a single scale or by summing across several scales. This approach will be referred to as a "unidimensional" approach. 11 12 Another approach that we will review is concerned with the content of the attitude rather than generating a single score. Such studies examine each attitude scale individually and as a part of the overall attitude. This approach will be referred to as a "content" study. This review will also point out ways in which attitude scores are used to classify, or type people. One typing procedure, called scission typing (Stephenson, 1953), defined by cuts across normal or other distributions on a single continuum. Scission attitude typing would use a single attitude score along a favorable-unfavorable con- tinuum to group individuals with similar attitude scores. Typically, this typing results in some percentage of respondents being classified as having a favorable attitude; some percentage unfavorable; and some percentage neutral. Another typing approach is multivariate typing, defined as many cuts across many distributions. This approach uses each person's score on each attitude item in the typing procedure. Stephenson's (1953) Q-methodology yields such a typing. Attitudes Reflect Mass Media Effects There are numerous studies of mass media effects which operate under the paradigm just described. Examples of this research are discussed below. This discussion is not meant to be exhaustive, but rather to expose the reader to some typical mass media attitude research. Much attention has been given to the transmission of political information and the cultivation of political beliefs. Many of these 13 were concerned with the effect of media content on attitudes toward Watergate as well as attitudes toward media coverage of Watergate (e.g., O'Keefe and Mendelsohn, 1974; Robinson, 1972; Robinson, 1974). Others were concerned with attitudes about the Vietnam War. Adams (1977), for example, found in survey research that a significantly greater number of combat-experienced veterans described as "poor" television's [Jerformance in providing the viewing public an accurate image of the war. A third of the non—combat veterans rated television coverage as "good.“ In a longitudinal field study conducted in 1967 and 1968, Brenner (1972), using Q-methodology, discovered a polarization in public attitudes toward Vietnam. Q-methodology, as discussed earlier, utilizes all of a person's responses to attitude itemS' to find persons with similar responses. It is an example of multivariate typing. Rather than summing across items, each item response is utilized. This results in a natural, or operant (Stephenson, 1953) grouping of individuals. These groups can be considered attitudinally homogenous. With a set of 63 attitude items, Brenner found four groups of people with essentially different attitudes which between 1967 and 1968 became more defined and more polarized. These attitude groups were characterized as (1) doves, whose belief structure provided evidence that they considered the Vietnam war a tragedy to be terminated as soon as possible; (2) highly defensive hawks, who saw the U.S. as having a Christian duty to stop communism anywhere in the world; (3) moderate, reasonable hawks, who kept military aspects of Vietnam in perspective and recognized the 14 right to dissent, and (4) self—centered hawks, who rejected the concept of the.U.S. as overseer of the world's rights, but who felt the U.S. can go anywhere to protect its own interests (Brenner, 1972). Interestingly, had Brenner averaged the scores of these respondents, as many researchers do, he would not have discovered these different attitudes, but rather, he would have reported a "meaningless" average. Brenner's study provides evidence for the need to examine attitude data for the existence of operant, or natural, attitude groups before analyzing data. This prevents making a faulty assumption that all respondents are the same and that the best statistic for representing public attitudes is the mean attitude score. The effect of polls, endorsements, issue and candidate infor- mation have also been studied. Paid political advertisements have their greatest effects on late-deciders; however, candidates were rejected after exposure to their ads rather than selected because of them (Atkin, Bowen, Nayman, and Sheinkopf, 1973). Exposure to the results of pre-election polls influences voting, but only when the polls suggest a degree of support relatively stronger than the message recipient expected (Atkin, 1969). Agenda setting has also been investigated. This research centers on the hypothesis that the news media communicate to the public the relative importance of various political issues by extend- ing differential coverage priorities to them (Liebert and Schwartzberg, 1977). The most popular approach has been to search for positive correlations between media attention to various issues and survey data l5 measuring perceived importance of these issues. Liebert and Schwartzberg (1977) report research that has shown a cumulative effect of media agenda setting upon the personal agendas of individuals and that agenda setting through newspapers seems to have its greatest effect on older voters and those who skim the medium. Douglas, Westley, and Chaffee (1970) examined the effects of a multimedia campaign designed to provide information and improve atti- tudes toward the mentally retarded. Using 21 Likert items, the authors used a summed attitude score and found a significant attitude change and a moderately strong positive correlation between information gain and attitude change. This is a good example of a unidimensional approach with scission typing. It is interesting to note that these authors used a summed score and assumed homogeneity of effects on respondents. If there were different attitudes, like those found by Mauldin et a1. (1978) and like those found by Brenner's Vietnam study, the results may be questionable. Hiett and his associates (1969) found no significant effect for simple exposure to a series of print and television ads favoring gun control. O'Keefe (1971) surveyed almost 1,000 persons to determine the effects of antismoking commer- cials on attitudes toward smoking. The majority of smokers believed that smoking was bad for their health, but few were inclined to stop. Nonsmokers thought the campaign was more effective than smokers did. Ward (1972) established that understanding of television com- mercials increases with age of the child and that as children grow older they are increasingly skeptical about the truthfulness of these 16 messages. Culley, Lazer, and Atkin (1976) investigated the beliefs of six a priori defined groups (students, townspeople, advertising agency persons, advertisers, Action for Children's Television [ACT], and government) on major issues regarding children's television adver- tising. Using 29 attitude items measured on 2 Likert-type scales, these researchers found a general lack of understanding of positions among the groups. For example, 70 percent of the students, townspeople, ACT and government agreed that, "Most advertisers on children's television are not really concerned about kids; they just want to sell their products" (Culley et al., 1976). Advertisers and ad agency persons disagreed with the statement. This study differs from other studies that have been discussed regarding different attitude groups (e.g., Brenner, 1972). This study defined the groups a priori while Brenner identified his groups on the basis of their attitudinal responses. The a priori assumption in the Culley study is that each of the groups are attitudinally homogenous. Brenner's study provided evidence that such an assumption may lead to misleading results. In addition to these studies of effects, the federal government has released two major reports of the effects of media on (1) attitudes toward pornography (Report of the Commission on Obscenity and Pornogra— lphy, 1970) and (2) the relationship between TV violence and antisocial attitudes and behavior (Television and Growing Up: The Impact of Televised Violence, 1972). Beyond the study of mass media effects, attitude is a useful concept in the study of mass media public policy and in providing a 17 consumer data base for managerial decisions himedia and advertis- ing. Attitudes Can Be Used for Public Policy_Decisions The need for attitude research in public policy regarding television station licensing can be considered a mandate. Since licensees are expected to operate their stations in the public inter- est and convenience, it seems licensees must have an understanding of community attitudes. In February 1971, the Federal Communications Commission released its Primer on Ascertainment of Community Problems (27 FCC 2d 650). This primer presents a set of guidelines for applicants to follow in identifying and responding to the "problems, needs and interests" of their communities. The Commission's requirement that "a sufficient number of members of the general public to assure a generally random survey must . . . be consulted" (27 FCC 2d 684), dictates some type of survey attitude research (e.g., Foley, 1972). This type of attitude survey typically follows a unidimensional approach and scission typing. However, multivariate typing has also been used. Adams (1972) used a multivariate approach in a study of broadcasters' attitudes toward their community service responsibility. His typing procedure also allowed him to study attitude content. Adams measured attitude using 40 statements which represented the major dimensions of broad- cast responsibility to the community. Broadcasters were asked to indicate their reaction to each statement along a scale ranging from 18 strongly agree to strongly disagree. Using Q-factor analysis Adams found five different types of broadcasters which clearly indicated strong ". . . differences in the patterns of opinions among broadcasters as to how the public interest of the community is to be served" (Adams, 1972, p. 419). This analysis not only uncovered five different attitude groups, but it allowed Adams to examine the content of each group's attitude as indicated by the responses to each attitude state- ment. In related policy research, Meeske and Handberg (1976) investi- gated news directors' attitudes toward the Fairness Doctrine. Their results indicate a lack of strong feeling for repeal of the Fairness Doctrine among the news directors. Attitude research has also been suggested as a solution to the task of identifying deceptive advertising faced by the Federal Trade Commission. Thus far, this body of law has grown out of case-by- case rulings that have been prescriptive rather than prescriptive. The result is that no overall definition or classification of deception in advertising exists (Gardner, 1975). To overcome this problem, Gardner (1975) suggests classifying deceptive advertising into three classes: Unconscionable Lie, Claim-Fact Discrepancy, and Claim- Belief Interaction. While the unconscionable lie is a matter of fact, the latter two involve consumer attitudes. Deception is defined operationally and behaviorally in terms of consumer attitudes. 19 Attitudes Can Be Used to Make Media Content Decisions Attitude research is important to decision makers because it provides answers to the many questions they have about the public, consumers, readers, and viewers. Publishers and editors want to know how people feel about their newspapers. Broadcasters are equally sensitive to public approval. A recent report issued by Yankelovich, Skelly and White, Inc., Young People and Newspapers (1976), describes the attitudes of young people toward newspaper content. Among the implications of the results lire 'that. young people would like to see more in the paper about the kinds of things they are interested in. Schweitzer (1976) suggests the implication to the editor is to report thoroughly important issues in international, national, and local news, but not necessarily in the same ggily fashion as has been traditional. This would allow more space to pursue some of the interests of the young readers in more depth and breadth. Stephenson (1964b) conducted several Q-studies to test his "Ludenic Theory" of newspaper reading. The attitude items used in the Q-sort referred to the reader's feelings and understanding of newspaper content. We found three attitude groups described as (l) the "well- rounded" reader, interested in national and international news and who desires a daily metropolitan paper; (2) a "pleasure-readerV interested in features, human interest and sports; and (3) nonreaders. An editor could use such information in meeting the demands of each group for specific content. 20 Nielson (1974) used an expectancy-value attitude model in investigating viewer attitudes toward various television programs, a unidimensional and scission—typing approach. Based on his results, Nielson concluded that attitude research was valuable for (1) deter- mining or selecting television programs for various audience segments; (2) for measuring how existing television programs are meeting audience groups' needs and wants; and (3) for providing information on how to change existing programs to improve audience satisfaction with pro- gramming. Stone (1973) investigated viewer attitudes toward tele- vision newswomen. The objective is not to schedule a newswoman in a situation that would irritate viewers. Stone found that preferences varied across situations and thus it is possible to schedule newswomen in a manner consistent with viewer preference. Monaghan, Plurrmer, Rarich, and Williams (1974) utilized 0— methodology to investigate viewer preference for new television program concepts. Respondents were asked to sort statements describing possible program content from least preferred to most preferred. They found two types of viewers: one preferred programs that were reality-oriented, that is, containing conflict between good and evil. The second preferred nostalgia, light humor and realistic science. The ability of the researcher to study the content of the attitude of each type permits the programmer to plan programs suited to the interest of each audience. Attitude data is also useful to media managers in selling advertising time or space. Traditionally, advertisers have purchased 21 media audiences described solely by demographic characteristics, such as age and sex. However, there is a growing demand among advertisers for descriptions of media audiences that go beyond demographics to include attitudinal characteristics, often labeled "psychographics" in the industry. A medium with a more complete audience description will be able to better meet the needs of an advertiser who is targeting his message to a particular target audience. Attitude Research Reveals Public Opinion One of the major concerns of media is content. While attitude research can tell an editor what his readers or viewers want, attitude research in the form of public opinion surveys becomes content. Public opinion is news. The Gallup Poll, the Harris Poll, the NBC Poll, Roper, and others provide attitude research that indicate the public attitude toward social and economic issues, as well as government officials and policies, and other current events. In its first issue, Public Opinion (1978) reports on the public attitude toward Carter's Presidency, the economy, foreign policy, 1978 Congressional elections, Sadat's trip to Jerusalem, the Panama Canal Treaty, and physical exercise, among other issues. Katona and Strumpel (1978) discuss the difficulties of economic forecasting. They report being able to improve economic forecasting by measuring and analyzing people's expectations (attitudes), ". . . for their expectations play a significant role in shaping sub- sequent economic behavior" (p. 10). 22 The United States Information Agency (U.S.I.A.) also uses attitude research to measure public opinion. As a part of its work in foreign countries, the U.S.I.A. surveys public opinion toward the U.S. and the effects of various information programs such as Voice of America. Usually this is a unidimensional, sciSsion typing study. In addition to the typing of individuals as favorable or unfavorable or neutral, multivariate typing has also been used to study public opinion. Brenner's study (1972) of attitudes toward Vietnam, discussed previously, is an example. Stephenson's study (1963) of public attitudes toward public utilities is another excellent example. Attitudes Help_in Making Marketing_and Advertising Decisions Attitudes have long been of interest in marketing and adver- tising. Successful marketing and advertising strategy is determined by how well the decision maker understands the consumer. As McKitterick (1969) stated, ". . . the principal task of the marketing function in a management concept is not so much to be skillful in making the consumer do what suits the interests of the business as to be skillful in con- ceiving and then making the business do what suits the interests of the consumer" (p. 79). Marketers need to know consumers' attitudes toward goods in the marketplace and which attitudes predispose them to a par- ticular purchase. Since men began to exchange goods and services, attitude research has existed in at least some crude form. Sellers have 23 always been interested in finding out if customers 1iked--disliked, wanted--did not want their products. However, since McKitterick's formalization of the marketing concept in 1951 and the development of improved methods to assess consumer attitudes, attitude research has become a major emphasis in marketing and advertising research. With the introduction of Colley's (1961) Definigg Advertising Goals for Measured Advertisigg_Resu1ts, commonly referred to as DAGMAR, attitudes were recognized as appropriate advertising goals. The DAGMAR approach to advertising planning is best summarized in its succinct statement defining an advertising goal. An advertising goal is a specific communication task, to be accomplished among a defined audience, in a given period of time. Thus, as Aaker and Myers (1975) stated,"It.is recognized that advertising is mass, paid communication that is intended to create awareness, impart information, develop attitudes. . . " (p. 100). While attitudes are appropriate goals, attitude research provides a measure of the effectiveness of a campaign in accomplishing its goals. In planning a message, an advertiser should consider the attitudes of his target audience for it is among those attitudes that his message will be perceived, interpreted and acted upon (Cox, 1961). Thus attitude data are useful for strategic and tactical message decisions: selling propositions, positioning, tone, setting, casting characters, and message structure (Mauldin et al., 1978). Attitudes can also be used in market segmentation. Smith (1956) defined market segmentation as a strategy of viewing a 24 heterogenous market as a number of small homogenous markets with different product preferences that is attributable to the desires of consumers for more precise satisfaction of their varying wants. In a recent article, Percy (1976) presented a general paradigm for attitude segmentation, multivariate typing, basing the actual segmen- tation on attitudes and proposing additional analyses to incorporate demographic and perceptual data and behavioral information about product class, brand, and media use into the development of marketing and advertising strategies and media plans. The argument favoring attitudes as independent variables in identifying homogenous groups, or segments, is based on the theoretical assumption that persons consume to satisfy needs and wants, an idea Adam Smith introduced in economic theory as "value in use" in 1776. Say described the concept in 1821 as "utility," and Alderson used the same concept in 1957 to justify defining products as "bundles of utilities." Haley (1968) used the concept in saying that segmentation should be based on attitudes, which he argued are causal, whereas demographic and behavioral variables are descriptive but not causal. Thus, the criteria for selecting a market segmentation variable do not come from specific consumer data but from assumptions about human behavior: that persons behave to satisfy wants and needs and that attitudes form to help persons achieve wants and needs (Mauldin et al., 1978). Attitudes have been used by many researchers in market seg- mentation. Bartos and Dunn (1976) for example, used a unidimensional 25 approach and scission typing to segment consumers into groups with similar favorable-unfavorable scores. Other researchers have used a multivariate approach to attitude segmentation. Adams (1972), dis- cussed previously, used such an approach. 'Haley (1968) used attitude segmentation, which he labels "Benefit Segmentation," to segment the consumer toothpaste market. Mauldin et a1. (1978) used a multivariate attitude segmented data to select a market target and choose a new product name. Stephenson (1953) has also argued in favor of this approach. Attitude and Mass Media Research: Conclusions In this review of attitude reserach in mass media, two points stand out. One is the heuristic nature of the concept attitude. It is versatile and as McGuire (1966) suggests, ". . . the creative phase of research is so idiosyncratic that each researcher is best left to his own preferred mode of conceptualization" (p. 477). Second, much of the research is survey. Mass media have mass audiences. Hence, to study mass media effects we often need to study mass audiences. The most effective and efficient way to study such mass audiences is survey research. However, these two points represent problems for the mass media researcher. First, the definitional question of what exactly an attitude is, both theoretically and operationally, is left to each individual researcher. Thus, we have many and diverse concep- tualizations. As a result, much of the research is not directly comparable or cumulative. 26 The second problem is that many of the analyses of attitude scales yield a single score which indicates degree of favorableness- unfavorableness, but does not reveal attitude content. Individuals with the same score may have different attitude structures. Thus, in developing a general definition of, and operations for, attitude, there is a need for operations which allow the researcher to examine responses to each attitude item, that is, to deal with the content of attitudes. In this manner, the researcher will be able to assess the respondents' overall attitude as well as the content of that attitude. This is consistent with Rokeach's (1968) definition of an attitude as a "set of beliefs." The researcher should be able to examine each individual belief as well as the set of beliefs as a whole. Lastly, we have seen that market segmentation research (Haley, 1968; Stephenson, 1953; Mauldin et al., 1978) and other attitude research (Adams, 1972; Brenner, 1972) have found groups with essen- tially different attitudes in survey research. The implication is that research which assumes homogeneity of respondents may yield meaningless results. Researchers should examine survey data through multivariate typing for the existence of different attitude groups bgjgrg_proceeding with any analyses. Briefly, we have seen the need to study attitudes. We have seen that we need to do it in surveys and that these attitude surveys should yield a measure of favorability-unfavorability. We have seen the need to look at the content of attitudes. And, lastly, we have seen the need to segment persons into groups which are attitudinally homogenous. 27 An Approach to These Research Needs Fishbein's attitude model (Fishbein and Ajzen, 1975) offers a possible solution to the lack of conceptualization and comparable Operations. Fishbein defines attitude as a unidimensional concept that is represented by a subject's location on a bipolar affective or evaluative dimension (good-bad) vis-a-vis a given object (Fishbein and Ajzen, 1975). Fishbein and Ajzen (1975) suggest that concepts such as opinion, satisfaction, prejudice, intention, value, and belief have been used in measures of evaluation and have confounded the distinction between attitude and other concepts. Distinctions have also been suggested by other authors (e.g., Rokeach, 1968; Triandis, 1971; Harvey, Hunt, and Schroeder, 1961; Katz, 1960; Osgood, Suci, and Tannenbaum, 1957). FISHBEIN ATTITUDE MODEL Working within a behavior theory framework, Fishbein(l963, 1967b, 1967c, 1967d; and Fishbein and Ajzen, 1972, 1975) derived his attitude model based upon principles of mediated (secondary or con- ditioned generalizations. Essentially the theory may be stated as follows (1963, p. 233): 1. An individual holds many beliefs about any given object, i.e., many different attributes, character- istics, goals, objects and values are positively or negatively associated with a given object. 2. Associated with each of these related objects is a mediating evaluative response. 3. These evaluative responses summate. 4. Through the mediation process, the summed evalua- tive response is associated with the attitude object. 5. Thus on future occasions the attitude object will elicit this summated evaluative response, i.e., an attitude. According to this theory an individual's attitude toward any object is a function of his beliefs about the object and the evaluative aspect of those beliefs. Expressed algebraically: where: bi = the strength of belief "i" about the attitude object "0," that is, the probability or improbability that 28 n: This 29 "o" is related to some other object "Xi" (e.g., the probability that Brand X is carbonated; that Carter is a Democrat). the evaluative aspect of bi, that is, the evaluation of "Xi"--its goodness or badness (e.g., the evaluation of c arbonation; Democrats). number of beliefs. approach is similar to Rosenberg's model (1956) which is based on a functional approach to attitudes (Katz, 1960). Rosenberg hypothesized that a person's attitude toward a given object would be a cognitive structure made up of beliefs about the potentialities of that object for attaining or blocking the realization of valued states. The more the object leads to attainment of positively valued states or blocks the attainment of negatively valued states, the more the person would have a positive attitude toward the object. Algebraically, Rosenberg's hypothesis may be expressed as follows: n—c II where: < do II 3 ll perceived instrumentality, the extent to which the person believes that the object "0" will lead to or block the attainment of value i. value importance, value “i's” importance to the person as a source of satisfaction. number of value. 30 Notice that Vi is not a measure of "importance," but a measure of "satisfaction" or "evaluation." Values, consequence, or attributes, then, are "important" strictly along a continuum bounded by "gives me maximum dissatisfaction" and "gives me maximum satisfaction." As Peak (1955) points out: It was assumed that the effect attached to an attitude object would be some function of (l) the judged probability that the object leads to good or bad consequences (2) the intensity of the effect expected from the consequences. (p. 154) In a mass media context, both models require the following information: (1) the extent to which a person believes that the object (i.e., newspaper, television program, candidate, issue, product, or brand) is related to or possesses a characteristic or attribute and (2) his evaluation of or satisfaction he would derive from each attribute. While the Rosenberg and Fishbein models are similar concep- tually, they also share the following admirable properties (Lutz and Bettman, 1977): 1. Strong foundations in psychological theory 2. Clearly specified constructs and measurement procedures 3. Theoretically derived attribute combination rules (i.e., multiplication and summation) These models have received considerable attention in marketing research (Wilkie and Pessemier, 1973; Lutz and Bettmann, 1977) both theoretically and in applied research (Tuck, 1973). Their use was 31 popularized by Myers and Alpert's (1968) concept of a "determinant attribute." This concept suggests that among all the functions of a product or brand, there are some attributes which have a greater influence on a consumer's attitude toward a product. The attribute(s) which are more closely related to preference are called "determinant attribute(s)." This is consistent with a current definition of a "product" suggested by Kotler. He defines a product as, ". . . a bundle of physical, service and symbolic particulars expected to yield satisfactions or benefits to the buyer" (Kotler, 1967, p. 289). Mass media fit easily into this definition. A newspaper, a television program, a public issue, a political candidate, etc., can be seen as "bundles of utilities." Thus, most mass media research deal with some "product," and each product consists of a set of particulars, or attributes. The notion of determinant attribute(s) implies a correlation between a consumer's evaluation of product attributes and his attitude toward that product. This does not necessarily establish a causal link, but one might expect high preference for a product perceived to have the determinant attribute(s). With the focus of research on identifying important attri- butes, these models began to receive greater attention from researchers. Under this general paradigm, the Rosenberg and Fishbein models became known as "multiattribute attitude models." In general, it has been established that these models are good predictors of overall evaluation or attitude, whereas their 32 ability to predict behavior has varied. This variance in predicting behavior is not surprising when one closely examines the models. In both cases, the dependent variable is attitude or overall affect, not behavior. In fact, Fishbein has developed a different model for the study of the relationship between behavior and behavior intention (Fishbein and Ajzen, 1975). Lutz and Bettmann (1977) pOint out the advantages of the multiattribute attitude models (p. 137): 1. The Diagnostic Tool. Multiattributemodels are not only a means for measuring consumer attitudes but also for diagnosing these attitudes and suggesting attitude change strategies. This is, by careful analysis of consumers' perceptions of product attributes, the manager can alter his market offering (most notably the pro- motion element) to favorable influence attitudes and behaviors toward his product. The most cogent statements of this rationale can be found in Boyd, Ray and Strong (1972) and Cohen (1974). 2. Intuitive Appeal. It makes good sense that consumers select products that possess desirable attributes and reject those which do not. Kotler's (1967) definition of a product was couched in these terms: "a bundle of physi- cal, service and symbolic particulars expected to yield satisfactions or benefits to the buyer" (p. 289). Sub- stitution of the word attributes for particulars in this definition does little to change its meaning but does clarify the centrality of product attributes in marketing theory. 3. Industry Acceptance. Although more practitioners have published psthographic studies than is the case with the multiattribute model, it is still unusual to find major marketing survey questionnaire that does not have a section on perceptions of brand attributes. Increasing acceptance by industry encourages further application and refinement of the multiattribute model. 4. Research Pragmatics. Data conforming to a multiattribute model are easy to collect and susceptible to analytical techniques ranging from simple to extremely sophisticated 33 ones. Unfortunately, the pragmatics of publishing and/or perishing dictate the interest of many researchers in this model. 5. Theoretical Relevance. A final reason for the popularity of multiattribute models is their linkage to basic psycho- logical processes of the individual and to the earlier works in psychology by Fishbein (1963) and Rosenberg (1956). These theories in turn are closely related to even earlier thinking by Lewin (1938). Thus a strong research tradition underlies these models as explanatory constructs that can help in yielding insights into consumer behavior. Since current interest focuses on the Fishbein model, which is perhaps the most influential model (Calder and Lutz, 1972), the following portion of this paper will focus on the Fishbein model. A discussion of the Fishbein model, its assumptions, its operations, and its limitations will be presented. In addition, a possible solution to the limitations of the Fishbein model in mass media attitude research will be outlined. Fishbein's Conceptualization of Attitude Before beginning a discussion of Fishbein's theory, it should be noted that one of the reasons his model has received so much atten- tion is that he has gone to great ends to clearly and cogently provide the theoretical foundations and operations for attitude theory. His text, Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research, written with Ajzen (1975), provides the most complete description of his work in a single volume. Because his work is clearly stated there, much of the following discussion is taken directly from that text. Fishbein's conceptualization of an attitude is consistent with Thurstone's (1929) which defined attitude as the amount of 34 affect for or against some object. Fishbein (Fishbein and Ajzen, 1975) suggests that attitude should be measured by a procedure which locates an individual on a bipolar affective or evaluative dimension vis-a-vis a given object. He uses the terms "affect" and "evaluation" synonymously. "Although it might be argued that there is a difference between a person's judgment that an object makes him feel good and his evaluation that the object is good, there is little evidence to suggest that a reliable empirical distinction between these two variables can be made" (Fishbein and Ajzen, 1975, p. 11). He limits attitude to the affect component of the venerated attitude triology: cognition, affect, conation. Cognition denotes knowledge, opinions, beliefs; affect refers to feelings toward and evaluation of some object; and conation refers to behavioral intention with respect to the object. Fishbein adds a fourth component, behavioral intention, to this trilogy because ". . . we are concerned with predispositions to behave rather than with behavior itself. . . " (Fishbein and Ajzen, 1975, p. 12). To better understand Fishbein's conceptualization, it is therefore useful to define these four broad categories: 1. Belief: Specifically, a belief links an object to some attribute. An object may be a person, a group of people, an institution, a behavior, a policy, an event, etc., and the associated attribute may be any object, trait, property, quality, character- istic, outcome or event. Thus, belief can be 35 operationally defined by a procedure which places a person along a dimension of subjective probability relating an object to some related attribute. For example, Fishbein measured the following belief in a study of political candidates: Lyndon B. Johnson is in favor of using nuclear weapons in Vietnam. probable improbable 2. Affect or attitude is the amount of affect for or againstsome object. Attitude is measured by a procedure which locates a person along a bipolar affectivecw'evaluative dimension. For example, Using nuclear weapons in Vietnam is . . . good bad 3. Behavioral Intention: Refers to a person's intention to perform various behaviors. Intention can be viewed as a special case of beliefs, in which the object is always the person himself and the attribute is always behavior. When a probability dimension links a person to a behavior, behavioral intention is measured. 4. Behavior: Refers to overt behavior. An observable agt_studied in its own right or used to infer beliefs, attitudes, or intentions. Under this conceptualization, beliefs are the fundamental building blocks. A person forms beliefs or learns about an object through 36 direct experience or communication from other sources. In this manner, he forms beliefs about himself, about other people, about institutions, behaviors, events, etc. The totality of a person's beliefs serves as the informational base that ultimately determines his attitudes, intentions, and behaviors. Thus a person's attitude toward some object is related to the set of his beliefs about the object but not necessarily to any specific belief. This conceptualization is consistent with Rokeach's definition (1968) and compatible with Stephenson's (1963) call for study of the content of an attitude. The Fishbein position is best summarized (Fishbein and Ajzen, 1975): Like most other investigators, we agree that an attitude can be described as a learned predisposition to respond in a consistently favorable or unfavorable manner with respect to a given object. It should be clear that since a person's attitude is assumed to be related to the total affect associated with his beliefs, intentions, and behaviors, we define response consistency in terms of overall evaluative consistency. Thus attitude is viewed as a general predispo- sition that does not predispose the person to per orm any specific behavior. Rather, it leads to a set of intentions that indicate a certain amount of affect toward the object in question. Each of these intentions is related to a specific behavior, and thus the overall affect expressed by the pattern of a person's actions with respect to the object also corresponds to this attitude toward the object. . Some attitudes may be relatively stable over time, and others may exhibit frequent shifts. At any point in time, however, a person's attitude toward an object may be viewed as determined by his salient set of beliefs about the object. (p. 15) Salient Beliefs Fishbein argues that although a person may hold a large number of beliefs about any given object, it appears that only a relatively small number of beliefs serve as determinants of his attitude at any 37 given moment. Fishbein's evidence is research of attention span, apprehension, and information processing that suggests individuals are capable of attending to or processing only five to nine items of information at a time (e.g., Miller, 1956; Woodworth and Schlosberg, 1954; Mandler, 1967). He therefore argued that a person's attitude toward an object is primarily determined by no more than five to nine beliefs about the object; these are the beliefs that are salient at a given point in time. It is, of course, possible for more than nine beliefs to be salient and to determine a person's attitude; given time and incentive, a person may take a much larger set of beliefs into account. Fishbein suggests that under most circumstances, a small number of beliefs serve as the determinants of a person's attitude. This is consistent with Myers and Alpert's (1968) concep- tualization of "determinant attribute." The Identification of Salient Beliefs In order to construct attitude items, Fishbein (1967d) recom- mends a free-response format in which a person's beliefs about a given object are elicited by asking him to list characteristics, qualities, and attributes of the object. He argues that salient beliefs are elicited first and thus, consistent with the attention span, apprehension and information processing theory, beliefs elicited beyond the first nine or ten are probably not salient for the indi- vidual (Fishbein, l967d). This thinking centers on Fishbein's learning theory approach to attitudes which suggests that attitudes are learned initially as a 38 part of concept formation. Once a concept is learned, the individual learns new things about it, that is, he associates many different objects, concepts, values, or goals with the attitude object. This set of associations the individual learns is viewed as a belief system--a "habit-family-hierarchy of responses" (Fishbein, l967d). The higher the response in the hierarchy, the greater the probability the responses is associated with the concept, that is, the stronger the belief. Stronger beliefs have a greater amount of evaluative response, and thus a greater effect on the overall attitude toward the concept. To determine salient beliefs for their study of political candidates, Fishbein and his associates conducted interviews with a small independent sample of the larger population they were interested in studying (Fishbein and Coombs, 1974). In a free-response format, persons were asked to answer the following two questions: (1) What are the characteristics, qualities, and attributes of each candidate? (2) What do you think are the relevant issued in this campaign? The resulting 24 belief statements are presented in Table l. The questionnaire was then administered to a sample of over 600 residents of two midwestern communities. Each belief statement was rated on a seven-point probable-improbable scale, and each attribute was rated on a seven—point, good-bad scale. The product of these two measures, summed over the 24 beliefs, served as estimates of attitude toward the two candidates. These estimates correlated .69 and .87 with direct measures of attitudes toward Johnson and Goldwater, respectively. 39 Table l. Belief Statements Used in Study of Attitudes Toward Presidential Candidates in the 1964 Election 01-wa 12. 13. 14. 15. 16. 17. 18. 19. 20. OkOCDNOS Lyndon B. Johnson (Barry Goldwater) is a Republican is a Democrat is consistent in his views is a conservative is a moderate is a liberal is physically healthy is mentally healthy is a political opportunist is in favor of our present foreign policy in Vietnam is in favor of the antipoverty bill is in favor of reducing the power of the Supreme Court is in favor of allowing military personnel to make decisions about the use of nuclear weapons is in favor of Medicare selected a well-qualified running mate for Vice President is in favor of political extremism is in favor of price supports for farm products is in favor of using nuclear weapons in Vietnam is in favor of swift enforcement of the Civil Rights Law is in favor of increased Social Security benefits 40 Table l--Continued 21. is in favor of reducing the power of the federal government 22. approves of the John Birch Society 23. is in favor of the Nuclear Test Ban Treaty 24. approves of the Americans for Democratic Action Source: M. Fishbein and F. S. Coombs, Journal of Applied Social Psy- chology, 1974, 4, 2, pp. 95-124. Reprinted by permission of V. H. Winston & Sons, 7961 Eastern Avenue, Silver Spring, Maryland 20910. 41 It is possible, however, that only the first two or three beliefs are salient for a given individual and that additional beliefs elicited beyond this point are not primary determinants of his attitude (i.e., are not salient). "Unfortunately, it is im- possible to determine the point at which a person starts eliciting nonsalient beliefs" (Fishbein and Ajzen, 1975, p. 218). Fishbein and Ajzen (1975) point out another problem in that the elicitation procedure itself may produce changes in a person's belief hierarchy. Previously nonsalient beliefs may become salient once they have been elicited. This implies that mere elicitation of beliefs may change a person's attitude. While listing his beliefs about an object, the person may recall some information he had forgotten or make a new inference on the basis of existing beliefs. The previously nonsalient beliefs may become important determinants of his attitude. Fishbein and Ajzen (1975) conclude that under these circumstances, the first few beliefs elicited will be highly related to the person's attitude as it existed prior to the elicitation of beliefs, but they may have a somewhat lower relationship to his attitude following elicitation. Similar problems emerge when a person responds to a standard set of belief statements, such as an attitude scale. Fishbein and Ajzen (1975) suggest, however, that although salient beliefs are viewed as the primary determinants of attitude, nonsalient beliefs can nevertheless be used to measure atti- tude. In fact, standard attitude scales may comprise in large part statements concerning nonsalient beliefs. Responses to these statements 42 are largely inferences consistent with the beliefs held by the person, and thus they, too, are likely to be predictive of his attitude (Fishbein and Ajzen, 1975). This elicitation experience may change his salient beliefs and thus affect his attitude. As with many other phenomena, attempts to assess salient beliefs may influence the phenomenon under inVestigation. Fishbein and Ajzen (1975) conclude "it appears impossible to obtain a precise measure of the beliefs that determine an individual's attitude since the number of salient beliefs may vary from person to person. However, a rough approximation can be obtained by considering the first few beliefs (five to nine) as the basic determinants of attitude" (p. 219). In survey research it is desirable to have information about the salient beliefs in a given population (modal salient beliefs). For example, marketing research frequently attempts to iden- tify the determinants of attitudes toward some product. To ascertain modal salient beliefs within a given population, Fishbein and Ajzen (1975) suggest that a representative sample of the population be asked their beliefs about the product. The most frequently elicited beliefs should be considered the modal salient beliefs for the population. This approach, however, has its limitations. The effect of accepting only modal beliefs is to include in a questionnaire beliefs that are generally held by most people. This may prevent the collection of information that would provide a better understanding. of attitudes. By including a more heterogenous set of beliefs, a researcher will more likely be able to discriminate between people 43 with different beliefs. A set of modal beliefs will not likely include such discriminating items. Fishbein and Ajzen (1975) suggest that one possible solution is to take the 10 or 12 most frequently mentioned beliefs; this would allow for imperfect correspondence in the salient beliefs of different components of the population. Another possibility is to use those beliefs that exceed a percentage of all beliefs elicited (Fishbein and Ajzen, 1975, p. 219). These procedures are consistent with other research. For example, in marketing research, the basic criteria for specification of attribute lists require that they be exhaustive, semantically meaningful, subject to unidimensional interpretation, and reflect possible variations in choice or use contexts (Wilkie and Pessemier, 1973). Most researchers (e.g., Hansen, 1969) agree that attributes must reflect consumer perceptual dimensions rather than product characteristics directly measurable and controllable by a decision maker (i.e., product manager, editors, programmers, etc.). Methods for attribute generation include expert judgment and unstructured group or depth interviews. The unstructured or depth interviews are also recommended by Fishbein and Ajzen (1975) and Stephenson (1953). Difference between Salience and Importance A point of confusion may arise when salience is equated to importance. Rosenberg's (1956) concept of "value importance" equates 44 importance with the evaluation of the associated attribute, but there is considerable evidence that this evaluation is not related to belief salience (Fishbein, 1963). The term importance has been used to refer to (l) the per- ceived importance of an attribute for the person,annd(2) its perceived importance as a defining characteristic of the object, or (3) its perceived importance as a determinant of the person's attitude (Fishbein and Ajzen, 1975). The first of these usages is highly related to the polarity of the attribute's evaluation; that is, highly positive and highly negative attributes will tend to be perceived as important (Feldman and Fishbein, 1963). The perceived importance of an attribute as a defining characteristic of an object is close related to the subjective probability of an association between object and attribute. Thus, if a person has a high probability (i.e., strongly believes) that a newspaper is conservative, he is also likely to believe that conservatism is an important characteristic of newspapers. As was true with belief strength, this measure of importance cannot be used to determine whether a belief is salient or not. Another interpretation deals with the perceived importance of an attribute as a determinant of a person's attitude. Studies of cue utilization have found that subjective estimates of each cue's relative importance as a determinant of a given judgment do not correspond ix> weights obtained in a multiple regression analysis (Fishbein and Ajzen, 1975). Fishbein thus concludes that none of the different interpretations of belief importance can be used to derive measures that will identify salient and nonsalient beliefs. 45 One method of attempting to estimate importance of attributes as determinants of attitude has been to correlate each belief (taking evaluation of the attribute into account) with the attitude. As in any multiple regression approach, these correlation or regression weights are viewed as objective indices of importance. Fishbein and Ajzen (1975) point out, however, that regression weights provide no evidence as to causality, and ". . . it is therefore inappropriate to assume that a high correlation indicates an important or deter- minant of attitude or that a low correlation is evidence that the belief is not an important determinant of attitude" (p. 222). However, Cohen (1974) has suggested that a multiple regression analysis of beliefs retains the maximum amount of information regarding each biei component. In marketing this permits analysis of brand perception and position- ing. For example, if we know the salient brand choice attributes, it would be useful to know how each brand "stacked up" on each at- tribute. In conducting empirical research with the Fishbein model, the necessary information for this analysis is available. And what is more, the analysis is extremely straight-forward, comparatively nonsubjective and inexpensive. Summary of Fishbein's Model Fishbein's model is based upon the relationship between beliefs about an object and attitude toward that object. It is a descriptive model that is applicable to any set of beliefs, whether they are salient or nonsalient, new or old. Although Fishbein argued that a person's salient beliefs determine his attitude, the model 46 itself is not predicated on an assumption of causality but deals merely with the relation between beliefs and attitude. Specifically, it provides a description of the way in which different beliefs (and the evaluations of the associated attributes) are combined or integrated to arrive at an evaluation of the object. Thus, attitude can be expressed by the following equation in which A is the attitude toward an object, action, or event; b is the beliefs about the object's attributes or about the act's consequences; and e is the evaluations of the attributes or consequences. Or, as Fishbein (Fishbein and Ajzen, 1975) wrote: Thus, according to the model, a person's attitude toward an object can be estimated by multiplying his evaluation of each attribute associated with the object by his subjective probability that the object has that attribute and then sum- ming the products for the total set of beliefs. Similarly, a person's attitude toward a behavior can be estimatedby multiplying his evaluation of each of the behavior's conse- quences by his subjective probability that performing the behavior will lead to that consequence and then summing the products for the total set of beliefs. The terms "attribute" and "consequence" are used in a very general sense to refer to any aspect of an object or behavior, respectivelya-that is, to any characteristic, quality, object, concept, value, or goal associated with the object or behavior. (p. 223) However, persons holding the same beliefs may have very differ- ent attitudes and persons holding different beliefs may have the same attitudes. Attitudes are based on the total set of the person's salient beliefs and the evaluations associated with those beliefs. When the same beliefs are held with different strength or when 47 evaluations of associated attributes differ, attitudes will also be different. Conversely, when different beliefs are held with equal strength and when they have identical evaluative implications, the same attitudes will result. Fishbein and Ajzen (1975) thus conclude that knowledge of a person's attitude provides little information about the particular beliefs he holds or about his evaluations of attributes associated with the attitude object. This is consistent with a major premise of this paper that to fully understand an atti- tude, we need to study its content. Table 2 presents a hypothetical analysis of twoindividualS' attitudes using the Fishbein model. Notice that both individuals had an attitude score of 18; yet, it is easy to see that they have essentially different beliefs about the object under study. If one were to use a scission typing approach, those individuals would be considered to have the same attitude when really they have essentially different positions. Multivariate typing would overcome this problem. Evidence in Support of Fishbein's Model Considerable evidence exists in support of Fishbein's model. For example, Fishbein (1963) first stated his formulation of an expectancyevalue model with reference to attitudes toward Negroes. He constructed a set of 10 modal salient beliefs for his subject population by taking the 10 attributes that were elicited most fre- quently in response to the question: "What do you believe to be the characteristics of Negroes?" The 10 attributes, ordered in terms of 48 Table 2. Hypothetical Scores for Persons A and B Belief b e b . Person A 1 l 1 1 2 2 -l -2 3 O O O 4 -l -l 1 5 1 3 3 6 3 2 6 7 -3 -3 _§l A0 = 18 Person B 1 2 2 4 2 l 2 2 3 -1 3 -3 4 3 3 9 5 -3 -3 9 6 0 -1 0 7 -2 3 ;§_ A0 = 18 49 frequency of elicitation, were dark skin, curly (kinky) hair, musical, athletic, friendly, tall, uneducated, unintelligent, hard workers, and lazy. A new sample of subjects then evaluated each attribute on five evaluative semantic differential scales; the sum over the five scales provided a measure of e. To provide a measure of belief strength (b), subjects rated the probability that "Negroes have dark skin," "Negroes are uneducated," etc., on a set of five probability scales in a semantic differential format (e.g., probable- improbable, likely-unlikely); again a sum over the five scales was obtained. The e and b measures could both range from -15 to +15, with high scores indicating positive evaluation or high probability of association. An estimate of each subject's attitude toward Negroes was obtained by multiplying each e by the corresponding b and summing the products. Finally, each subject's attitude toward Negroes was assessed directly by asking him to rate the concept "NegroeS"n all other salient attributes. Here persons look for objects that are satisfactory on all salient attributes. Thus one may consider noncompensatory models to be "satisficing models." 62 These models have not attracted much attention in consumer research. The major types of noncompensatory models are: (l) conjunc- tive_models, which reflect extreme stress on the negative end of attribute scales; brands are rated high only if minimum levels on all attributes are exceeded; (2) disjunctive models which place stress on the positive pole; brands are rated high only when seen as superior on one or more relevant attributes; and (3) lexicographic models which posit sequential attention to the order of salience of attributes; preference for a brand is established through consideration of only the most salient attribute. Equivalence of two or more brands on this dimension introduces them to the next most salient attribute and so on until the choice is clear. Wilkie and Pessemier (1973) report several studies (e.g., Heeler, Kearney, and Mehaffey, 1973) that have compared these alterna- tives with the basic model, with conflicting results. The distinction between models of information processing and models.cfl’ processed information is an important one, since post- decision cognitive structure might vary from that utilized in active decision making as a function of time pressure, risk variables, cog- nitive consistency operators, and expreimental factors. Fishbein's basic linear compensatory model was developed as a static approach for describing an existing attitudinal structure. Most marketing studies have adopted this perspective of the model. The difference in purpose between this model and dynamic utility or decision models suggests that results from the basic model must be carefully evaluated before translations to information processing tasks are made. STATEMENT OF THE PROBLEM In this review, we have seen that there are certain diffi- culties with the model. However, as an attitude measurement model, Fishbein's approach has received empirical and psychological valida- tion. It seems to be an appropriate solution to the lack of conceptualization of attitude in mass media research. It allows one to derive a single measure of overall affect as well as examine the individual beliefs that comprise an attitude. One difficulty, however, remains. We have reviewed research that showed a lack of attitudinal homogenity about subjects of survey research and demonstrated a need for oeprant multivariate typing of persons. Wilkie and Pessemier (l973) echo this position, ". . . finding groups that are homogenous enough for cross-sectional analyses may be as important as improving the ways in which individual behavior may be more accurately modeled" (p. 438). A plausible solution to this problem would be operant multi- variate typing using the "e" score of the Fishbein model. This approach is, in essence, the same as attitude segmentation. Fishbein defines attitudes as a procedure which locates an individual along a dimension of favorable-unfavorable. The "e" score represents such a measure, thus we have attitude segmentation. The use of attitude to segment is based on the theoretical assumption that persons consume to satisfy their needs and wants, an 63 64 idea prevalent in economic theory for over 200 years. Since attitudes reflect needs and wants, finding homogenous groups means finding groups of individuals with the same attitude. Having identified homogenous groups, the Fishbein model can be further used in the study of the content of the attitude, the beliefs and evaluations that provide its structure. HYPOTHESES Thus far, the need for a general conceptual and operational definition of attitude in mass media research has been established. The Fishbein expectancy-value model was offered as a possible solution, provided the model was used only in conjunction with operant multi- variate typing and operations which permit the study of attitude content. The purpose of this dissertation was to design and test a method which would meet these criteria. Operational procedures were established for (1) operant multivariate typing; (2) the construction of factor statement arrays to permit the study of attitude content; and (3) the identification of consensus and discriminating items. Qperant Multivariate Typing As Wilkie and Pessemier (1973) pointed out, finding homogenous groups is perhaps the most important step in using an expectancy-value model in survey research. This study used e scores for multivariate typing of respondents into homogenous groups. This position is based on the theoretical assumption that persons consume to satisfy needs and wants. Since attitudes reflect needs and wants, finding homogenous groups means finding groups of individuals with the same attitude. Furthermore, since attitude is a set of beliefs, typing by attitude means typing along multiple beliefs. Thus, we have multivariate typing. 65 66 Operationally, the e scores in the Fishbein model were 0- factor analyzed to identify attitudinally homogenous groups, which will be referred to as attitude segments. Three different attitude segments were discovered. A discussion of this procedure is presented in the methodology section. Since group membership was determined by the correlation of persons' scores rather than by some a priori classification established by the researcher without regard for attitude scores, the Q-factor analysis resulted in natural, or operant groups. If multivariate typing results in homogenous segments, we should find that persons belonging to a segment have the same or very similar scores on each b and e item. There should be little deviance from the group's mean. If this is the case, attitude segments should have less variance about their b and e scores than a typical random sample or the traditional a priori behavioral segments. To test this proposition, a comparison of variances was undertaken. However, there would be no need to compare variances if the Fishbein model did not provide significant correlations between A0 and Zbiei for each attitude segment. Therefore, model performance was tested first. ModelgPerformance To test the importance of operant multivariate typing by attitude, it was necessary to compare the performance of the Fishbein model for attitude segments with the performance of the model for non- attitude segments. Traditionally, performance is operationalized 67 as the correlation between the measure of overall affect and the sum of the b - e scores. Before attitude segmentation can be used or before the model can be accepted, attitude segments must yield significant correlations. If the groups are homogenous, attitude segments should have higher correlations than non-attitude segments. Correlations for the attitude segments were compared to correlations for behavioral segments and a random sample group. The random sample of individuals was selected from the original sample of respondents. This sample represents typical survey research procedures where no attempt is made to type people into homogenous groups. Usually in these cases, the researcher assumes homogeneity. Other groups were "user" segments. Usage represents the traditional segmentation in marketing and advertising research. Under this approach, the researcher classifies individuals on the basis of their usage of the object under study. Thus there are heavy users, light users, and non-users (e.g., Twedt, 1967; Tuck, 1973) and for newspapers, there are frequent readers, infrequent readers, and nonreaders (News- paper Advertising Bureau, 1978). The assumption behind this approach is that homogeneity is indicated by similar behavior. If persons behave in the same manner for different reasons, this assumption may be tenuous. Since the attitude object for this study was a newspaper, the non-attitude groupings were Heavy Readers, Medium Readers, and Light Readers. 68 Hypothesis 1 It was hypothesized that the Fishbein b - e scores would be more correlated with the dependent variable, overall affect, when the model was used with the attitudinally homogenous groups than for groups assumed to be homogenous or for groups assumed homogenous because they have similar behaviors. However, before this hypothesis was tested, the performance of the model for each group was tested. It was hypothesized that the correlation between overall affect and the summed b . e scores for each group, the attitude segments, the random sample group, and the reader groups, would be significantly different from zero. Hypothesis 1: There is a significant linear n relationship between A0 and 2 biei for each 1—1 of the seven test groups. 1. Statement of Hypothesis: H0: r=0 H o 1. r f O 2. Find critical value from t distribution for a = .05, degrees of freedom = n - 2. 3. Test Statistic: t = r/(n - 2)/(1 - r2) 69 4. Decision Rule: If t greater than critical value, reject H0; If t less than critical value, do not reject H0. Hypothesis 2 Hypothesis 2: The correlation between A0 and n 2 biei for each attitude segment will be i=1 significantly different from the correlation for the random sample and from the correlation for each reader segment. 1. Statement of Hypothesis: where: ra = correlation for each attitude segment r1 = correlation for the random sample and for each reader segment 2. Find critical value from t distribution for a = .05, degrees of freedom = n - 3. 70 Test Statistic: Jnlé 3 (1 + rxz) t: /2(2(rxy . Fzy' rxz) + 1 - (rzm + rzy + r2 z) where: rXy = correlation of A0 with Zbiei for an attitude segment rZy = correlation of A0 with Zbiei for random sample or reader segment rxz = correlation of Zbiei for attitude segment with zbiei for random sample or reader segment Since persons included in Q-factors could also be members of the random sample and/or reader segments. the r's were treated as related, or dependent. Thus, a test for differences between dependent r was used rather than the more familiar r to z transfor- mation and Zutest (Bruning & Kintz, 1968, p. 191). Decision Rule: If t greater than critical value, reject Ho. If t less than critical value, do not reject H0. 71 Hypothesis 3 Kerlinger (1973) wrote, "(The) main technical function of research design is to control variance" (p. 306) and offered a possible solution: ". . . choose subjects so that they are as homogenous as possible" (p. 309). Operant multivariate typing may serve as a means of controlling variance in mass media survey research. However, before one can draw that conclusion, evidence must be offered in support of the premise. This evidence should be a measure of the variability within the groups. If a group is homo- genous, it should have less variance. Ifaigroup is not homogenous, it should have more variance. If operant multivariate typing yields more homogenous groups than the typical random sample or the tradi- tional reader segments, the attitude segments should have less variance about the e and b items. However, the first question is, "Is there a difference between the average variances of the b and e items for the seven groups?" Hypothesis 3: The variances about the b and e items among attitude segments, the random sample, and the reader segments are significantly different. 1. Statement of Hypothesis: H0: u] = p2 = W3 = p4 = U5 = U5 = My H]: p] f p2 # W3 # p4 f U5 f p5 f p7 72 where: u = the average variance about the 37 b and the 37 e items for each of the groups 2. Find critical value: ' ' “Fm-oxa-um-u 3. Test Statistics: F = MSB/MSRES 4. Decision Rule: If F is greater than the critical value, reject H0 and proceed with Scheffé and Tukey contrasts Since a significant F tells us that all group means are not identical and since we cannot determine the location or the magnitude of the difference on the basis of the F alone, the Scheffé method of multiple contrasts and the Tukey method of simple contrasts were used to determine which differences were contributing to the significant F. The following hypotheses were tested with the Scheffé method after a significant overall F was found. Hypothesis 3A: The average variance for the b and e scores of the attitude groups is significantly less than the average variance for the b and e scores of the reader segments. 73 1. Statement of Hypothesis: _ UF1 + uF11 + uF111 = 11H + “M + “L Ho ' 3 3 = uF1 + “F11 + “F111 1'H + “M + “L ”1 3 < 3 2. Find critical value: “(J ' nl-aFJ-1,N-J 3. Test Statistic: CA W 4. Decision Rule: If test statistic is greater than the critical value, reject H0. Accept H1, if D has a negative value. Hypothesis 3B: The average variance for the b and e scores of the attitude groups is significantly less than the average variance for the b and e scores of the random sample. 1. Statement of Hypothesis: H = PFI + PFII + “FIII _ o 3 ' “Rs 74 _ “F1 + “F11 i “F111 < u “1 ' 3 RS 2. Find critical value: " (J 'Th—fa-i ,N-J 3. Test Statistic: 4. Decision Rule: If test statistic is greater than the critical value, reject H0. Accept H] if D has a negative value. Hypothesis 3C: The average variance for the b and e scores of the reader segments is significantly less than the average variance for the b and e scores of the random sample. 1. Statement of Hypothesis: pp + PM + UL _ “o = 3 ‘ “Rs “H + PM + “L 3 < “Rs 2. Find critical value: ”J - T)]‘?\J'19N‘J 75 3. Test Statistic: 1.1L 8 W 4. Decision Rule: If test statistic is greater than the critical value, reject H0. Accept H0 if D has a negative value. The Tukey method was used to contrast each group's mean to each of the other groups' mean after a significant overall F was found. The procedure tells us which of the means differ significantly. Hypothesis 3D: The average variance of the b and e scores for the individual atttitude groups will be significantly less than the average variance of the b and e scores for the individual reader segments and the random sample. 1. Statement of Hypothesis: H0 = ”Fl = URj I —I l ‘ “F1 < “Rj where: uFi = average variance of the b and e scores for each attitude segment 76 pR = average variance of the b and e 1' scores for each reader segments and the random sample 2. Find critical value: 1-an,J(N-l) 3. Test Statistic: Ix.j ' X 1*I/VMSRES'" 4. Decision Rule: If test statistic is greater than critical value, reject H0. Accept H1 if X j"x.j* has a negative value. METHODOLOGY To test these hypotheses, several methods were necessary. These methods, the selection of subjects, and attitude object will be discussed in this section. Sample Subjects for this study were 100 Michigan State University students enrolled in an introduction to advertising class during spring quarter, 1978. Questionnaires were administered to 237 student in* the class; then 100 completed questionnaires were randomly selected for analysis. Random selection was accomplished with a table of random numbers. Students were selected as subjects because they were accessible at low cost. Random selection of subjects was important only to statistical assumptions, not to generalization. The purpose of this study was to test a method, not to generalize from a sample to a population. Only 100 subjects were selected because an absolute maximum of 100 variables can be submitted to the Statistical Package for the Social Sciences (SPSS) (Nie, Hull, Jenkins, Steinbrenner, and Bent, 1970, p. 490), which was used for the statistical analyses in this study. 77 78 Attitude Object The attitude object for this study was the Michigan State University student newspaper, the State News. This object was selected with the following criteria: 1. The object was familiar to subjects being sampled Had multiple attributes Had heterogenous utility, and hOON Was. related’to the mass media The State News is a mass medium circulated throughout the Michi- gan State University campus. Copies are available in at least one location in all dormitories and in most class room buildings. There is no charge for the newspaper other than a voluntary contribution requested during each quarter's registration. The newspaper is published five days a week with a daily circulation of approximately 40,000. Any student enrolled and in attendance at Michigan State's main campus can be expected to be familiar with the State News. Like other newspapers, the State News is decomposable into multiple attributes. Researchers, practitioners, and readers have long recognized the multiattribute nature of newspapers. These attri— butes include national and international news, local news, retail advertising, classified advertising, and sports. Empirically, Swanson (1955) defined 130 categories of newspaper content. More recently, Mauro and Weaver (1977) used R-factor analysis to decompose 244 items of content into four major factors: (1) community, national, and world news; (2) sports; (3) personal entertainment; and (4) general service items and business listings. 79 Theoretically, newspapers also have heterogenous utility. Such heterogeneity of attribute evaluations exists among objects for which there is individuality of choice. Stephenson (1967) calls this "convergent selectivity." Stephenson uses "convergent selectivity" to describe behavior which is voluntary, where more or less acceptable differences in taste or minor matters of opinion exist. It exists when an individual has the opportunity to exist for himself. Moral and ethical sanctions are largely bypassed. According to Stephenson (1967), the mass media, plays, art, and the theatre offer opportunities for convergent selectivity. Thus, the State News, a mass medium, can be considered to have heterogenous utility, and a suitable attitude object for this study. Questionnaire The questionnaire was developed according to Fishbein's pro- cedures (Fishbein and Ajzen, 1975, pp. 218-219) with a slight variation. Theoretically, the questionnaire had to present each subject with a standard set of belief statements about the attitude object and request the subject to indicate how strongly he believed each of the statements. In an effort to obtain a standard set of belief statements that would be as salient as possible for the sample under consideration, an independent sample of twenty Michigan State University students were asked in a personal interview, prior to development of the actual questionnaire, what characteristics of the newspaper they thought were important. These "open-ended" responses were tabulated and a sample of beliefs were selected for the questionnaire. 80 To represent the beliefs of the students, the independent sample was judgmentally selected so that all key beliefs had a chance to be represented. The sample of students included: 1. 7 males and 13 females 2. 5 freshmen, 4 sophomores, 6 juniors, and 5 seniors 3. 4 lB-year-olds, 4 l9-year-olds, 5 20-year-olds, 5 21-year-olds, l 22-year old, and 1 23-year-old 4. All different majors All classifications, ages, sex, and majors were represented because the introduction to advertising class serves all majors and typical enrollment includes all classifications, sex, and majors. The interview guide, presented in Appendix A, differs slightly from Fishbein's approach. The difference is that the interview guide does not solicit beliefs by only asking, "What about the State News is important to you?" as Fishbein's approach suggests. Rather, the interview guides focuses on a segment of behavior in which the object is normally encountered (Merton, Fiske, and Kendall, 1956). This permits the respondent to describe characteristics that are salient during normal experience with the attitude object. The beliefs arise naturally, without prompting, as desired by Fishbein. In addition, when respondents failed to mention any attribute, the interview guide asks the respondent about the item specifically. This procedure recognizes that a person's attitude toward an object is based on mul- tiple attributes and that the person's evaluation of each attribute is, in turn, based on multiple attributes of that attribute. By focusing 81 on the natural situation in which the object is encountered and by focusing on each attribute, the interview guide should produce a qualitative assessment of an individual's beliefs. While the procedure differs slightly from Fishbein's, the results should not influence the performance of the model. Fishbein clearly states that it is impossible, regardless of technique, to determine at which point a person starts eliciting salient beliefs (Fishbein and Ajzen, 1975). Given a set of pre-selected beliefs, the theory asks for a measure of (1) the evaluative aspect of each belief and (2) the strength of each respondent's belief that the object is associated with each attribute. The heart of the questionnaire, then, consisted of 2 items for each of a series of 47 attributes. The independent variables were the evaluative and the belief scores. The dependent variable was the evaluation of the State News: Dependent Variable: State News Like Dislike Independent Variable, Evaluative (e) score: A newspaper that provides me complete coverage of national and international news. Good Bad Independent Variable, Belief (b) score: The State News provides me complete coverage of national and international news. Probable Improbable 82 The 47 attributes were selected under the premise that all salient beliefs should be represented. While Fishbein selects only the most frequently occurring (modal) beliefs, such a sample may not be representative of the population beliefs. This study sought to represent all beliefs because deletion of a key belief, although not frequently mentioned, may create a sample of beliefs not truly representative. Once the questionnaire was constructed, it was pretested among a group of 50 students. This pretest revealed two problems which needed remedy. First, the questionnaire was fatiguing. Comple- tion of 94 items (47 - 2) tired the subjects. Second, several of the statements were considered by the students to be duplicates. As a result, ten items were deleted from the final questionnaire. This deletion reduced the number of items by 20 (10 - 2), an amount that was considered to be sufficient to solve the fatigue problem. The questionnaire consisted of three parts. The first section included demographic and behavioral questions and asked students to provide their overall attitude toward six newspapers, including the State News. The second section contained the evaluative (e) items and the last section contained the belief (b) items. The final questionnaire is presented in Appendix B. Data Collection During spring quarter 1978, the questionnaire was administered to an introduction to advertisting class. Since the instrument used a 83 semantic differential scale, instructions for using the scale prepared by Osgood, Tannenbaum, and Suci (1957, pp. 82-84) were utilized. Two-hundred thirty-seven questionnaires were returned, completed and usable. From this population of completed questionnaires, 100 were randomly selected. These questionnaires were coded and the data were keypunched onto computer cards for analysis. Level of Measurement It is generally assumed for the tests used in this study that the variables are measured at a level that permits parametric statistics. However, the typical criticism of most scales including the semantic differential used in this study, is that they measure at the ordinal level. While it could be argued that they are interval scales, it is better to base the decision of whether to use parametric statistics or non-parametric statistics on existing evidence. The evidence suggests using parametric statistics. Based on studies by Labovitz (1967, 1970) and Baker, Hardyck, and Petrinovich (1966), Bohrnstedt and Carter (1971) conclude, ". . . under almost any conceivable research situation, statistical tests are robust enough to allow the researcher to use them with little fear of gross errors regardless of whether or not he has an interval or ratio scale so long as his ordinal measure is monotonically related to the underlying true scale" (p. 131). In this study, this monotonic rela- tionship can be assumed and parametric statistics are therefore used. 84 QeFactor Analysis Given the sample of 100 randomly selected respondents, the first step was to determine the number of attitude segments, or Q-factors, present in the data. Likewise, it was necessary to es- tablish the behavior segments and the sample segment. To identify the attitude segments, or factors, present in the data, Q-factor analysis was used. The basic input for Q-factor analy- sis is the correlation matrix of correlations between subjects. R-factor analysis, the more commonly used method, begins with a matrix of correlations between variables. Thus Q-factor analysis classifies respondents into factors or groups with similar profiles of responses. To facilitate the distinction, it is best to remember that in 0, variables refer to respondents and in R, variables refer to variables. Since the variables are the respondents, Q-factor analysis shows patterns of relationships among respondents. Within a factor are respondents who provided responses that correspond to some degree (a tendency to evaluate attributes in a similar way), thus indicating some similarity in attitude (Schlinger, 1969, p. 56). Q-factor analysis was preferred to cluster analysis because: First, cluster analyses are ". . . mathematically 'messy'" (Nunnally, 1967, p. 365). Numerous semiarbitrary decisions must be made about the number of clusters and the composition of clusters. Second, ". . . these methods require as much computational time as a formal factor analysis would" (Nunnally, 1967, p. 365). 85 Having decided to use Q-factor analysis, the next step was to decide whether to use principal components analysis or classical factor analysis. Principal components analysis is ". . . a relatively straightforward method of transforming a given set of variables into a new set of composite variables or principal components that are orthogonal (uncorrelated) to each other" (Nie et al., 1970, p. 470). No particular assumption about the underlying structure of the variables is necessary. It simply asks what would be the best linear combination of variables. The first principal component is the best linear combination; the second principal component is the second best linear combination; and so on (Nie et al., 1970). The model can be expressed mathematically as: Zj = ajF] + asz + o o . + ajnFn where each of the n observed variables is described linearly in terms of n new uncorrelated components F1, F2, . . . , Fn’ each of which is a linear combination of the n original variables (Nie et al., 1970, p. 470). Classical-factor analysis assumes that the observed correla- tions are the results of some underlying regularity in the data. "This regularity is attributed to various determinants, some of which are shared by other variables in the set while others are not shared by any other variable" (Nie et al., 1970, p. 471). The part of a variable that is influenced by the shared determinants is called common. The major assumption is that the assumed determinants will account for all the observed relations in the data and be fewer in 86 number than the variables. The part of the variable not shared with others and influenced by idiosyncratic determinants is called unique (Nie et al., 1970, p. 470). The model can be expressed as: zj = aleli-aj2F2+-.. .i-aijmi-dej where: j = 1, 2, . . . , n z- = variable j in standardized form F1 = hypothetical factors Uj = unique factor for variable j a.. = standardized multiple-regression coefficient of variable j on factor i (factor loadings) d. = standardized regression coefficient of variable j on unique factor j The following correlations are assumed to hold among the hypothesized variables: r = O (Fing) where: i = 1, 2, . . . , n J =19 2, 9 n 1'” and r O (uj.uk) where: j f k 87 The unique factor Uj is assumed to be orthogonal to all the common factors and the unique factors of other variables. The unique portion of a variable is not related to any other variable or that part of itself which is due to the common factor. "Therefore, if there is any correlation between two variables j and k, it is assumed to be due to the common factors" (Nie et al., 1970, p. 471). Theoretically, the classical-factor analysis was the best suited for this study. That is, there is an assumed regularity in the data that can be explained by a common set of beliefs, or attitude. Individuals with similar attitudes, as indicated by their responses to the e scales, will load on a common factor or attitude. The variance of an individual unexplained by the analysis is due to the uniqueness of that individual. Thus, individuals who share beliefs will load on the same factor and that part of the individual not accounted for by the analysis is the uniqueness of the individual. This uniqueness can be attributed to the beliefs: the individual's beliefs were not all represented and/or there are parts of an individual which are idio- syncratic and can be attributed to his unique life experiences. In execution, the difference between these two methods is whether the diagonal of the correlation matrix is altered before performing the factor analysis. Classical analysis replaces the main diagonals of the correlation matrix with communality estimates before factoring. Communality refers to the amount of variance of a variable that is shared by at least one other variable in the set. It is 88 expressed as hz. The portion of a variable that is unexplained represents the unique variance and is expressed as 1 - h2 (Nie et al., 1970, p. 476). Communalities are calculated by squaring the fac- tor loadings on each factor in a solution and summing across factors: 2 = 2 2 2 hi ai] + ai2 + . . . + ain where: i = l, 2, . . . , i n=1,2, ,n Since classical—factor analysis replaces the diagonal of the correlation matrix with the communalities before the solution is determined, estimates of the communalities must be used. In SPSS, initial estimates of the communalities are given by the squared multiple correlation between a given variable and the rest of the variables in the matrix (Nie et al., 1970, p. 480). When the deter- minant of the matrix is smaller than 10'8 or the matrix is singular, the absolute value of the largest element in each column is used instead of the squared multiple correlation. In addition, SPSS provides an iterative procedure with the PA2 (classical-factor analysis) mode for improving estimates of communal- ity (Nie et al., 1970, p. 480): First, the program determines the number of factors to be extracted from the original or unreduced correla- tion matrix. The program then replaces the main diagonal elements of ths correlation matrix with estimates of communalities, the R estimates. Next, it extracts the same number of factors from the reduced matrix, and the variances accounted for by these factors become new communality estimates. The diagonal elements are then 89 replaced with these new communalities. This process continues until the differences between the two suc- cessive communality estimates are negligible. This procedure staps if two successive sets of communality estimates differ from each other by not more than 0.001 (Nie et al., 1970, p. 493). A third consideration in factor analysis is the method of rotation. Rotation seeks to simplify the initial-factor matrix. Since the study seeks to identify different, or independent groups of persons who represent different attitudes, a method which results in independent factors and maximum loadings on a factor was needed. VARIMAX rotation offered such a solution. VARIMAX centers on simplifying the columns of a factor matrix: it maximizes the loadings of a factor. It defines a simple factor as one with "only 1'5 or 0's in the columns" (Nie et al., 1970, p. 485). Having selected the type of rotation to use, the next decision was how many factors are in the data. This is a problem with factor analysis because the exact configuration of any factor solution is not unique; one factor solution can be transformed into another without violating the basic assumption or mathematical pr0perties of a given solution (Nie et al., 1970). Therefore, it is the researcher's solution that satisfies the theoretical and practical needs of the research problem. Most researchers use some rule of thumb for deciding the number of factors. There is no agreed-upon formula. 90 In order to interpret a table of factor loadings, a researcher must do two things: (1) deciderthe cut-off for significant and insignificant factor loadings, and (2) decide which factors are "real" and significant (Schlinger, 1969). Factor loadings, which indicate the relationship of each respondent to each factor, are keys to determining the number of factors. Factor loadings are like correlation coefficients in that they range from +1 to -1. The higher the loading (either + or -) the more representative that respondent is (either + or -) of that particular factor. Significance of factor loadings was determined using a criterion suggested by Stephenson and Damby (cited in Schlinger, 1969). The criterion was 2.58(l//N) where N is the number of items. This means that a factor loading was considered significant if it was as large as 2.58 times the standard error for a zero-order correlation. The 2.58 value is equivalent value of z for a .01 level of significance. Significance of factors was determined using Humphrey's Rule (Fruchter, 1954): 1. Find the product of the two highest significant loadings on a factor 2. Find the standard error of a zero order correlation coefficient at .05 level of significance (2(1//N)) 3. If the product found in Step 1 does not exceed the value found in Step 2, the factor is not con- sidered significant 91 There is an additional problem in determining significant factor loadings. Sometimes respondents have significant loadings on more than one factor. Such a person is said to have confounded loadings. These respondents represent persons for whom there are not enough similar persons in the sample to form a factor. There is no agreed upon formula for handling confounded loadings, and the following procedure was developed for this study: 1. Square the two highest significant loadings of a subject Divide the largest squared factor loading by the smaller If the result of Step 2 is 1.5 or greater, consider the larger loading the more significant loading. The value of 1.5 means that the larger loading accounts for one and a half as much variance as the smaller. This is similar to an F test whiCh divides one variance by another: where 111 = 50, 112 = 50, F = 1.50 is significant at the .75 level of significance If the result of Step 2 is less than 1.5, the respond- ent is considered confounded, not loaded on any factor, and deleted from further analysis With these procedures the number of significant loadings and the number of factors were determined. However, if there were less than five significant loadings on a factor, the factor was not considered significant. The number 5 is peculiar to this study because sufficient 92 numbers of subjects were needed to accomplish the statistical analysis planned. Five is consistent with a minimum cell size of 5 for a chi-square analysis, without a correction factor (Blalock, 1972, p. 285). The search for significant loadings and factors was accom- plished on the initial factor matrix. Evidence for three factors was found. A VARIMAX rotation was then executed with the specified three factors. The rotated factor matrix is presented in Appendix C. QeFactor Solution The three factor solution accounted for 47.4 percent of total variation as follows: Eggtgn Variance .lL FI 34.3 38 FII 7.8 18 F111 __5_._3 _1__4 47.4 70 Eighteen of the respondents were confounded and twelve were not loaded.~ The confounded respondents were loaded accordingly: Confounded on: _n_ F I and FII 11 F1 and FIII 3 FII and F111 3 F1, FII, and FIII 93 To maintain equal cell size, the number of subjects to be used in further analysis was limited to the number of subjects significantly loaded on the smallest factor. Since FII (n=14) was the smallest, the 14 subjects with the highest loadings on FI and FII were selected for further analysis. All 14 from FIII were used. Reader Segments and the Random Sample Having identified the factors, or segments, in the data and the necessary cell size, it was necessary to select user segments for comparisons. The user segments were decided a priori and con- sistent with other Fishbein studies and marketing studies to be Heavy Readers, Medium Readers and Light Readers. Heavy Readers were operationalized as respondents who reported reading the State News five days a week and reading 80-100% of the content. Medium Readers were defined as respondents who reported reading the newspaper five days a week and 50-79% of the content. Light Readers were defined as respondents who reported reading the newspaper 1-3 days a week and 50% or less of the content. As n was limited to 14, the 14 subjects who maximized the reading criteria for each of the three categories were selected for analysis. In addition, a simple random sample of 14 was selected from the sample of 100 using a table of random digits. At this point, there were seven segments of 14 persons each: (1) 3 Q-fact-ors; (2)3 reader segments; and (3) 1 simple random sample. 94 Since the respondents were selected without regard for which factor they belonged to, some persons in the reader segments and the simple random sample were also significantly loaded on factors. The composition of each group is outlined in Table 3. Hypothesis 1: Significance of Pearson Product-Moment Correlation Coefficients To compute the model, the score of each b item is multiplied by the corresponding e item and then summed across all items for each person. The sums for all persons are then correlated with the dependent variable, the person's evaluation score for the State News on the 7-point Like-Dislike scale. This study, like most others, used the Pearson product-moment correlation coefficient to determine the relationship between A0 (Like- Dislike) and Zbiei. The product-moment correlation coefficient was preferred to regression because the researcher had no real control over either variable (Hays, 1973). The assumptions for the Pearson product-moment correlation coefficient (Nunnally, 1967, p. 126, and Glass and Stanley, 1970, p. 106) are: 1. Linear relationship between variables 2. Normal distributions 3. Homoscedasticity The Fishbein model and measurement are generally assumed to fit these assumptions. 95 Table 3. Groups with Matched Subjects FI FII FIII __Group Loadings Loadings Loadings Heavy (n=3) .65 (n=3) .41 (n=3) .51_ Readers .48 .46 .50 .43 .50 .58 Medium (n 6) .56 (n=2) .66 (n=l) .66 Readers .72 .54 .43 .53 .47 .65 Light (n=5) .79 (n 2) .49 (n=2) .63 Readers .69 .64 .48 .46 .43 .71 Random (n 5) .71 (n 3) 77 (n=3) .77 Sample .43 .65 .65 .67 .61 .61 .69 .79 96 The linear relationship is proposed by the model and theoretically sound. The model proposes that as the Zbiei items increase, the more favorable the person's attitude. A straight line would seem to do well in describing this relationship. Likewise, most previous research has tested for and found a linear relationship between A0 and Zbiei. Nunnally (1967) simply states that unless there is a marked curve in the relationship the linear measure gives much the same results as does a curvilinear measure. Heteroscedasticity will reduce the amount of information provided by the product-moment correlation coefficient. It will not reveal, in the case of heteroscedasticity, that the relationship was stronger at certain levels of the variables and weaker at others. . Generally, the product-moment correlation is robust, that is, not drastically influenced by violations of the assumptions. As Nunnally (1967) wrote, "Unless one of the assumptions were seriously violated, inferential statistics would not_be highly erroneous" (p. 126). The SPSS (Nie et al., 1970) was used to compute the Pearson product-moment correlation coefficients and tests for significance. The SPSS uses the following formula (Nie et al., 1970, p. 280): N N N 1 Z XiYi - 2 xi 12 Yi/N) N 12 1/2 . - 2 Y. /N» i=1 '1 ]} 97 2. Test of Significance: t = r/(n-2)/(1-r2) df = n - 2 3. A t value larger than I2.16 is significant at the .05 level using a two-tailed test. Hypothesis 2: Tests for Differences between Correlation Coefficients After testing for model performance among all segments, the differences between r's was tested to satisfy the second hypothesis. Since the Random Sample segment and the reader segments included persons from the Q-factors, the correlations were not inde- pendent. To compare related correlations, the following formula was used to test for difference between dependent correlations (Bruning and Kintz, 1968, p. 193). 1 t _ /h-3 (l-Frxg) f2(2(rxy . rzy . rxz) +1 .. (r§y+ rgy'I' rgz) 2. df = n - 3 3. A t larger than 2.201 with 11 df is significant at the .05 level using a two-tailed test. A t larger than 1.796 with 11 df is significant at the .05 level using a one-tailed test. 98 Hypothesis 3: Differences between Ayerage Variances The third hypothesis questions the amount of variance about the b and e items. Does Q-factor analysis yield groups whose b and e scores are more homogenous than the typical random sample or the traditional behavioral segments? Since there are a set of b items and a set of e items, there are two question: (1) Is the average amount of variance about the b items less for the Q-factors than for the random sample and less than the reader segments? (2) Is the average amount of variance about the e items less for the Q-factors than for the random sample and less than the reader segments? Since the question involves 7 different means, analysis of variance is suggested. However, this analysis of variance design is slightly different, but most appropriate, from the more common ANOVA designs. The difference is that this design compares scores for items rather than subjects. There were 37 items compared across 7 different, but not independent groups rather than 37 subjects compared across 7 different treatments. This design had to account for the same elements being measured under each of the 7 treatments. It had to account for correlated observations. A repeated measures design (Winer, 1971, p. 261) best met these conditions. The design is portrayed in Table 4. The variance for each evaluative (e) item and each belief (b) item was compared across all groups. Thus item 1 had variances of .22, 1.912, 1.258, 2.571, 1.67, .885 and .951 under the respective 99 1M l_l._.u ... ... ... ... HHmL-u HmF CGQE w I_.—- ... ... ... ... HHKF Hmh PMHOH ca ca mcx ... ... ... ... HHucx Hmcx : wm mm ... ... ... ... ... HHewx Hawk N we _a ... ... ... ... ... HHmrM Harm F cam: Papoh pgmwe anuwz x>mmz mnmEmm HHH HH H EmpH mcmuwmm emvcmm mcouomu Aucmspmmchv asoco :mwmmo mmczmwmz emummamm .w mpnwp 100 segments. The total of these scores is 9.467; thus the numerical value of P] is 9.467. The other values of P's are obtained by summing the entries in their respective rows. The numerical values for the T's are obtained by summing the columns. For example, T1, is the sum of the variances for the 37 e items under F1. The grand total, G, is obtained either by summing the P's or by summing the T's. The F ratio MStreatment MSresidual is used in testing hypotheses about differences among the variances. The analysis of variance table for the sources of variation that exist appears as Table 5. For a .05 level test of the hypothesis that u] = p2 = u3 = u4 = p5 = p6 = p7, the critical value for the F ratio is F.95(6,216) = 2.09. The assumptions (McSweeny, 1976) for the model are the same as for the Two-Way Fixed-Effects ANOVA plus equal pairwise correlations for the different levels of the fixed independent variable: 1. Normal distribution Equality of variance Independence boom Equal or proportional cell sizes 5. Homogenous correlations: In all pairs or levels of the fixed factor, the correlatiOn (across the 101 1 1 1 a1 A_ aVAF Hv_ Ha _ Ha=_m> _mocpwcu m to mucaom F u.: saw: <>oz< muum$$m1umxwz sz1ozh mcu com opus» <>oz< .m mpnmh 102 population of random effects) of the scores under the other level of the pair must be the same" (Glass and Stanley, 1970, p. 465). While assumptions are sometimes readily made, one must consider what happens if they are violated. The assumption that observations are sampled for normal populations has received considerable attention. Citing Pearson (1931), Box and Anderson (1955) and Scheffé (1959), Glass and Stanley (1970) state, "With respect to the probability of a type I error, we can safely conclude that the ANOVA assumption of normality is of almost no importance" (p. 372). Regarding homoscedasticity, Glass and Stanley (1970) state, ". . . the effect of heterogenous variances on the level of signifi- cance in the F-test is negligible" (p. 372). This study has equal n's, so heteroscedasticity should be no problem. When heterogenous correlations are suspected special measures must be taken to assure the validity of the F-test of the null hypothesis about the fixed main effects. A contingent hypothesis testing procedure, suggested by Glass and Stanley (1970, p. 470, reprinted by permission of Prentice- Hall, Inc., Englewood Cliffs, New Jersey) was used: 1. If F = MS /MS% exceeds the l - a percentile point in the F- gistribution with l and J - 1 degrees of freedom, reject H0: 2a? = O at the a—level of sig- nificance. (The conservative test.) 2. If F = MSA/MS 8 falls below the l-a ercentile point in the -distribution with I - and (I - 1) (J - 1) degrees of freedom, do not reject H Za2= O at the a-level of significance. (TRe apparent test.) 103 3. If F = MSA/MSAB falls between] and F1’ 161 a one must resor multi- varigtl’ {Achhi gue)known as Hotelling's T2 test (see Scheffé, 1959, or Winer, T971). With these considerations and the design of this study, it was con- cluded that the repeated measures ANOVA was a robust test. Scheffé Tests In ANOVA when H0 is rejected, the researcher does not always know which difference between means was responsible for the signifi- cant difference. In such a case, multiple comparisons of means are necessary. Since the t-test is not appropriate (Glass and Stanley, 1970, p. 382) and Winer (1970, p. 270) states that the Scheffé method is acceptable for multiple comparisons with a repeated measures design, the Scheffé method was used to compare differences between means after a significant difference was found in the repeated measures ANOVA. The only difference between a repeated measure Scheffé test and a typical ANOVA is that Msresidual (interaction) takes on the value of MS for a repeated measures design (Winer, within 1970). Glass and Stanley's (1970, p. 390, reprinted by permission of Prentice-Hall, Englewood Cliffs, New Jersey) procedure for accomplish- ing a Scheffé test was used: I Step 1. Specify and estimate w. The coefficients c], . . . , cd that determine the contrast of interest must be specified. Once this is done, 3 is obtained by substituting sample means for population means: Suppose there are [seven] population means u], . . . , u7. Table [6] lists contrasts . . . 104 Table 6. Contrasts and Coefficients Values of Contrast c1 c2 c3 c4 c5 c6 c7 “F1+“F11"“F111 _ “n+“M+“L _1_ 1 1_ 0 __1_ _1 _1_ 3 3 3 3 3 3 3 3 “F1 +“F11““F111 1 1 1 3 ‘ “Rs ‘3' 3 3 '1 0 0 0 “Hi“MWL 1 1 _1_ 3 ' “Rs 0 0 0 ‘1 '3' '3 3 Step 2. Step 3. Step 4. Step 5. 105 [used in this analysis]. Any contrast m can be estimated by replaging u's with sample means, i.e., an estimate of w of ip is given by A _ ‘1) = C-lx..| +C27.2 + o o o +CJ-X-OJ For example; if m = p2 - u3, then an estimate of 11115 X.2 - X.3. Find an estimate of the variance of O. The one- factor ANOVA that precedes application of the S-method contains a MS . This value, the c's, and the sample sizes are substifiuted into the following formula to find 6“- 82:MS ii E—g—+Ooo :3. 1P W n1+n2 "J where MSw is the "mean-square within," cj is the constant that multiplies the jth mean, and _ nj is the number of observations in the jth group. Find 8 . Take the positive square root of 82, which as found in step 2. Form the ratio of O to 8A. The value of O obtained in step 1 is divided by the value of 89 found in step 3. Compare the absolute value of the ratio with the test statistic. The absolute value of the ratio found in step 4 is compared with the square root of (J — 1) times the 100(1 - a) percentile in the F-distribution with degrees of freedom J - l and N - J. That is, the absolute value of the ratio is compared with /TJ - l)1_aF _] N- . The hypothesis that p = O is rejegted if the absolute value of the ratio exceeds the square root of (J - 1) times the percentile point, i.e., A . _ -|_‘P_| FEJECt ”0‘ w ’ 0 ‘f a@ > “13"‘11-dFJ-1,Nsd 106 Tukey Test To conduct the Tukey contrasts, Glass and Stanley's procedures (1970, p. 383, reprinted by permission of Prentice-Hall, Englewood Cliffs, New Jersey) were used: Step 1. All J(J - l)[2 comparisons between sample means of the form X3 -‘X * are computed. For exa_p1e, if three treatments gre compared, then X 1 -X.2, X 1 - X 3, and X 2- X;3 are calculated. Step 2. All comparisons X j* are divided by JMSw7n, where MSw ls theJ mean- square within factor levels from the one- way ANOVA and n is the number of observations in any one group. Step 3. The 100(1 - a) percentile point in the Studentized range distribution with degrees of freedom J and J(n - l) is found. . . . This percentile point is denoted ]_ an J(n- 1). The Studentized range is the difference between the largest and the smallest means of J independent samples eachcf size n from a normal population, divided by 15 n. There is a family of distributions of the St dentized range, since a different distribution results for all pairs of values of J and n. The two parameters used to identify a particular Studentized range distribution are J, the number of samples, and J(n - l), the degrees of freedom for MSN . Step 4. All J J - l)/2 differences X'o divided by 45 n are compared with the percentile point. It 15 concluded that X - and X * are significantly different, i. e. , present evidenge that pj and uj* are different, if |X_j -X j*I over /Ms;7fi'is greater than 1_an J(n- l)' Limitations There are two major limitations of this study that need to be enumerated before the results are discussed. First is sample size. An n of 14 is easily criticized. However, this is more of a l07 mechanical than a theoretical problem. The SPSS limits input of variables to be Q-factored to l00 because of the large work space required. However, there are other programs available commercially which will handle a larger number of variables. Hells, for example (l979), uses a procedure which Q-factors over 2,000 persons. Such a procedure was not available for this study. The second problem is the number of items which can be used in the questionnaire. Since each item is measured twice, adding a few questions greatly increases the time required of the respondent to complete the questionnaire. Fatigue can then become a problem. An intuitive solution would be to reduce the number of items. However, reducing the number of items reduces the information obtained. A short questionnaire may not provoke fatigue, but it surely retrieves less data. Another problem with reducing the number of items is the effect of the reduction in the correlations between individuals to be used in the Q—factor analysis. A reduced number of items leads to less stable correlations and less stable factors. Some happy medium must be struck between fatigue and stable correlations. While it was beyond the scope of this study to test a solution to this problem, this was a problem faced in this study. RESULTS Hypothesis 1 The first hypothesis was tested with simple correlations between the independent variable (Zbiei) and the dependent variable (A0). Table 7 contains the results. Table 7. Pearson Product-Moment Correlations Group r pg_ F1 .63 .05 FII .50 .05 FIII .79 .05 Random Sample .58 .05 Heavy Readers .74 .05 Medium Readers .64 .05 Light Readers .52 .05 All of the groups had significant correlations at the .05 level of significance. The model works for the random sample and both types of segmentation. The three attitude segments had an average correla- tion of .64 compared to the reader segments' average of .63 and the random sample's .58. Since they were all significant, the null for Hypothesis 1 was rejected and the alternative accepted. 108 109 Hypothesis 2 The second hypothesis required a comparison of the correlation coefficients. A test for differences between dependent correlations was used to compare these coefficients. Table 8 contains the results of these comparisons. Since there were no significant differences the null for Hypothesis 2 was accepted. The Q-factors did not yield better model performance than the random sample or the reader segments. Table 8. Comparison of Correlation Coefficients Differences Random' Heavy Medium Light Sample Readers Readers Readers Factors diff.* sig.** diff, SigL diff. sig. diff. sig. F1 .05 --- -.l6 --- --0l --- .ll --- FII -.08 --- -.24~ --- -.l4 --- .02 --- FIII .2l --- .05 --- .l5 --- .27 --- *Difference was calculated by subtracting from the factor, e.g., rFI ' rRS = .05. **Critical value for .05 level of significance, one-tailed test = l.796. Hypothesis 3 The third hypothesis required a repeated measures design ANOVA to test for differences between average variance for all b and e items across groups. The average variances for the e items are presented in Table 9 and the averages for the b items appear in Table 10. 110 Table 9. Average Variances for e Items Group X Rank* FI 1.06 2 F1 I 1 . 1’ 4 FIII 1.49 5 RS 1.50 6 H 1.65 7 M .91 l L 1.47 4 *1 = least average variance, 7 = greatest-1: average var1ance. TablelO. Average Variances for b Items Group XV Rank* FI 1.69 2 FII 2.10 5 FIII 2.16 6 RS 1.74 3 H 2.00 4 M 1.55 1 L 2.56 7 *l = least average variance, 7 = greatest average variance. 111 Visual inspection shows that the attitude segments generally have less variance about the e items. They tend to have more variance about the b items. The analysis of variance for the e items, presented in Table ll, indicates the existence of significant differ- ences between the means. Table l2, which presents the ANOVA for the b items, also indicates significant differences. Table ll. Repeated Measures ANOVA for e Items Sum of Squares df MS F Significance Total 99.55 258 Items 25.3l 36 Groups 16.98 6 2.83 l0.68 .00l* Residual 57.26 2l6 .265 *Conservative test: F > .95 F1 6 = 5.99 Apparent test: F > .95 f6,216 = 2.09 Table 12. Repeated Measures ANOVA for b Items Sum of Squares df MS F Significance Total l90.3 258 Items 50.6 36 Groups 26.1 6 4.35 8.27 .00l* Residual ll3.6 2l6 .526 .99 *Conservative test: F > .95 F = 5 2.09 Apparent test: F > .95 F6,216’ ll2 Both the conservative test and apparent test yield signifi- cant differences for b and e items across groups. Therefore, the null for Hypothesis 3 was rejected. Hypotheses 3A--3B--3C To determine what means contributed to the significant differences, Scheffé tests were conducted. The results of the Scheffé tests for b and e items are presented in Tables l3 and 14, respec- tively. In neither set of contrasts are there significant differences. The attitude groups, as a whole, and the reader groups, as a whole, do not differ from each other or from the random sample for either the b or e items. Therefore the null hypotheses 3A, 38, and 3C were not rejected. Hypothesis 3D To determine how the individual group means differed, Hypothesis 30 was tested with multiple Tukey contrasts. Table 15 presents the Tukey contrasts for b items, and Table 16 presents the Tukey contrasts for e items. Table 15 indicates there are few differences between group variances for b items. In two of nine comparisons attitude segments have less variance than Reader Segments and in two others they have more variance. Likewise, there is no real difference between the Random Sample and the Reader Segments or the attitude segments. Thus, one can conclude that there is no difference between the groups' variances about the b items and accept that part of Hypothesis 3D. 113 Table 13. Scheffé Contrasts--b Items Contrast w W/Ofi Significance* u +u +u 11+u +u FI F1; FIII _ H 3M L _.05 .72 ns u +11 +11 F1 F1; FIII - “Rs .24 2.47 ns uH'HM‘UL 3 _ “RS .30 3.08 ns */6.95F6,252 = 3°55 Table l4. Scheffé Contrasts--e Items Contrast w $/8@ Significance* u +u +u 11+u,+u F1 F1; FIII _ H 3” L _.11 1.57 "S u -+u -+u F1 Flfi FIII - “Rs -.27 2.7 ns 11 +11 +11 H 3M L " HRS -.16 1.6 ns */6 F = 3.55 .95 6.252 114 Table 15. Tukey Contrasts--b Items Contrasts Difference Difference/(MSresln)1l2 Signi ficance/l evel UFI - UFII -.41 3.44 ns “FI _ “FIII -.47 3.94 significant (.10) “FI - “RS -.05 .42 ns uFI - UH -o3] 2.60 "S UFI‘UM +.14 1.17 "S “FI — “L -.87 7.30 significant (.01) “FII-“FIII -.06 .50 ns UFII " MRS +.36 3.02 115 um - 11H +.1o .84 ns “FII'UM +.55 4.61 significant (.01) “FII - “L -.46 3.86 significant (.10) “FIII " “RS +.42 3.52 ns “FIII "IJH +.16 1.34 ns “FIII-“M +.51 5.12 significant (.01) “FIII ‘ 11L +.40 3.36 "S “RS'UH -.26 2.18 ns HRS-“M +.19 1.59 "5 “RS - 11L -.82 5.88 significant (.01) “H - “M +.45 3.78 ns “H‘UL -.56 4.70 significant (.05) 11M- “L -l .01 8.47 significant (.01) 115 Table 16. Tukey Contrasts--e Items Contrasts Difference Difference/(MSres/n)V2 Significance/level “FI"UFII -.08 .95 ns uF1-uF111 -.43 5.08 significant (.01) HFI"HRS -.44 5.20 significant (.01) “FI"“H -.59 6.97 significant (.01) )JFI "UM +.16 1.89 ns “FI"“L -.41 4.85 significant (.01) UFII-UF111 -.35 4.14 significant (.10) “FII"“RS -.36 4.25 significant (.05) “FII"“H -.51 6.03 significant (.01) “FII ' 1.1M +.23 2.72 ns “FII - 11L -.33 3.90 significant (.10) “FIII"“RS -.01 .12 nS “FIII ‘IJH -.16 1.89 "S “FIII - 11M +.58 5.85 significant (.01) “FIII - 1.1L +.02 .24 n5 “RS"UH -.15 1.77 ns URS"UM +.59 6.97 significant (.01) HH"HM +.74 8.74 significant (.01) lfll'uL +.18 2.13 ns “M"“L -.56 6.62 significant (.01) 116 Table 16 indicates that the attitude segments generally have less variance about the e items. FI has less variance than the Random Sample group, the Heavy Readers, and the Light Readers. FII has less variance than the Random Sample, the Heavy Readers and the Light Readers. FIII, on the other hand, has variance significantly greater than Medium Readers and is not significantly different from the others.. In four of the nine comparisons between attitude segments and Reader segments, presented in Table 17, the attitude segments have significantly less variance. Table 17. Significant Differences between Attitude Segments and Reader Segments: e Items __— Reader Segments Attitude Segments Heavy Medium Light Random Sample F1 SIGLT* ns SIGLT SIGLT FII SIGLT ns SIGLT SIGLT FIII ns SIGGT** ns ns *SIGLT==significantly less than **SIGGT==significantly greater than Therefore, Hypothesis 3D is rejected for differences between e item variances for attitude segments and Reader segments. Two of three comparisons of attitude segments to the Random Sample are significantly different. FI and FII have average variances 117 less than the Random Sample. Therefore, Hypothesis 30 related to differences between the Random Sample and the attitude segments is rejected. A comparison of reader segments to the Random Sample shows that reader segments generally are not different from the Random Sample. In only one of three comparisons is there a significant difference. Therefore, Hypothesis 3D regarding differences between reader segments and the Random Sample is accepted. Generally, there is no signifi- cant difference. Discussion This study was designed to test a procedure for selecting homogenous subjects to be used in conjunction with the Fishbein attitude model in mass media attitude research. To accomplish this, multivariate typing, using Q—factor analysis of the correlations of subjects' e scores was used. A test of the Fishbein attitude model for each Q-factor type revealed that the model performed significantly. A comparison of r's between the Q-factor types and typical user segments and a random sample showed no significant improvement in the model with the Q-factors. A repeated measures ANOVA of the variance about the evaluation of attributes and the beliefs that the State News possessed the attri- butes revealed significant differences. A Scheffé test revealed these differences were not due to differences between the groups taken as a whole. Tukey contrasts of the variances about the beliefitems discovered 118 significant differences, but they were not attributable solely to the attitude segments. It was concluded that multivariate typing did not yield groups more homogenous in their beliefs. 0n the other hand, contrasts of the variances about the evaluative items revealed a number of significant differences attributable to the attitude segments. While the results were not unanimously in favor of attitude segments, it was concluded that multivariate typing yields groups more homogenous in their evaluation of attributes. While this conclusion may be questioned, it is based on the statistical tests of the hypotheses and other evidence provided by the analysis of the data. First, FI and FII have less variance than all other groups except Medium Readers. Why Medium Readers have less variance is unknown. Perhaps it is due to the fact that a large number of respondents classified as Medium Readers were significantly loaded on F1 as shown in Table 3, page 95. Almost half of the Medium Readers are significantly loaded on F1. Heavy and Light Readers had much more variance. Unlike Medium Readers, these two groups were more diverse in the number of persons classified as Heavy or Light Readers and also loaded significantly on a factor. The success of behavioral typing seems to depend on the number and strength of operant factors unknown to the researcher without multivariate typing. A researcher would never know before hand whther or not a behavioral segment would be more homogenous. 0n the other hand, FI and FII's homogeneity is the result of multivariate typing. The researcher would know beforehand the groups tend to be homogenous. The reason FIII does not yield significantly less variance 119 could be attributed to the fact that it is the third factor extracted. from the correlation matrix. As more and more factors are extracted, each new factor accounts for less and less variance. Another reason could be the lack of items in the questionnaire which have sufficient power to discriminate a more homogenous factor. Thus with a reader segment or a random sample, the researcher operates in uncertainty that the groups will be homogenous. Multivariate typing allows the researcher to operate in more certainty. He will be able to examine each factor carefully before analysis, and he will have an opportunity to explore the lack of homogeneity of any particular group. Based on this logic, it was concluded that multivariate typing of groups on the basis of their evaluations of attributes tends to yield more homogenous groups than random sampling or segmenting on the basis of reading behavior. While the attitude groups have homogenous preferences, the data indicate that attitude segments have as hetergenous beliefs, or percep- tions, as the behavioral segments and the random sample. This is not difficult to understand. Persons may have similar likes and dislikes and yet have different experiences with an attitude object. These different experiences lead to different perceptions while the preferences remain similar. Such groups would be easier to deal with than groups with heterogenous evaluations and homogenous beliefs. To change the overall attitude of a group with heterogenous beliefs, a communicator plans communication to alter the negative beliefs. A media decision maker can 120 alter his content or programming or use communications to alter such beliefs. These plans are made with the existing, and homogenous, pre- dispositions of the segment or segments in question. Such efforts are more likely to be successful in improving a group's attitude than an effort which tries to alter existing evaluations. Cox (1961) supports this proposition, "The closer the match between the appeals used and the individual's predispositions, the more likely he is to expose him- self to the advertisement, and to act as desired" (p. 51). Therefore, one could conclude that operant multivariate typing results in attitude segments which are homogenous in a way that permits planning communication or media content that is consistent with what people like and dislike. FACTOR ARRAYS One of the major implications of the use of operant multi- variate typing and the Fishbein attitude model in mass media attitude research is the opportunity to study the attitude content for each segment. To study the attitude content, a set of stereotypical responses for each segment is needed. Such a set of responses can be called a "factor array." A factor array is a ranking of e items from good to bad and b items from probable to improbable, created from the scores of respondents significantly loaded on a particular factor. In other words, a factor array is a prototypical set of responses to the questionnaire that represent that beliefs and evaluations of a factor rather than a single individual. The analysis and interpretation of factor arrays permit the researcher to examine the attitudes repre- sented by each factor (Schlinger, 1969). Scores from respondents having significant loadingscn1each factor are used to construct the array. It is important to take into account the relative magnitude of respondents' factor loadings in constructing an array so more weight can be given to persons who have larger loadings because they are more representative of the factor than are persons with smaller loadings. For this reason, per- sons are weighted and factor arrays are based on the weighted sums of 121 122 raw scores assigned to each b and e item by the persons who represent the factor. The first step in constructing factor arrays is to weight each person. Stephenson (1953) and Schlinger (1969) suggest the formula: where w weight of respondent j 1 II factor loading of respondent j For example, person #1 had a factor loading of .79 on F1. The weight for that person would be: w _ .79 1 1- (.79)2 W] = 2.142 Once the weights have been assigned to persons, the next step is to multiply each respondent's original raw scores by the appro- priate weight. The same weight is used for the b and e items. If a person's loading on a factor is negative, his raw scores must be reversed (for example, a -3 score would become a +3, a ‘2 becomes +2, etc.) and the reversed scores weighted (Schlinger, 1969). This procedure prohibits multiplication of two negative numbers which would yield a positive number. A positive number would agg_to the weighted sum of an item rather than detract which should occur for persons with negative loadings. "This is perfectly acceptable, so long 35.211 of 123 the loading signs on that factor are changed, i.e., for every respondent n in the study" (Schlinger, 1969, p. 58). As an example of the weighting procedure, Table 18 presents the raw score, the weight, the weighted score, and the summed weighted scores for the three persons. Table 18. Respondent Weighting Procedure Factor Raw e1 Weighted Raw bi Weighted Person Loading Weight Score e1- Score Score bi Score 1 .79 2.1 2 4.2 -2 -4.2 2 .77 1.9 l 1.9 -l -l.9 3 .76 1.8 3 5.4 -2 -3.6 Total 11.5 9.7 Average weighted score 3.8 -3.2 The weighted summed scores can now be converted to Z scores using the standard formula: _Xi'u Zi ' s where: Zi = Z score for weighted sum for item i xi = weighted sum for item i u = average weighted sum for all b or e items 5 = standard deviation of the weighted summed scores 124 However, there is a problem in directly comparing the Z scores. Since the Z is standardized on the basis of the group mean and standard deviation, Z's from different factors may represent different positions along the +3 to -3 scale. However, the Z does provide a measure of the relative distance, in standard deviations, a particular item is from its group mean. Thus an item with a large Z is very likely very different from the mean of the group. Likewise, a factor with a Z for any particular item that is very different from the other factors' Z's on the same item is an item that makes the particular factor different from the others. At the same time Z scores are computed, it is convenient to divide the summed weighted scores by the number of persons significantly loaded on the factor to determine the average weighted sum for each b and e item. This procedure simply reduces the size of the weighted scores to a size comparable to the original +3 to -3 scales. This does not, however, reduce the scores to scores that are directly comparable to the original scales since they are weighted and the original scores are not. Once the Z scores are calculated, e items can be ordered from "most" good to "most" bad. The largest scores represent the items which are the "most" good while the largest negative scores represent the "most" bad. Accordingly, as the e items are ranked, the corresponding Z score for the b measure can be recorded. The larger scores represent those items which are "most" probable and vice versa. The result is then a prototypical set of responses to the e and b items representing the factor rather than a single individual. 125 Once the factor arrays are completed, it is an easy matter to compare the order of items between factors. A simple list of the items and their corresponding b and e scores is constructed, and since the scores are normalized to a Z distribution, they can be directly compared. This is simply a quick method of determining which items factors differ on. The first step in this comparison is to identify items for which the difference between each of the factors is less than i l Z-score. Items which fall into this category are considered "consensus items." A "consensus item" is interpreted as an item for which all factors are in essential agreement. As “consensus items" are searched for, the researcher is also looking for discriminating items. "Discriminating items" are defined operationally as items for which the difference between any one factor and the average Z-score for the remaining factors is f 1 Z-score. Theoretically, they represent items for which the factors are in essential disagreement. For example, the Z-scores for e items 1—5 are examined in Table 19. The Z-scores for the three factors differ by less than f l Z-score for items 3 and 5. They are considered "consensus" items. The remaining three, 1, 2, and 4, differ by more than t l Z-score so they are analyzed for discrimination. The Z-scores for F1 are compared to the average Z-scores for the other factors, F11 and FIII. The difference between FI Z-scores for these three items are greater than 1 so they are considered discriminating items. They are beliefs or evaluations that help the researcher understand FI and how it differs from the others. 126 Table 19. Consensus and Discriminating Items e Items FI FII FIII l 1.9 -.5 -.7 2 1.9 -.7 -l.0 3 1.0 .9 .7 Consensus 4 .8 -.5 -.3 5 -.2 .3 0.0 Consensus Calculations: e Items FI Average Z (FIIi-FIII) Difference l 1.9 (-.5)+(-.7)=(-l.2)+2=(-.6) l.9-(—.6)=2.5 2 1.9 (-.7)+(-l.0)=(-l.7)+2=(-.85) l.9-(—.85)=2.75 4 1.0 (-.5)+(-.3)=(-.8)+2=(-.4) l.0-(-.4)=1.4 Having constructed the factor arrays and identified the consensus and discriminating items, the researcher is left with the task of interpreting the arrays. Schlinger (1969) outlines three steps which must be taken in analyzing a factor array (p. 59): l. Analyze each array. What does the factor array "say" about the factor type? What does the prototypical response indicate about the needs, wants, and atti- tudes that factor projects? 2. Compare and contrast factors. Analyzing differen- tiating items will put each factor into sharper focus and perspective. Consensus items will reveal what beliefs and attitudes the factors share. 3. Compare other data from respondent interviews. Demo- graphic and behavioral characteristics will also help put each factor into sharper focus and perspective. 127 In addition, for a factor that proves difficult to interpret, or if the researcher would like to validate his interpretation, he may have a subject with a high, pure loading on the factor project on discriminating items for that factor (Mauldin, 1978). Factor Interpretations In order to convey some notion of the kinds of information this technique generates, the findings of the factors are discussed briefly. The reader should be cautioned that this analysis is not a generaliza- tion. The sampling used for this study does not permit generalization of results to any group other than the population of 237 students in the advertising class which supplied the final sample of 100. The following analysis is based on the 14 respondents from each factor that were used in this study. A demographic and behavioral description of the three factors is presented in Table 20. Consensus Items Three of the 37 items (Table 21) emerged as consensus evalua- tions. Subjects value reviewers who are knowledgeable (3). They like information about restaurants so they know what to expect (5). They are more neutral about pictures that give you an idea of what the news is without reading the story. However, belief that the State News had these characteristics varied across factors. Thirteen of the 37 items (Table 22) emerged as consensus beliefs. Subjects believe the State News covers East Lansing and MSU more than the world or nation (21). They believe the State News is entertaining and easy to read (24, 32, 31). Table 20. Demographic and Behavioral Profiles of the Factors 128 Characteristic FI FII FIII Age: 7 19.79 20.57 19.57 Sex: Male 86% 43% 7% Female 14% 57% 93% Classification: Freshman 28.6% 35.7% 35.7% Sophomore 28.6% 42.9% 21.4% Junior 28.6% 14.3% 28.6% Senior 14.3% 7.1% 14.3% Residence: 0n campus 78.6% 78.6% 71.4% Off campus 21.4% 21.4% 28.6% Number of Newspapers Read: X' 2.86 2.71 2.50 Like-Dislike ._ State News:. 1.07 1.00 1.57 Days mtg _ Hgyg Read; X 4.29 4.36 4.07 Percentage __ Read: X 44.29 58.93 63.93 Zbiei 'X 19.14 33.21 68.79 129 m. N. m. m. 28m .33 .99 m. a.- m. _e N mum umwwwmmmpu uoow .Nm m. m. o.o N. 28m .33 .99 9.2:... mcmcmoc uzocuwz pzonm mp mam: mg» was; mo m.- o.P- m.- N «our cm :0» m>wm was“ mmgspowa mo mac; .w R; N; m; E 28m .33 .99 cm so» cog: pumaxw op was: o.o m. N.- N gecx so» om pcmczmpmmc pzonm cowpmscoacm .m :9 o; E Z 88w .83 .99 m:P3mw>mL N. m. o.P N 0cm aux» was: “scam Zoe; on: mcmzmw>mm .m mcoum HHHJ NH; Hg .6“: .m>< msauN m mamcmmcou .PN mpnmh 130 m. N. m. o. .mcoom .uu: .m>< cmpmp cowummcm>cou N. m.- m.- N to. 800. “an NMN>NLp at. “as. memo. cmNNNL .o N. N. o. m. macaw .uo: .m>< N. o.o a. N mucoam m.:meoz “scam mmpowug< .v m. m. o. o._ .mcoum .upz .m>< .mzmc newton _Nm m.uw .303: mxwp New» u.:oc so» om N. o.o m. N mam: msp a: xmmgn “any mmgspuwa mcwpmmgmacH .NN o._ P._ N.N N. .Loom .nuz .m>< N. m. N. N new. co :3. .mcwx._.m .Nm m.N m.N m. o.— acoum .upz .m>< =mmma pace» ucouwm: m :o mwmcopm Pmoop can o._ m. m. N mama pcogm mg» co mzm: mumpm ucm chowumz .o_ ¢.P m.N o.F ¢._ macaw .epz .m>< swamp a com uoom N.N o._ m. N .xcouwn .zccam men “as“ Lopmum mgp op mgwuuwg .Nm m.N N.N m._ 9., mcoum .uuz .m>< ucmwmgmuc: op xmmm .maxuumczummw .Loszc mpupNN m spNz mwpowucm Locum; pan m.N N._ m. N .mzw: mcmgon »_puwcpm p.=mcm was“ mmpuwuc< .wm m.N N.— N.N ¢._ mcoum .cpz .m>< cowpm: Lo upcoz on» “zone mmPUNpcm cog“ egos m. o.N m. N am: use mcwmcmA gnaw uzonm mmPuNme _mqu .NN mcoum HHHN NH; NJ .nuz .m>< mewuH a mamcmmcou .NN mpamh 131 c._- N.N- w.N- macaw .caz .c>< wpaoaw N._- m.N- N Fccowwwmmoaa mo mamam>oo mpmpasoo ._ N.- a.- _.N- macaw .caz .c>< meaa m.N- N.N- N as a: mxcp ucc acop mam was“ wmpovpa< .NN Nc. N. c.c macaw .caz .c>< mswww mscw m.- m.- N mg» co meUNpam wsoamE=c op macaw mo no. < .mN w. c. w. macaw .caz .c>< m. N. N wcc cmwawwcha cccc .Nw w. N. w. macaw .caz .c>< vpaoz uwsocoom vac FcuNproa N. N. N mg» cw :o mcpom wN was: zocx ms mum; .m macaw Ham Ha .caz .m>< cmccaacca--NN mNch 132 Subjects are less sure the State News has interesting pictures to break up the editorial content (27), has articles on women's sports (4), has trivia filler items (6), has good classified ads (37), and devotes a lot of space to the same issue (28). Subjects are also unsure how well the State News provides political and economic news . Students do Hgt_believe the State News has complete coverage of professional sports. While there is consensus belief for these items, factors varied significantly in their evaluation of these attributes. Twelve items (Table 23) emerged as both consensus evaluative (e) and consensus belief (b) items. Subjects value news of serious issues and events (30). They like understanding the issues (34) but they don't like articles which include every little detail (12). Subjects believe the State News reports the serious issues and events (30) and they believe it doesn't report every little detail (12). However, subjects were less sure the State News gave them proper background on all stories (34). Subjects value informative headlines (18) but they are not sure whether or not the State News has such headlines. Subjects like an entertainment section that tells them what's going on (29) and they like human interest stories and pictures (25). Likewise, they believe the State News provides both attributes. Subjects place less value on newspapers written by young people (33), advertising supplements (23) and movie reviews (19). 133 _._ w. c. N.N macaw .caz .c>< N. m. o.N N Nana a; N; o.N w; macaw .33 .w>< mcEca wmgcs 3c .cccm Parmamfi. Na ac czca pcccc cc mcwcm w.uc;3 ms m. N. a. N wNNmp amga :chamw acmecwcpamacm c< .mN N. a. 0.0 N. macaw .cuz .m>< m.- a.- N.- N o.N N.N w._ w.N macaw .caz .m>< mNaNpac map ccma ca ac: ac amspmsz mcwamc ca ccwpceaca m. w. o.N N law gmcccm ms m>Nm was» wmcwpccm: .mN _.F c. w._ _._ macaw .caz .c>< a. N. o. N c.N N._ w.N N.N macaw .caz .m>< wa=m>m ccc w. c._ o.N N wmcwww wccwamw Ncma map ucccc wzmz ..om macaw HHHa NH; Ha .ca: .m>< wemaH c ccc m wcwcmwccu .mN mNch 134 c._ w._ c._. c._ macaw .caz .c>< m._ N.N m. N N.N w._ ¢.N m.N macaw .ca: .m>< o.N N. a. N wmaauawg ccc meacpw awmampcw sass: .wN N.. m. w.a c._ macaw .caz .c>< w. N. m. N N.N N.N _.N c.~ macaw .caz .c>< _Ncamc mNaaNN Nam>m mcwuaccma “sonar: w. w. o.o N cowacsacacw uccaaccsw map so» wm>Nw .NN c. w. c. a. macaw .caz .c>< a.- N.. o.o N zacuw mpcsz w._ m.N c.N w.N macaw .cu: .m>< mg» ccauwamccc caa H cw wacm>m ccc wmcwwm cc chacsachN ccccam m.. N.N N. N -xaac pacamaaacga meac w wzmz .cm macaw NHHa HHa Ha .ca: .m>< cacaaaccammwN mNch 135 N. N. N.N N. macaw .caz .c>< N. c. w.- N c._ N.N c. N.N macaw .caz .c>< . amcce m>cw ms c—m; c.c w.- N.- N acaa waamemNcccw ccawaaam>c< .wN w.N N.N N.N w.N macaw .ca: .c>< N.N c.N N.N N c.N N.N w. c.N macaw .caz .c>< Nwazs _. a.- c.c N ma_N m_ccmc cach aca amaaaaz .ww _.N _.a N._ c.N macaw .caz .c>< N. w. w. N N._ c._ N.N N.N macaw .caz .c>< N.- N. N.- N Amcce m>mw ms QNm; “can wccaccu .mN macaw NNHa NHL Ha .ca: .m>< cmzcwucoo 11mm mpncp 136 w. w. N.N w. macaw .caz .m>< N.- w. m. N c aw: cu m. N.N w. w. macaw .cuz .m>< cmacNma wchcmacc; ccc wchp -cNNccmac can wacamcaa ammpchc> N. w.- m.- N m pzcca cmEachN NNmz me wammx .N w. w. w. N.N macaw .caz .c>< N.- N. o. N c wNaa=EcaacN w. o.N o.N w. macaw .cpz .m>< .wcha .meaN>Npac achaNaacacapxm :N .wmwwaNa aNmaa :N . . . chcc N.- N.- a.- N m mac wucmccuw amapc was: me wNNmN .oN a.- N.- w.- N.N- macaw .caz .c>< N.N- a.- N.N- N c Nazca c.N N.N N.N c. macaw .caz .c>< NE mawcz a.ccc N cw . . . cc N macamc me>ce map mNNN NNN3 N NN N. N. w.- N m NNmp cca N cw . . . wsz>ma mN>cz .wN macaw NNa NNN Na .cpz .w>< cmchaccc--wN mNch 137 On the other hand, subjects strongly believe the State News is written for young people (33). They are less sure the State News has good advertising supplements (23) and they tend to Hgt_believe it has good movie reviews (19). Subjects also placed less value on news about other students (10) and campus organizations (7). The subjects believe the newspaper has both attributes. In order to improve the State News effort should be placed on providing better background information (34) and on writing better headlines. Likewise, the State News should provide better movie reviews (19). A11 reviewers who write for the State News should be well informed and knowledgeable (3). With the movie reviews, the State News should also provide better information about restaurants and eating places (5) and things to do in the area (29). Factor 1: Sports Enthusiasts Factor I is composed primarily of men (see Table 20). They read the State News an average of 4.286 days a week and they read 44.286 percent of the newspaper. 'Their Zbiei score is the lowest of the three factors indicating they have a less favorable attitude toward the State News. In order to stereotype the psychological profile of F1, this factor was named "Sports Enthusiasts." This factor's discriminating items are presented in Table 24. They very much like complete coverage (20) of all types of men and women's sports (14), college (2) and professional sports (1). 138 Regarding the State News, F1 believes the newspaper is lacking in its comprehensive sports reporting (20) and in its coverage of college (2) and professional (1) sports. 0n the other hand, they believe the State News covers women's sports (4) and sports that don't always make the headlines (14). While subjects value summing news sections (11), F1 values them less than the other segments. They do believe the State News has a summary news section. Subjects don't like trivial filler items (6). The State News could improve the attitude of the Sports Enthusiasts by providing more comprehensive sports reporting and more college and professional sports news. The State News should avoid using trivial filler items, especially in the Sports Section. Factor II: Hard News Readers Factor II is composed of 6 (43 percent) males and 8 (57 per— cent) females (see Table 20). They read the State News about four days a week and they read almost 60 percent of the newspaper. Their average Zbiei score was greater than FI but less than FIII. FII, whose discriminating items are presented in Table 25, is interested in hard news. Subjects value complete coverage (36) of national and international news (35). Unfortunately for the _S_ta_t_g News, they do not believe the State News provides such news. They like the national and state news on the front page and local news on a "second front page" (16). They believe the State News has such a layout. 139 w: N.- N.- w.- w. amaaN :chcwam>:ca ww.N- wN.- N.N- o.N- aca cccm asp NcN>Nau mac was“ wEmaN amNNNa .w wa wc. wN.N N.N N.N NNcamc ac acN c acccaaa chcscaa NzccuNz :cNNc: cca cNacz may _aN :c chcm w.pa;3 wo.N- wo.N o.o N.N “so cha sac N cams: amaccwzmc c cN :chamw < .NN w: mm. wN.- w. N.N :N mucaNaNpacc :ca N wNaccw map meN wchN N.N m.- N. m.N -ccm; map mace wzcch p.:cc Nag» wuaccw paccc wzmz .NN w: mm. mo. 9. m. N.N a.- m. c.N wuaccw w.:mEc3 ucccc meaNaa< .c w: N.- m.N- w.N- N.N- ow.N a.- m.N m.w wpaccw Ncchwmecaa ac mmcam>ca mNmNcEcu .N w: mm.- wN.- N.N- N.N- wN.N ww.a m.N m.m wpacaw mmmNNca ac mmcam>ca mpmNQEco .N ¢.N- o.N- «.N- w.N- waNuwNpcpw Ncch>Nch ccc Emma .wmacaw xcc .wchcccpw .mezcmgaw wc m.N N.N- N.N w.m wchsy gazw nuNz .wNacaw No chpacama m>chmsmaaEcu .ON .NNNo N ~m>< N macaw wchpchc>w .cuz .m>< wEmNN chpcsNaawNo N acuaca .em mNnmN 140 «.N- o.N- N.N- w.N- c waNuwNpcpw Ncch>Nch ccc Emma .wmacaw xcc .wchcccpw .mescmcaw wc N.N N.N- N.N w.m m wchcu gasw cuN: .waaccw Nc chuaccma m>chmcmaQEco .ON .NNNQ N «ww<. N macaw wamNNmm .cpz .m>< cmchaaca-.cN mNch 141 Subjects want to know what is happening in the political and economic world (9) but are not sure the State News keeps them in- formed. 0f some value to FII is college (2) and professional sports (1) news but, like FI, they don't believe the State News provides such sporting news. Three beliefs also help discriminate 1:11. These subjects are less sure than the other factors that the State News does not have well-informed reviewers (3). They do not beleive the State News has a variety of Letters to the Editor (17). On the other hand, these subjects believe more than the others that the newspaper pro- vides information about restaurants in the area so one knows what to expect (5). Factor III: Ludenic Readers ~ Factor III is composed primarily of females. They like the State News more than the other factors. They read the newspaper about four days a week and they read it more completely (64 percent) than the other factors. Their Zbiei was the largest of the three factors. FIII, whose discriminating items are presented in Table 26, is named the Ludenic Reader because these persons so well fit Stephenson's Ludenic Theory of Newsreading (1964). These persons enjoy reading the newspaper. It is subjective play. After reading the newspaper, both good and bad news, these persons would say that it was absorbing, interesting, and enjoyable. 142 w: N. 0.- w.- o.o wN.N- we. N.- c. wuaccw mmmNNca ac mwcam>ca mpmNcEcu .N w: ww. mm.N- N.N- N.N- oN.N- w. w.- w. wpaccw Ncchwmecac No mmcam>ca mumNcsco .N w: o.o w.N- m.N- a.- N.N m.m- N.N- N.N- mENu NE c: mxcp ccc mch mac ucga meaNaa< .NN m: N. —vo' —.o No cNacz aNEcccam ccc N.N N. c.N N.N NcaNNNNcc ms“ :N :c chcm wN “ca: 3c=¥ me wumN .m w: wo.. mm. m. m. cmmcc uccaN cccamwa c cc meacpw w.N N. N.N N.N NcacN ccc mmcc accaa mg» no wzm: macaw ccc Ncchpcz .wN w: wm.- mm.N- w.N- N.N- wzm: w.N N. N.N N.w NcchuccamucN ccc Ncchucc Nc mmcam>ca mpmNcEco .wm w: wm.- ww.- N.N- N.N- wcmccc; ucsz ac wuamaam Eampuuaccw ccc -mch map N.N m.» w.N m.N ch>Nm . . . NNm>Nuamwcc .NNmpmNcEca wmcwwN wam>cu .wm .NNNQ N sm>< N macaw wchpchc>m .cpz .m>< wEmaN chacaNENaach NN acaaca .wN mNch 143 N.N- w. N.N- w.- aaccp mEcw may on ccza amauca waaccp w: w. w. w. N.N ac NpmNac> c cu cmac>mc wacaacm map on wampuma .NN N.N w.- w. c.N accaccpwma achaNNacc c ca on so» cmaz Namcxm w: c. N.- m. w.N ca acgz 2ccx so» cw waccazcawma acccc chacEacacN .w wN.N wN.N- N.N- w.- w: we. mm. m. N.N mcazma>ma mac cha ucsz socx ca: wamzma>mm .m .aaNo N .w>< N macaw wamNNmm .caz .m>< o.N N.N- m.- N.- waapwaucpw chca>acca ccc Ecma .wmacaw xcc .wmcacccaw .mescmaaw wc wN.N- wN. o.N- N. wmcasp gacw saaz .waaccw ac mcaaaccma m>chmgmacEcu .ON N.N- o.N N.- w. Ncc sacm acmz ca aczz mcaamc N.N- N. N.N- o.o :ca N cw amcacmz mg» acccc me wNNmN acca :capamw < .wN w: mm.. mm. o.o w. cwzma 9_Nacc NNc w.aN .3c3. mNNN Nmma a.ccc N.N- m. a.- N. no» cw meaNaac map c: acmac pcsa wmacaaac chNwmamacN .NN .aaao N .m>< N macaw wccapchc>m .caz .m>< cmcaaaacc--wN mNch 144 w: wo.- wN.N- N.N- w.- c.N- N. N.- N. wpaccw Ncccawwmacac ac mmcam>ca mamNcEcu .N w: w.- w.- a.- o.o mcwwa o.N- m.N- m.N- a- mEcw may go wmaacaw wscamE=c cu macaw ac acN < .wN w: N.- m. N.N w.N auscN c aca cccm c.N w.- N. c.N .Naaaac .Ncaca mac acaa acaNcc mac ca wamaama .Nw w: .1 N. w.- N. wo.N ww.- c. c.N wcc cmNancha cccw .Nm w: wo.- wN. N. m. , =wzmc ccaacc NNc w.aN .zcz. maNN Nmma a.ccc ch N.N w.- N. c.N cw meaapac map a: xcmac acca wmaaaaac acaawmamch .NN oo.N m. m.N N.N zcc :acm acmz cu “cg: mcaamc ww.N ww.- m. N.N cca N cw amaacmz map uzccc me wNNma accp ccauamw < .wN w: wm.- we. N. N. amacN ww.N wN.N- a. a.N :caucwam>:ca aca cccm use Nca>Naa mac acga wsmpa amNNNa .c .aaE .fiNN N macaw wcc 5c: Nc>w .ca: .m>< wsmaN caNachsNaach NNN acaaca .cN mNch 145 c.N- N.N N.N- N.- waapwaacaw Ncaca>acca ccc . Ecma .wmacaw xcc .wmcacccpw .meccmsaw wc wmcaap mm.N- mm. N.N- a. scam spa: .wuaccw ac acaaacama m>awcmgmacscu .ON w.N- w.N N.- w. chacEcaaca .wczNa .wmaua>aaac achaNaacacaNxm ca .wmwcha aamcu w: N. .1 N.- o.N ca . . . mcacc mac wucmccpw amgpc acgz ms wNNmN .oN oo.N m. m.N c.N Ncc :acm acmz ca pcgz mcaamc ww.N mm.- m. N.N :ca N cw amspcmz map pcccc me wNNma acca ccapamw < .wN wN.N wm.- w. N.N waaccu w: w. c. o.N N.N ac zumaac> c ca cmuc>mc acpacm may ca wampuma .NN .aaao N .w>< N macaw wamaNmm .cuz .m>< c.N- N.N N.N- N.- waaawaacpw Nccca>acca ccc Ecmp .wmacaw xcc .wmcacccaw .meccmsaw wc ww.N- mm. N.N- a. wmcacu nacw spa: .wpaccw ac chNaccma m>awcmcmaqsco .oN .aaao N .m>< N macaw wccaucch>w .mp3 .m>< cmchacca--cN mNch 146 They like trivial items they can use in their conversations (6); they like interesting pictures (27); and they like Letters to the Editor that are fun, good fora laugh (32). They want news about the weather so they can decide how to plan their day (26). They look good classified ads (37). They are more neutral toward lots of space being devoted to the same issue (28). While they like comprehensive sports news and news of professional sports, they like this news much less than the other segments. Regarding FIII beliefs, we find the Ludenic Reader believes the State News provides the types of news they like except for pro- fessional sports news. These subjects are less sure the State News covers student activities (10) and they believe more than others that the State News carries Letters to the Editor devoted to a variety of topics (17). To satisfy Ludenic Readers, the State News really doesn't have to change anything, just keep up the good work. Adding professional sports news and more comprehensive sporting news would not offend this group. Discussion The evidence from this analysis indicates that the State News is perceived to be a college newspaper targeted to college students. It has content that is fun and entertaining as well as informative. Subjects like information on entertainment in the local area and they want to learn information from a well-informed reviewer. 147 The State News is perceived generally to cover the important issues and events without excessive detail but does not provide enough background information for the reader to understand the whole story. In addition to this general attitude toward the State News, there are three independent attitudes. One, the Sports Enthusiast, is primarily concerned with sports news and is dissatisfied with the coverage in the State News. The second, the Hard News Reader, wants more hard news, complete reporting of national and international news as well as complete political and economic news. Unfortunately for the newspaper, this group does not believe the State News provides what it-wants. The third attitude, the Ludenic Readers, is differentiated from other groups by its preference for entertaining content. They want their newspaper to be fun, and they believe the State News meets their expectations. With this description of the market, the State News could improve the attitude of its readers by deleting and adding certain content items, as discussed. The State News product could be altered to meet the expectations of the market, in general, and each of the attitude segments. This could be accomplished without adding or deleting matter that would offend any particular segment. IMPLICATIONS AND CONCLUSION In addition to the practical implications for decision making, this study has implications that are theoretical and methodological. Theoretical Implications Theoretically, this study found strong support for use of the Fishbein attitude model in mass media attitude research. There is clear evidence for the need to discriminate between beliefs and evaluations in attitude studies. For example, consider the average weighted scores for item 20 and item 1, presented in Table 24, page 139. FI, the Sports Enthusiasts, place great value on comprehensive reporting of sports and complete coverage of professional sports. They do not believe the State News provides either. There is a difference between what these persons liked and what they believed about the State News. This is strong evidence that a beliefs-only model could perform like the Fishbein model. This study also provides evidence that use of the model without multivariate typing is likely to result in a meaningless average. Consider the average weighted scores for item 20: F1 = 3.6; FII = 0.1; and FIII = 0.4. Grouping all persons together would yield approximately 1.4, a score not indicative of any of the factors' scores. Multivariate typing helps overcome this problem by reducing extraneous and systematic variance. This variance arises from (1) the 148 149 number of attitudes, represented by the number of operant factors; (2) the similarity/dissimilarity between attitude segments; and (3) the distribution of individuals across attitudes, represented by the number of persons significantly loaded on a factor. Table 3, page 95, demon- strates this conclusion. Each reader segment and the random sample is composed of persons significantly loaded on a factor. The homogeneity of any reader segment or random sample could be attributed to the factors. For example, FI represents about half of the Medium Readers. There is also evidence that attitude segmentation is preferable to behavioral segmentation. However, other methods of segmentation need to be compared to attitude segmentation to determine which yields the more homogenous groups. Other variables, such as sex, age, income, etc., have yet to be tested. There is evidence in this study that sex may not work well. Table 20, page 128, reveals that F1 is predominantly male, FIII is predominantly female; and FII'Hscomposed of an equal number of men and women. Certainly this method needs to be tested with larger samples. While this study was limited in sample size by the SPSS Q-factor pro- cedure, other procedures were available which will factor larger numbers of persons. William Wells (1979) reports using a procedure which 0- factors over 2,000 individuals. This study tested the Fishbein model among attitude segments with reduced variance, and the model continued to perform. However, the theoretical importance of reduced variance may lead to a methodologi- cal problem. Variance is the basis for most statistical tests. If we reduce variance to zero, the statistical tests will not perform. For 150 example, while FIII had the highest correlation between A0 and Zbiei, FIII had more variance. The variance contributes to a larger correla- tion but detracts from homogeneity. This is a problem that must be addressed in future research. Methodological Implications One of the most important methodological implications of this study is an operational definition of homogeneity. This concept has been used extensively in the literature but it has not been operation- . ally defined or tested. Homogeneity can be defined as the average amount of variance about each attitude item--the less variance a group has, the more homogenous it is, and vice-versa. Also of great importance is the partial validation of a method that operantly identifies groups that are homogenous enough for cross- sectional analyses. This means the Fishbein attitude model can be used in mass media attitude research. The result will be research which yields (1) a summary measure of a group's attitude (Zbiei) which will indicate the "primeness" of the group and (2) information that permits the study of attitude content for planning messages and/or media content. The data should help further our knowledge of this mass media. The procedure for constructing factor arrays is a method for analyzing attitude content. We have seen how such an analysis helps the researcher understand attitude content and how that understanding can be used in decision making. 151 Operant multivariate typing when used with the Fishbein attitude model should yield a more precise measure of attitude. This in turn should aid in a more precise assessment of mass media effects. The pro- cedure provides homogenous groups which could be used in experimental, as well as in survey research designs, to assess media effects. Experimentally, the procedure would help control extraneous variance due to the heterogeneity of subjects. Grouping people who have similar predispositions should yield a more valid test of communication effects. If predispositions effect how messages are perceived, a test of message effects should account for subjects' predispositions. Other- ' wise, the message effects may be diluted by the different predispositions within an experimental group. While random selection and random assignment is often used to overcome this problem, the study provides evidence that a random sample is likely to be composed of persons with different attitudes and therefore much more variance. For survey research, the procedure suggested here would help in overcoming the "meaningless average." No longer do subjects with dif- ferent attitudes need to be lumped together in assessing the public attitude. The procedure allows the researcher to identify and analyze the different attitudes that may exist. Practical Implications With more precise definitions of attitudes, mass media decision making should be more effective. Having groups with homogenous predis- positions permits accounting for selective exposure, selective percep- tion and selective retention in planning communication strategy. 152 Likewise it allows mass media institutions like newspapers, magazines, and television and radio stations to plan their content and programming with a consumer orientation. Knowing what people want and designing a system to meet those wants is the basic premise of the marketing concept, :1 concept being adopted by many media, especially newspapers. Public-policy decision making can also be accomplished with a better understanding of the complexity of issues and what beliefs and evaluations surround the issues. Ascertainment studies will provide a more precise definition of community needs. Stations will be able to better plan programming for each unique segment of viewers. In addition, the methods outlined here allow media to plan for a general audience where there is uniformity of opinion. Similarly, in determining deceptive advertising, the procedure would account for the effects of both beliefs and evaluations of different segments. The method would better inform policy makers who is being affected and how. The Fishbein attitude model and multivariate typing will also help improve content decisions for media. The discussion of factor arrays and their implications for the State News is a good example. Table 24, page 139, shows evidence that the State News is not perceived by the Sports Enthusiasts to provide good sports coverage where there is a strong need for such content. With more precise attitude data, advertisers will also be able to make decisions better. The segmentation will facilitate target audience selection and message strategy development. A company could attack a single segment or it could decide to target a unique product 153 and promotional effort to each segment. Knowledge of the expectations of consumers in each segment will also provide information for planning product improvements or development of new products to meet unfilled needs. Conclusion The purpose of this study was to design and test a method for mass media attitude research using the Fishbein attitude model. The method used operant multivariate typing to identify homogenous groups so the model could be used in survey research. While the data do not over- whelmingly support the contention that attitude groups are more homogen- ous than the typical random sample or traditional user segments, there is evidence that attitude segments are more homogenous. Since the Fishbein attitude model performed significantly for each attitude segment, use of the model in conjunction with operant multivariate typing is recommended. However, the results of this study and its limitations provide clear direction for further research in this area. Methods outlined here should aid researchers in the use of the Fishbein attitude model and multivariate typing. Likewise, an opera- tional definition of homogeneity is provided. These methods should provide more precise attitude research. More precise data should aid in mass media decision making and in the furtherance of our knowledge of the mass media and their effects. APPENDICES APPENDIX A FOCUS INTERVIEW GUIDE TO THE INTERVIEWER: What we are looking for as a result of this interview is a description of how the subject reads the State News; verbatim opinion statements about what the subject likes and dislikes in the paper; reasons for using the aper (e.g., to find out national news, movie listings, sale ads, etc.) and some personal information. Purpose: My class project involves describing in detail how an individual uses a newspaper. Demographic Information: Class: F___S__ J___Sr__ G__ Age: Sex___ Live On-Campus Off-Campus Race or Ethnic Background Socioeconomic Background: Lower;___ Low-Middle____ Upper-Middle___ UPPER... Do you read a newspaper regularly? Which one(s)? Why? Do you ever read other papers? Which? Under what circumstances would you? (Give the subject a copy of the State News, have her/him go through the paper as s/he normally would and describe what s/he notices, reads, likes, dislikes . . . everything involved in reading the paper. Record description and comments. PROBE.) (If the subject doesn't mention some of the following sections, ask her/him to comment. For example: "You didn't mention the little 'Inside Friday' box, do you usually read it?" or "There was no Enter- tainment Page today, but when there is do you read it?") SECTION Front Page 154 155 Focus World/Nation, etc. Second Front Page Opinion Page Entertainment Book Reviews Sports Classifieds It's What's Happening Comics TV Listings Ads (What kinds of ads attract your attention?) Advertising Supplements How often do you read the State News? Where do you pick it up? When? Where and when do you usually read it? What would you say is the best newspaper? The worst? Why? APPENDIX B NEWSPAPER STUDY Student Number Initials 1. What is your age? 2. Sex? (Circle one) 1. Male 2. Female 3. Year in college? (Circle one) 1. Freshman 2. Sophomore 3. Junior 4. Senior 5. Graduate Student 4. Where do you live? (Circle one) 1. On-campus 2. Off-campus 5. Type of residence? (Circle one) 1. House Apartment Dormitory ’Fraternity/Sorority House 01 -§ 00 N o o o 0 Mobile Home or Trailer 156 157 Which of these newspapers do you read? (Check the ones you read). ______MSU State News Lansing State Journal Detroit Free Press Detroit News New York Times ______Wall Street Journal ______ Other (please specify: ). For those newspapers that you read, please indicate the extent to which you like or dislike each one by placing a check on the scale position that indicates your liking or disliking. MSU State News Like : : : : : : Dislike Lansing State Journal Like : : : : : : Dislike Detroit Free Pres Like : : : : : : Dislike Detroit News Like : : : : : : Dislike New York Times Like : : : : : : Dislike Wall Street Journal Like : : : : : : Dislike Other (please specify: ) Like : : : : : : Dislike 158 8. Do you read the State News? (Circle one) 1. Yes 2. No 9. How many days a week do you usually read the State News? (Circle one) 0 1 2 3 4 5 10. What percentage of the State News do you usually read? 11. How completely do you read the State News? Completely : : : : : : not completely The next portion asks you to consider several statements about newspapers in general. Please indicate whether the statements describe a newspaper that you believe is good or bad. Place a check of X mark on the scale position that indicates the extent of the good-bad of the newspaper: The closer you mark to the word "good," the better the newspaper and the closer you mark to the word "bad," the worse the newspaper. The middle position should be marked only if you don't know how good or bad the characteristic is or if you think it is neither good nor bad. Make only one mark per scale and mark lg a blank space, not on the border between spaces. 1. A newspaper that provides complete coverage of professional sports. Good : : : : : : Bad 2. A newspaper that provides complete coverage of college sports. Good : : : : : : Bad 3. A newspaper with reviewers who know about what they are reviewing. Good : : : : : : Bad 4. A newspaper with articles about women's sports. Good : : : : : : Bad 5. A newspaper with information about restaurants so you know what to expect when you go to a particular restaurant. Good : : : : : : Bad 10. 11. 12. 13. 14. 15. 159 A newspaper with filler items that are trivial but good for conversation later. Good : : : : : : Bad A newspaper that keeps me well informed about volunteer programs and organizations and happenings related to MSU. Good : : : : : : Bad A newspaper with lots of pictures that give you an idea of what the news is about without reading anything. Good : : : : : : Bad A newspaper that lets me know what is going on in the political and economic world. Good : : : : : : Bad A newspaper that tells me what other students are doing . . . in their classes, in extracurricular activities, clubs, intramurals. Good : : : : : : Bad A section in a newspaper where I can find out what's going on in the world and nation without thumbing through a lot of detail. Good : : : : : : Bad A newspaper that gives you the important information without reporting every little detail. Good : : : :: :1 : Bad A newspaper with coupons that help me save money. Good : : : : : : Bad A newspaper that carries news about sports that don't always make the headlines, like the sports I can participate in. Good : : : : : : Bad A newspaper column that tells me everything that's happening on campus in capsule form. Good : : : : : : Bad 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 160 A newspaper with national and state news on the front page and local stories on a "second front page." Good : : : : : : Bad A newspaper that carries Letters to the Editor devoted to a variety of topics rather than a bunch of letters all devoted to the same topic on one day. Good : : : : : : Bad Headlines that give me enough information to decide whether or not to read the article. Good : : : : : : Bad A newspaper that reviews movies so I can tell if I will like the movie before I go . . . so I don't waste my money. Good : : : : : : Bad A newspaper with comprehensive reporting of sports, with such things as schedules, standings, box scores, team and individual statistics. Good : : :* :: : ' :: Bad A newspaper that has local articles about East Lansing and MSU more than articles about the world or nation. Good : : : : : : Bad A newspaper with articles that are long and take up my time. Good : : : : : : Bad A newspaper with advertising supplements that help me save money. Good : : : : : : Bad A newspaper with articles that aren't strictly boring news, but rather articles with a little humor, feature—type, easy to under- stand. Good : : : : : : Bad A newspaper with human interest stories and pictures. Good : : : : : : Bad 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 161 A newspaper with a section that tells me about the weather so I can decide what to wear each day. Good : : : : : : Bad A newspaper with interesting pictures that break up the articles so you don't feel like "Wow, it's all boring news." Good : : : : : : Bad A newspaper that devotes a lot of space to carrying numerous stories on the same issue. Good : : : : : : Bad A newspaper with an entertainment section that tells me what's going on about town or if there's any good, old movies coming back. Good : : : : : : Bad A newspaper that carries news about the real serious issues and events. Good : : : : : : Bad A newspaper that is relaxing, fun to read. Good : : : : : : Bad A newspaper with Letters to the Editor that are funny, bitchy, good for a laugh. Good : : : : : :' Bad A newspaper that is written for young people like myself. Good : : : : : : Bad . News stories that report background information on issues and events so I can understand the whole story. Good : : : : : : Bad A newspaper that provides me a complete coverage of national and international news. Good : : : : : : Bad 36. 37. 162 A newspaper that covers issues completely, objectively . . . giving the long- and short-term effects of what happens. Good : : : : : : Bad A newspaper with good classified ads. Good : : : : : : Bad The last portion asks you to consider several statements about a specific newspaper, the State News. Please mark the scale position which best shows how much you believe the newspaper is described by the statement: the closer you mark to the word "probable," the stronger your belief the statement is correct and the closer you mark to the word "improbable," the less you believe the statement is correct. The middle position should be used if you don't know or are not sure the statement describes the newspaper. 1. The State News provides complete coverage of professional sports. Probable : : : : : : Improbable The State News provides complete coverage of college sports. Probable : : : : : : Improbable The State News has reviewers who know about what they are reviewing. Probable : : : : : : Improbable The State News carries stories about women's sports. Probable : : : : : : Improbable The State News has information about restaurants so you know what to expect when you go to a particular restaurant. Probable : : : : : : Improbable The State News has filler items that are trivial but good for con— versation later. Probable : : : : : : Improbable The State News keeps me well informed about volunteer programs and organizations and happenings related to MSU. Probable : : : : : : Improbable 10. 11. 12. 13. 14. 15. 16. 17. 163 The State News has lots of pictures that give you an idea of what is happening without reading anything. Probable : : : : : : Improbable The State News lets me know what is going on in the political and economic world. Probable : : : : : : Improbable The State News tells me what other students are doing . . . in their classes, in extracurricular activities, clubs, intramurals. Probable : : : : : : Improbable The State News has a section where I find out what's happening in the world and nation without thumbing through a lot of details. Probable : : : : : : Improbable The State News gives you the important information without reporting every little detail. Probable : : : : : : Improbable The State News carries coupons that help me save money. Probable : : : : : : Improbable The State News carries news about sports that don't always make the headlines, like the sports I can participate in. Probable : : : : : : Improbable The State News' What's Happening column tells me everything that's happening on campus in capsule form. Probable ': : : : : : Improbable The State News carries national and state news on the front page and local stories on a "second front page." Probable : : : : : : Improbable The State News carries Letters to the Editor devoted to a variety of tapics rather than a bunch of letters all devoted to the same topic on one day. Probable : : : : : : Improbable 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 164 State News' headlines give me enough information to decide whether or not to read the article. Probable : : : : : : Improbable The State News movie reviews tell me if I will like a movie before I go . . . so I don't waste money. Probable : : : : : : Improbable The State News carries comprehensive reporting of sports, with such things as schedules, standings, box scores, team and individual statistics. Probable : : : : : : Improbable The State News has local articles about East Lansing and MSU more than articles about the world or nation. Probable : : : : : : Improbable The State News has articles that are long and take up my time. Probable : : : : : : Improbable The State News has advertising supplements that help me save money. Probable : : : : : : Improved The State News has articles that aren't strictly boring news, but rather articles with a little humor, feature-type, easy to under- stand. Probable : : : : : : Improbable The State News carries human interest stories and pictures. Probable : : : : : : Improbable The State News has a section that tells me about the weather so I can decide what to wear each day. Probable : : : : : : Improbab1e The State News has interesting pictures that break up the articles 50 YOU don't feel like "Wow, it's all boring news." Probable : : : : : : Improbable 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 165 The State News devotes a lot of space to carrying numerous stories on the same issue. Probable : : : : : : Improbable The State News entertainment section tells me what's going on about town or if there's any good, old movies coming back. Probable : : : : : : Improbable The State News carries news about the real serious issues and events. Probable : : : : : : Improbable The State News is relaxing, fun to read. Probable : : : : : : Improbable The State News carries Letters to the Editor that are funny, bitchy, good for a laugh. Probable : : : : : : Improbable The State News is written for young people like myself. Probable : : : : : : Improbable State News' news stories report background information on issues and events so I can understand the whole sto y. Probable : : : : : : Improbable The State News provides me a complete coverage of national and international news. Probable : : : : : : Improbable The State News covers issues completely, objectively . . . giving the long- and short-term effects of what happens. Probable : : : : : :' Improbable The State News has good classified ads. Probable : : : : : : Improbable THANK YOU. APPENDIX C VARIMAX ROTATED FACTOR MATRIX AFTER ROTATION WITH KAISER NORMALIZATION Person FI FII FIII P1 ..79345 .33435 -.OO714 P2 .67196 .16033 .06895 P3 .25287 .49473 .18038 P4 .59639 .06250 .24314 P5 .03498 .72233 .17639 P6 .62876 .23370 .36191 P7 .60683 .20559 .50687 P8 .32677 .03253 .65936 P9 .49368 .49065 .13445 P10 .68629 .03422 .07099 P11 .24304 .39398 .20436 P12 .45650 .15418 .25498 P13 .65044 .29661 .30063 P14 .40245 .44072 .38728 P15 .28818 .21973 .51325 P16 .56008 .29128 .14085 P17 .44640 .33381 .03352 P18 .36377 .36337 .21187 P19 .20828 .60710 .49339 P20 .44486 .45978 .00163 P21 .52700 .39190 .27022 P22 .61026 .50029 .11708 P23 .06034 .76812 .00632 P24 .30565 .34579 .38371 P25 .10199 .04613 .42882 P26 .42884 .13939 .29917 P27 .43608 .46062 .26787 P28 .67057 .16614 .03812 P29 .36367 .47469 .02195 P30 .39854 .38942 .34844 166 167 Person FI FII FII P31 .68025 .33400 .00838 P32 .52372 .07337 .26881 P33 .09955 .65645 .28183 P34 .43375 .31887 .21889 P35 .51502 .10152 .19088 P36 .71614 .06974 .13588 P37 .46862 .43259 .51175 P38 .49860 .20838 .45913 P39 .42590 .08679 .59617 P40 .35055 .30340 .10511 P41 .51712 .31839 .01277 P42 .54382 .39865 .29573 P43 .07349 .68989 .39179 P44 .50378 .46683 .02490 P45 .38888 .08174 .13658 P46 .29360 .15476 .12958 P47 .52922 .14234 .21019 P48 -.25017 .40124 .50430 P49 .62703 .13841 .37344 P50 .71367 .09991 .06692 P51 .42133 .32991 .18003 P52 .03810 .34203 .49783 P53 .43294 .47418 .32781 P54 .63738 .19557 .00794 P55 .49386 .63823 .02978 P56 .50814 .20275 .19672 P57 .41916 .22040 .17006 P58 .68624 .37665 .13452 P59 .16034 .73923 .38669 P60 .43253 .54411 .05054 P61 -.12445 .26939 .30785 P62 .50157 .11701 .19053 P63 .26421 .52330 .51907 P64 .49075 .03449 .50683 P65 .49004 .64967 .12825 P66 .46831 .29072 .03097 P67 .52666 .68937 -03822 P68 .49308 .60764 .32919 P69 .50827 .05114 .54317 P70 .00436 .37938 .09490 168 Person FI FII FIII P71 .20474 -.06231 .62513 P72 .58738 .51293 .20129 P73 .77202 .22447 .21740 P74 .03613 .13688 .18478 P75 .10303 .30581 .52922 P76 .42790 .39983 .12987 P77 .64980 .27288 .02939 P78 .02261 .08579 .19339 P79 .30141 .41306 .14590 P80 .48443 .37703 .34822 P81 .76788 .04592 .24694 P82 .14287 .61963 .52364 P83 .36721 .47041 .41571 P84 .27470 .45840 .08234 P85 .40891 .17349 .48383 P86 .28959 .27579 .39561 P87 .25199 .49933 .29878 P88 .58829 .30624 .29147 P89 .11501 .62097 .32460 P90 .48671 .23958 .66011 P91 .60469 .04983 .25606 P92 .29691 .43406 .62160 P93 .55653 .52302 .38067 P94 .40440 .07706 .58486 P95 .18425 .36891 .53856 P96 .57141 .14152 .58422 P97 .13893 .01453 .46384 P98 .58946 .18705 .01927 P99 .66764 .16911 .23201 P100 .00402 .77450 .02912 BIBLIOGRAPHY Aaker, D. A. & Myers, J. G. Advertising_management. Englewood Cliffs, N. J.: Prentice-Hall, Inc., 1975. Achenbaum, A. A. Knowledge is a thing called measurement. In L. Adler & I. Crespi (Eds.), Attitude research at sea. Chicago: American Marketing Assoc., 1966. Adams, A. A. Broadcasters' attitudes toward public responsibility: an Ohio case study. Journal of Broadcasting, 1972, 16:4, 407-420. Adams, A. A. A study of veteran viewpoints on TV coverage of the Viet- nam War. Journalism Quarterly, 1977, 54:2, 248-252. Alderson, W. Marketing and executive action. Homewood, 111.: Richard D. Irwin, 1957. Allport, G. W. Attitudes. In C. Murchison (Ed.), Handbook of social 5 cholo . Worchester, Mass.: Clark University Press, 1935, 798-884. Atkin, C. K. The impact of political reports on candidate and issue preferences. Journalism Quarterly, 1969, 46, 515-521. Atkin, C. K., Bowen, L., Nayman, O. B., & Sheinkopf, K. G. Quality versus quantity in televised political ads. Public Opinion Quarterly, 1973, 37, 209-224. Axelrod, J. N. Attitude measurements that predict purchases. Journal of Advertising_Research, 1968, 8, 3. Baker, B. 0., Hardyck, C., & Petrinovich, L. Weak measurements vs. strong statistics: an empirical critique of S. S. Stevens' proscriptions on statistics. EdUcational and Psychological Measurement, 1966, 26, 291—301. Baroody, W. J., Jr. From the Publisher. Public Opinion, March-April 1978, 2. Bartos, R., & Dunn, T. L. Advertising_and consumers: new perspectives. New York: American Association ofiAdvertising Agencies, 1976. 169 170 Bass, F. M. Fishbein and brand preference: a reply. Journal of Marketing Research, 1972, g, 461. Bass, F. M. 8 Wilkie, W. L. A comparative analysis of attitudinal predictions of brand preference. Journal of Marketing Research, 1973, 10, 141-145. Bennett, P. D. & Harrell, G. The role of confidence in understanding and predicting buyers' attitudes and purchase intentions. Journal of Consumer Research, 1975, gg2, 119-117. Bettman, J. R., Capon, N. & Lutz, R. J. Cognitive algebra in multi- attribute attitude models. Journal of Marketing Research, 1975, lg, 151-164. (a) Bettman, J. R., Capon, N. & Lutz, R. J. Information processing in attitude formation and change. Communication Research, 1975, 2, 267-278. (b) Bettman, J. R., Capon, N. & Lutz, R. J. Multiattribute measurement models and multiattribute attitude theory: a test of construct validity. Journal of Consumer Research, 1975, 1:4, 1-15. (c) Bettman, J. R., Capon, N. & Lutz, R. J. A multimethod approach to validating multiattribute attitude models. In M. J. Schlinger (Ed.), Advances in consumer research (Vol. 2). Chicago: Association of Consumer Research, 1975. (d). Blalock, H. M. Social Statistics. (2nd ed.) New York: McGraw-Hill, 1972. Bohrnstedt, G. & Carter, T. M. Robustness in regression analysis. In H. L. Costner (Ed.), Socialogical'methodology. San Fran- cisco: Jossey-Bass, 1971. Box, G. E. P. & Anderson, S. L. Permutation theory in the derivation of robust criteria and the study of departures from assumption. Journal of the Roygl Statistical Society (Series B), 1955, 17, 1-26. Boyd, H. W., Ray, M. L. & Strong, E. C. An attitudinal framework for advertising strategy. Journal of Marketing, 1972, 66, 27-36. Brenner, D. J. Dynamics of public opinion on the Vietnam War. In S. R. Brown & D. J. Brenner (Eds.), Science, psychology and conmunication. New York: Teachers College Press, 1972. Brown, R. Social psychology. New York: Free Press, 1965. Bruning, J. L. & Kintz, B. L. Computational handbook of statistics. Glenview, 111.: Scott, Foresman and Company, 1969. 171 Bruno, A. V. 8 Wildt, A. R. Toward understanding attitude structure: a study of the complementarity of multi-attribute attitude models. Journal of Consumer Research, 1975, 2:2, 137-145. Calder, B. J. 8 Lutz, R. J. An investigation of some alternatives to the linear attitude model. In M. Venkatesan (Ed.), Proceedings of the third annual conference of the Association for Consumer Research. Chicago: Association for Consumer Research, 1972, 812-815. Campbell, 0. T. 8 Fiske, D. W. Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin, 1959, 66, 81-105. Cohen, J. B. Toward an integrated use of expectancy-value models. In G. D. Hughes 8 M. L. Ray (Eds.), Buyer/consumer information processigg. Chapel Hill: University OfXNorth Carolina’Press, 1974. Cohen, J. B., Fishbein, M. 8 Ahtola, O. T. The nature of expectancy- value models in consumer attitude research. Journal of Marketing Research, 1972, s, 456-460. Colley, R. H. Defining advertising goals for measured advertising results. New York: AssoCiatTOn of NationaliAdVertisers, 1961. Cox, D. F. Clues for advertising strategists: II. Harvard Business Review, November-December 1961, 160-182. Culley, J. D., Lazer, W. 8 Atkin, C. K. Experts look at children's television. Journal of Broadcasting, 1972, 26:1, 3-22. Douglas, 0. F., Westley, B. H. 8 Chaffee, S. H. An information campaign that changed community attitudes. Journalism Quarterly, 1970, 31, 479-487, 492. Feldman, J. M. Note on the utility of certainty weights in expectancy theory. Journal of Applied Psychology, 1974, g2, 727-730. Feldman, S. 8 Fishbein, M. Social psychological studies in voting behavior: III. Party affiliation and beliefs about candidates. American Psychologist, 1963, 16, 374-375. Fendrich, J. M. A study of the association among verbal attitudes, commitment, and overt behavior in experimental situations. Social Forces, 1969, gs, 347-355. Festinger, L. Behavioral support for opinion change. Public Opinion Quarterly, 1964, 26, 404=417. 172 Fishbein, M. An investigation of the relationships between beliefs about an object and the attitude toward that object. Human Relations, 1963, 16, 233-240. Fishbein, M. (Ed.). Readings in attitude theory and measurement. New York: Wiley, 1967. (a). Fishbein, M. Attitude and the prediction of behavior. In M. Fishbein (Ed.), Readings in attitude theory and measurement. New York: Wiley, 1967. (b) Fishbein, M. A behavior theory approach to the relations between beliefs about an object and attitude toward the object. In M. Fishbein (Ed.), Readings in attitude theory and measurement. New York: Wiley, 1967. (c) Fishbein, M. A consideration of beliefs and their role in attitude measurement. In M. Fishbein (Ed.), Readings in attitude theory and measurement. New York: Wiley, 1967. (d) Fishbein, M. 8 Ajzen,-I. Attitudes and Opinions. In Annual Review of Psychology,.1972, 26:.487-544. Fishbein, M. 8 Ajzen, I. Belief, attitude, intention and behavior. Reading, Mass.: Addison-We51ey, 1975. Fishbein, J. 8 Coombs, F. 5. Basis for decision: an attitudinal analysis of voting behavior. Journal of Applied Social Psycholggy, 1974, 5, 95-124. Foley, J. M. Ascertaining ascertainment: impact of the FCC rimer on TV renewal applications. Journal of Broadcasting. 1 72, 16:4, 387-406. Fruchter, B. Introduction to factor analysis. Princeton: Van Nostrand, 1954. Gardner, 0. M. Deception in advertising: a conceptual approach. Journal of Marketing, 1975, 62, 40-46. Gilbert, G. M. Stereotype persistence and change among college students. Journal of Abnormal Social Psychology, 1951._fl§. 245-254. Glass, G. V. 8 Stanley, J. C. Statistical methods in education and psycholggy. Englewood Cliffs, N.JT: Prentice4Hall, Inc., 1970. Guttman, L. A basis for scaling qualitative data. American Sociologjr cal Review, 1944, s, 139—150. 173 Haley, R. I. Benefit segmentation: a decision—oriented research tool. Journal of Marketing, 1968, 66, 30-35. Hansen, F. Consumer choice behavior: an experimental approach. Jour- nal of Marketing Research, 1969, 6, 436-443. Harvey, O. J., Hunt, 0. E., 8 Schroeder, H. M. Conceptual systems and personality organization. New York: Wiley, 1961. Hays, W. L. Statistics for the social sciences (2nd ed.). New York: Holt, Rinehart 8 Winston, Inc., 1973. Heeler, R. M., Kearney, M. J. 8 Mehaffey, B. J. Modeling supermarket product selection. Journal of Marketing Research, 1973, 16, 34-37. Hiett, R., Youngren, H., Freund, 0., Kennerly, J., Schanilec, W., Wong, H. T. 8 Rucher, B. W. A study of the effectiveness of gun control advertising. Journalism Quarterly, 1969, 56, 592-594. Holbrook, M. B. Comparing multiattribute attitude modesl by optimal scaling. Journal of Consumer Research, 1978, 6:1, 165-171. Holbrook, M. B. 8 Hulbert. J. M. Multi-attribute models: a comparative analysis. In M. J. Schlinger (Ed.), Advances in consumer research (Vol. 2). Chicago: Association for Consumer Research, 1975. Katona, G. 8 Strumpel, B. A new economic era. Public Opinion, March- April 1978, 9-11. Katz, D. The functional approach to the study of attitudes. Public ijnion Quarterly, 1960, 24, 163-204. Katz, D. 8 Braly, K. W. Racial stereotypes of 100 college students. Journal of Abnormal Social Psychology, 1933, 26, 280~290. Katz, 0. 8 Stotland, E. A preliminary statement of a theory of attitude structure and chan e. In S. Koch (Ed.), Psychology: a study of science (Vol. 3 . New York: McGraw-Hill, 1959. Kerlinger, F. N. Foundations of behavioral research. New York: Holt, Rinehart, and Winston, Inc., 1973, pp 309-311. Kotler, P. Marketing_management: analysis, planning, and control. Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1967. Krech, D., Crutchfield 8 Ballachey, E. L. Individual in society. New York: McGraw—Hill, 1962. 174 Krugman, H. New studies of brain functioning show need for marketing strategy revisions: Krugman. Marketing News, March 25, 1977, pp. 1; 7. Labovitz, S. Some observations on measurement and statistics. Social Forces, 1967, 46, 151-160. Labovitz, S. The assignment of numbers to rank order categories. American Sociological Review, 1970, 66, 515-524. LaPiere, R. T. Attitudes vs. actions. Social Forces, 1934, 16, 230- 237. Lehmann, D. R. Television show preference: applications of a choice model. Journal of Marketing Research, 1971, 6, 47-55. Lewin, K. The conceptual representation and measurement of psychologi- cal forces. Durham, N.C.: Duke University—Press, 1938. Liebert, R. M. 8 Schwartzberg, N. S. Effects of mass media. In Annual Review of Psychology, 1977, 66, 141-173. Likert, R. A. A technique for the measurement of attitudes. Archives of Psychology, 1932, No. 140. Lutz, R. J. An experimental investigation of causal relations among cognitions, affect, and behavioral intention. Journal of Consumer Research, 1977, 6:4, 197-208. Lutz, R. J. 8 Bettman, J. R. Multiattribute models in marketing: a bicentennial review. In A. G. Woodside, J. N. Sheth 8 P. 0. Bennett (Eds.), Consumer and industrial buying behavior. New York: Elsevier North-Holland, Inc., 1977. Mandler, G. Verbal Learning. In T. M. Newcomb (Ed.), New directions in Psychology (Vol. 3). New York: Holt, 1967. Mann, J. J. The relationship between cognitive, behavioral, and affective aspects of racial prejudice. Journal of Social Psychology, 1959, 42, 223-228. Markin, R. J. Consumer behavior: a cognitive orientation. New York: Macmillan Pub. Co., 1974. Mauldin, C. R. Personal communication, June 15, 1978. Mauldin, C. R., Sutherland, J. C. 8 Hofmeister, J. F. Operant atti- tude segmentation and marketing decisions. Operant subjec— tivity, 1978, 1:2, 38-50. 175 Mauro, J. B. 8 Weaver, 0. H. Patterns of newspaper readership. American Newspaper Publishers Association Research Report, July 22, 1977. Mazis, M. B., Ahtola, O. T. 8 Klippel, R. E. A comparison of four multi-attribute models in the prediction of consumer attitudes. Journal of Consumer Research, 1975, 6:1, 38-52. McDougal, W. An introduction to social psychology. London: Methuen and Co.,TLtdL, 1908. McGuire, W. J. Attitudes and opinions. In Annual Review of Psychology, 1966, 11, 475-515. McGuire, W. J. The nature of attitudes and attitude change. In G. Lindzey 8 E. Aronson (Eds.), The handbook of socialppsy- chology (2nd ed.). Vol. 3. Reading, Mass.: Addison-Wesley, 1968. McKitterick, J. B. What is the marketing management concept? In B. M. Enis 8 K. K. Cox (Eds.), Marketing classics. Boston: Allyn 8 Bacon, Inc., 1969. McSweeney, M. Notes, exercises and diagnostic measures for education 969 B. Department of Education. ‘MiChigan State University, 1976. Meeske, M. D. 8 Handberg, R., Jr. News director's attitudes toward the fairness doctrine. Journalism Quarterly, 1976, 66:1, 126-128. Merton, R. K., Fiske, M. 8 Kendall, P. L. The focused interview. Glencoe, 111.: Free Press, 1956. Miller, G. A. The magical number seven; plus or minus two: some limits on our capacity for processing information. Psychologi- cal Review, 1956, 66, 81-97. Moinpour, R. and Wiley, J. B.~ An approach to the resolution of multicollinearity in the attribute structure of attitudes. In M. Venkatesan (Ed.), Proceedings of the third annual conference of the Association for CfinsumerlRESearEfi. Chicago: AssoEiation for Consumer Researdh:71972. Monaghan, J., Plummer, J. T., Rarich, D. L. 8 Wiliams, D. A. Predict- ing viewer preference for new TV program concepts. Journal of Broadcasting, 1974, 16:2, 131-142. Myers, J. H. 8 Alpert, M. I. Determinant buying attitudes: meaing and measurement. Journal of Marketing, 1968, 66:4, 13-20. 176 Nakanishi, M. 8 Bettman, J. R. Attitude models revisited: an individual level analysis. Journal of Consumer Research, 1974, 1:1, 16-21. Nielson, R. P. A generalized attitude model for television programs. Journal of Broadcasting, 1974, 16:2, 153-160. Newcomb, T. M., Turner, R. H. 8 Converse, P. E. Social psychology, New York: Holt, Rinehart, and Winston, 1965. Newspaper Advertising Bureau. Identifying prospects for the daily newspaper: frequent readers, infrequent readers. New York: Newspaper Advertising Bureau Inc., July 1978. Nie, N. H., Hull, C. H., Jenkins, J. G., Steinbrenner, K. 8 Bent, D. H. Statistical package for the social sciences (2nd ed.). New York: McGraw-Hill, 1975. Nunnally, J. C. Psychometric theory, New York: McGraw-Hill, 1967. O'Keefe, M. T. The anti—smoking commercials: a study of television's impact on behavior. Public Qpinion Quarterly, 1971, 66, 242- 248. O'Keefe, G. J., Jr., 8 Mendelsohn, H. Voter selectivity, partisanship, adn the challenge of Watergate. Communication Research, 1974, _1_: 345-367 . Osgood, C. E., Suci, G. J. 8 Tannenbaum, P. H. The measurement of meaning. Urbana: University of Illinois—Press, 1957. Peak, H. Attitude and motivation. In M. R. Jones (Ed.). Nebraska symposium on motivation. Lincoln: University of NeErasEa Press, 1955. Pearson, E. S. The analysis of variance in cases of non-normal variation. Biometrika, 1931, 66, 114-133. Percy, L. How market segmentation guides advertising strategy. Journal of Advertising Research, 1976, 16:5, 11-22. Primer on ascertainment of community problems, 1972, 27 FCC 2d 650. Report of the Commission on Obscenity and Pornography. Washington, D.C: GPO. 1970. Robinson, J. P. Perceived media bias and the 1968 vote: can the media affect behavior at all? Journalism Quarterly, 1972, 56, 239- 246. 177 Robinson, J. P. Public opinion during the Watergate crisis. Communi- cation Research, 1974, 1, 391-405. Rokeach, M. Open and closed mind. New York: Basic Books, 1960. Rokeach, M. Beliefs, attitudes, and values. San Francisco: Jossey- Bass, 1968. Rosenberg, M. Cognitive structure and attitudinal effect. Journal of Abnormal Social Psychology, 1956, 66, 367-372. Sarnoff, I. Psychoanalytic theory and social attitudes. Public Opinion Quarterly, 1960, 66, 251-279. Say, J-B. Treatise on political economy. London: Longman, Hurst, Rees, Orme, and Brown, 1921. Schmidt, F. L. 8 Wilson, T. Expectancy value models of attitude measurement: a measurement problem. Journal of Marketing Research, 1975, 12, 366-368. Scheffé, H. The analysis of variance. New York: John Wiley, 1959. Schlinger, M. J. Cues on Q-technique. Journal of Advertising Re- search, 1969, 6:3, 53-60. Schweitzer, J. C. Newspaper Reading Interests of the Young. Paper presented at Association of Educators in Journalism Confer- ence, Madison, Wisconsin, August 1977. Scott, J. E. 8 Bennett, P. 0. Cognitive models of attitude structure: value importance is important. Proceedings, fall conference, American Marketing Association, 1971. Selltiz, C., Wrightsman, L. S. 8 Cook, S. W. Research methods in social relations. New York: Holt, Rinehart and Winston, 1976. Sherif, C. W., Sherif, M. 8 Nebergall, R. E. Attitude and attitude change. Philadelphia: Saunders, 1965. Sheth, J. N. Reply to comments on the nature and uses of expectancy— value models in consumer attitude research. Journal of Marketing_Research, 1972, 6, 462-465. Smith, A. An inquiry into the nature and causes of the wealth of nations (177611 New York: Modern Library, 1937} Smith, W. R. Product differentiation and market segmentation as al- ternative marketing strategies. Journal of Marketing, 1956, 21, 3-8. 178 Smith, M. B., Bruner, J. S. 8 White, R. W. Opinions and personality. New York: Wiley, 1956. Stephenson, W. The study of behavior. Chicago: University of Chicago Press, 1953. Stephenson, W. Public images of public utilities. Journal of Adver- tising Research, 1963, 6:4, 34-39. Stephenson, W. The ludenic theory of newsreading. Journalism Quarterly, 1964, 31, 367-374. Stone, V. A. Attitudes toward television newswomen. Journal of Broadcasting, 1974, 16:1, 49-62. Sutherland, J. C. Attitude segmentation of the Greenville Daily News market, unpublished paper, Michigan State University, 1977. Swanson, C. E. What they read in 130 daily newspapers. Journalism Quarterly, 1955, 66, 411-421. Talarzyk, W. A reply to the response to Bass, Talarzyk 8 Sheth. Journal of Marketing Research, 1972, 9, 465-467. Thurstone, L. L. 8 Chave, E. J. The measurement of attitudes. Chicago: University of Chicago Press, 1929. Television andrgrowing up: the impact of televised violence. Report to the Surgeon General, United States Public Health Service. Washington, D. C.: GPO, 1972. Tittle, C. R. 8 Hill, R. J. Attitude measurement and prediction of behavior: an evaluation of conditions and measurement tech- niques. Sociometry, 1967, 66, 199-213. Triandis, H. C. Attitude and attitude change, New York: Wiley, 1971. Tuck, M. Fishbein theory and the Bass-Talarzyk problem. Journal of MarketinLResearch, 1973, 16, 345-348. Tuncalp, S. 8 Sheth, J. N. Prediction of attitudes: a comparative study of the Rosenberg, Fishbein and Sheth models. In M. J. Schlinger (Ed.), Advances in consumer research (Vol. 2). Chicago: Association for Consumer Research, 1975. Twedt, 0. Some practical applications of 'heavy-half' theory. In J. F. Engel, H. F. Fiorillo 8 M. A. Cayley (Eds.), Market segmentation: concepts and applications. New York: Holt, Rinehart. and Winston, Inc., 1972. 179 Udel, J. C. Can attitude measurement predict consumer behavior? Journal of Marketing Research, 1965, 66, 46-50. Ward, S. Children's reactions to commercials. Journal of Advertising Research, 1972, 16, 38. Wells, W. 0. Personal communication, January 30, 1979. Wilkie, W. L. 8 Pessemier, E. A. Issues in marketing's use of multi- attribute attitude models. Journal of Marketing Research, 1973. 19, 428-441. Winer, B. J. Statistical principles in experimental design (2nd ed.). New York: McGraw-Hill, 1971. Woodworth, R. S. 8 Schlosberg, H. Experimental psychology, New York: Holt, 1954. Youngrpeople and newspapers. New York: Yankelovich, Skelly 8 White, 1976.