Role of Attribute Generality in Cognitions of Political Candidates

Glenn S. Omura, Michigan State University
ABSTRACT - Following empirical developments in the study of the attribution process and voter decision making, it was proposed that image-related attributes would be more generally used than issue-related attributes to describe political candidates, but that the generality of the two types of attributes would vary by the level of office, voter party membership, and partisan or nonpartisan voting behavior. The proposal was partially supported and implications to cognitive structure theory and political marketing were given.
[ to cite ]:
Glenn S. Omura (1979) ,"Role of Attribute Generality in Cognitions of Political Candidates", in NA - Advances in Consumer Research Volume 06, eds. William L. Wilkie, Ann Abor, MI : Association for Consumer Research, Pages: 635-640.

Advances in Consumer Research Volume 6, 1979      Pages 635-640

ROLE OF ATTRIBUTE GENERALITY IN COGNITIONS OF POLITICAL CANDIDATES

Glenn S. Omura, Michigan State University

ABSTRACT -

Following empirical developments in the study of the attribution process and voter decision making, it was proposed that image-related attributes would be more generally used than issue-related attributes to describe political candidates, but that the generality of the two types of attributes would vary by the level of office, voter party membership, and partisan or nonpartisan voting behavior. The proposal was partially supported and implications to cognitive structure theory and political marketing were given.

INTRODUCTION AND BACKGROUND

Although political party affiliation is a major determinant of voter choice of a candidate (Campbell, 1960: 121), there is little doubt that a candidate's image and position on issues are also important voting determinants. For example, in the election of Eisenhower in 1956, and Nixon in 1972, despite the fact that over half of the voters were Democrats, both Republican presidential candidates won their elections (Agranoff, 1970: 33). The basis for the above results transcends party affiliation congruency between voters and a candidate. The purpose of this paper is to explore the theoretical roles of a candidate's image and a candidate's position on given issues in voter cognitions of the candidate.

The candidate's set of image attributes as voter determinants first attracted serious study during the 1950's, when the Eisenhower versus Stevenson campaign was researched. The positive image of Eisenhower was found to lead to greater likelihood of voting for him across party lines (Campbell et al., 1954: 140). On the other hand, the Humphrey versus Nixon versus Wallace election in 1968, was found to be based on issue voting behavior among Wallace supporters, but party affiliation drew votes for Humphrey and Nixon (Converse et al., 1969: 1084). Other studies show similar mixed results with regard to the relative influence of issues and images in voter choice of political candidates (Agranoff, 1970).

Although party affiliation of candidates and voters are relatively fixed, the image and issues position of candidates are not. Both the image appeal and the issues position of a candidate ideally should be congruent with those desired by a plurality of the voting public. The image of a candidate ideally should match the cognitive creation based on voter attitudes, beliefs, and values. The issue profile of a candidate ideally should match the position of the voters. Also, the image and issues strategies of a candidate must be coordinated and ought not be considered mutually independent. For example, in the Johnson versus Goldwater election, it was found that strictly on issues, Johnson was overwhelmingly favored, but his wheeler-dealer style and earlier conflict of interest charges lowered his advantage. Simultaneously, Goldwater's integrity and sincerity benefited this latter candidate, while his stand on social welfare, domestic, and foreign policy issues eroded the number of his supporters (Converse et al., 1964: 331).

It is difficult to suggest more specific conclusions because past research has been relatively unsystematic. For example, the conclusions mentioned above dealt only with presidential contests. If research were to be compared across different levels of office, then stronger statements on the relative effects of image and issues might be made. A recent study by O'Keefe et al., (1976) provided some data to clarify the influence of the candidate's position on the issues and the candidate's image with respect to level of office. By asking voters what was their major reason for selecting a candidate, it was found that in a presidential race, issues were mentioned more often than image characteristics. However, there was little difference between the percentage of those mentioning issues versus image reasons for the gubernatorial or senatorial races. Although situational variables may well account for the above differences, an inference may be that in higher level offices, issues are a more determinant set than image in voter decisions. Accordingly, image may be a more determinant set in lower offices. Unfortunately, the study cited above is somewhat confounded since the authors drew conclusions from the 1972 presidential race versus the 1974 gubernatorial and senatorial races.

In apparent contrast to the previous study, another recent study found that the set of image characteristics appear to be a more crucial factor than the set of issue positions in a presidential state primary election (Williams et al., 1976). Williams et al. Suggest that the reason for the greater importance of image variables is that although voters may disagree with a candidate on a few specific issues, a positive evaluation of that candidate on more general image characteristics implies a candidate's favorable performance across many situations. The apparent difference in conclusions between the O'Keefe et al. and Williams et al. studies may have been due to the fact that the former study dealt with the final election, whereas the latter study dealt with a primary election. Primaries and lower level offices may tend to create the situation which elicits voter use of general image attributes rather than specific issue attributes, possibly because of Low involvement. It has been suggested by others that the degree of voter involvement varies directly with the level of political office being contested (Krugman, 1965; Yalch, 1975). Thus, the two studies may in fact have reported comparable conclusions. It was the conclusion by Williams et al. that image attributes may be more general than issue attributes which served as the impetus to this study.

THEORY AND HYPOTHESES

The domain of attributes relevant to the voter consists of those which the voter may ascribe or attribute to some candidate. This notion is referred to by some as differentiation. Wegner (1977) defines differentiation as "the total number of attributes given by the observer [to describe an actor]." This definition is comparable to those given by others; for example, Schroder et al. (1967) define differentiation as the "number of elementary dimensions in a complex cognitive structure (p. 165)." It has been noted that a characteristic of those attributes in the differentiation domain that should be of concern in cognitive structure studies is generality. General attributes are those which are used to describe a variety of actors; where the greater the frequency of attributions, the more general the attribute. General attributes, distinct from specific attributes, have been found to be more abstract, better defined (more stable) (Wegner and Buldain, 1977), and possess more discriminating power (Wegner, 1975). Regarding the last characteristic mentioned, Schroder et al. (1967) refer to the "capacity of a dimension to distinguish stimuli" as discrimination (p. 174). They note further that one condition of an abstract attribute is the "presence of many finely separated categories on the dimension, in terms of which many aspects of stimuli can be discriminated (p. 175). To researchers of cognitive structure, they suggest a measure of discrimination developed by Scott (1962), which Wegner (1975) later used to determine that general attributes had more discriminating power. [Actually, Wegner used a more recent measure of discrimination developed by Scott (1969).]

It is apparent from the above discussion that attribute generality, abstractness, and discrimination are positively related. This relationship provides a theoretical basis of support for Williams et al. (1976). Referring to the apparent generality of image attributes, they state that "perhaps voters approach candidate selection from an abstract perspective, making greatest use of such general attributions (p. 47)." The focus of this study was with attribute generality, relating to the suggestion by Williams et al. that perhaps image attributes are more general than issue position attributes of a political candidate. Therefore, hypotheses were developed to test whether image and issue attributes of political candidates were distinguishable on the basis of their generality. If the theoretical nature of image and issue attributes could be clarified, that is, in terms of their generality, it was felt that voter cognitions of political candidates could be better understood.

Before the hypotheses could be formally developed, however, several considerations had to be taken into account to reflect theory, past research, and the practice of political marketing. First, since the occurrence of general attributes is a function of all the attributes which may be used, both attribute characteristics of differentiation and generality had to be accounted for in the hypotheses. Second, as noted earlier, the frequency of reported usage of image versus issue attributes varied by the level of public office. Thus, level of office had to be treated as a potentially important factor. Third, as noted earlier, large numbers of voters tend to vote along party lines. Hence, party membership had to be treated as a potentially important factor, although no theoretical link between party membership and use of general attributes is immediately apparent. Also, since partisan voting is historically inconsistent, potential interacting effects between level of office and party membership had to be taken into account. Further, since partisan voting is inconsistent and there are large numbers of voters registered "independent," it was felt that understanding of voter behavior could be enhanced if voters were categorized by whether their voting behavior was partisan, non-partisan, or independent, and use of general attributes by the voter categories subsequently explored. Hence, candidates who attempt to insure partisan voting, non-partisan voting, or wish to attract independents in their campaign strategy will conceivably benefit by appropriately varying the generality of attribute descriptions. The following research hypotheses were developed to reflect the previous discussion and the above considerations. [The hypotheses were given alternative wordings since from a theoretical perspective, the wordings were dependent on the results of H1. Image attributes were to be termed general attributes only if H1 were given empirical support.]

H1: Frequency of usage of image attributes is greater than issue attributes at each level of office; i.e., image attributes are more general than issue attributes.

H2: Frequency of usage of image attributes is greater, relative to issue attributes, at lower level offices than at higher level offices; i.e., usage of general attributes decreases with increasing level of 3 office relative to usage of specific attributes. [Note that H2 is not inconsistent with H1.]

H3: Relative frequency of usage of image/issue attributes varies with party affiliation and level of office; i.e., usage of general relative to specific attributes varies with party affiliation and level of office.

H4: Relative frequency of usage of image/issue attributes varies with patterns of voting behavior (partisan voting, crossover or non-partisan voting, and independent voting) and level of office; i.e., usage of general relative to specific attributes varies with patterns of voting and level of office.

METHODOLOGY

Operational Definitions

The approach taken to operationally define attribute generality (or specificity) was a modification of a method used in previous research (Wegner, 1977; Wegner and Buldain, 1977). The original approach was based on free response elicitation of actor attributes followed by a content analysis of the responses with attributes as the unit of analysis. In order to get a measure of a subject's use of general attributes, Wegner (1977) used a generality ratio, which was the number of attributes a subject used to describe more than one actor divided by the total number of different attributes the subject used across all given actors. Since the focus of this study was on the generality of image versus issue attributes, this concept of a generality ratio was modified in order to provide a measure of the generality of a group of attributes rather than of subjects. The operational definition used for the generality of attributes was

Gi = gi/Di

where:

Gi = generality score for attribute category i (image or issue).

gi = number of different attributes in attribute category i used by a voter to describe all given political candidates.

Di = differentiation score or the number of different attributes in attribute category i used by a voter to describe at least one of the given candidates.

An attribute was "used by a voter to describe" a candidate if the voter could indicate a positive or negative assignment of that attribute. If a voter could not or did not assign an attribute to a candidate, then it was assumed that that attribute was not used to develop a cognition of the candidate.

The above operationalization provided an overall measure of the generality of image and issue attributes. Since the level of office was expected to have an effect on the use of general attributes, the above definition was modified to reflect level of office. The generality definition used for the hypotheses testing was

Gij = gij/Dij

where:

Gij = generality score for attribute category i for public office j.

gij = number of different attributes in attribute category i used by a voter to describe both given political candidates contesting for office J.

Dij = differentiation score or the number of different attributes in attribute category i used by a voter to describe at least one of the candidates in office j.

Procedure and Data Collection Instrument

The data necessary to examine the hypotheses were gathered in September, 1976. Mail questionnaires were sent to a randomly selected group of registered voters in one voting district of a Boston, Massachusetts suburb. A return rate of approximately 30% yielded 335 usable questionnaires.

Voting intentions were determined by asking, "If today was election day, who would you vote for?" A number of candidates running for three offices in the selected voting district were listed. The offices were State Senator, U.S. Senator, and U.S. President. These offices were deemed to be sufficiently important to interest and involve voters, but cover a wide enough range where differences as hypothesized between levels of office would appear. [This assumption is later tested.] Opposite each name, the respondent was asked to circle "YES" or "NO". No "undecided" alternative was given, but if a respondent failed to circle a "yes" or "no", it was assumed that the respondent was undecided and so classified. The names of the candidates listed that are of relevance to this study are shown in Table 1. The respondents were given no indication of the contested office or party affiliation relevant to each candidate and the candidates were presented alphabetically without regard to office sought. Voters were also asked to report their political party affiliation.

TABLE 1

LIST OF OFFICES AND CANDIDATES

Use of an attribute by a voter to describe a candidate was determined by listing in rough alphabetical order, image and issue attributes and asking the respondent whether each specific candidate possessed the attributes. The respondent simply indicated "TRUE", "FALSE", or "DON'T KNOW". The number of "true" or "false" indications was the g value in the generality score. A "don't know" report was assumed to indicate that the respondent did not use the respective attribute in the cognitive representation of the candidate. A total of 6 image and 21 issue attributes were originally selected from the descriptions of various candidates given by the local print media and candidate promotional print brochures. In order to reduce the possibility of including trivial attributes and to balance the number of image and issue attributes in the computation of the generality scores, a decision rule was developed to select a subset of the 27 attributes. An attribute was included only if at least one-third of the voter sample responded "true" or "false" to an attribute on at least one of the two candidates in each office. The figure of one-third was arrived at by examining the raw data and determining what figure was needed to balance the number of image and issue attributes. Hence, six image and six issue attributes were finally selected and are described in Table 2.

TABLE 2

LIST OF IMAGE AND ISSUE ATTRIBUTES

RESULTS

The first hypothesis predicted that image attributes would be found to be more general than issue attributes. In order to test this hypothesis, generality scores were computed for both categories of attributes and a series of paired sample t tests conducted. Table 3 displays the results. Regardless whether the data were collapsed across the level of offices, or each office was examined independently of one another, the first hypothesis was supported. The generality scores for image attributes are significantly (p < .01) larger than issue attributes.

TABLE 3

COMPARISON OF GENERALITY SCORES

The second hypothesis predicted that general attributes would decrease in usage with increasing levels of office relative to usage of specific attributes. This and remaining hypotheses were tested by computing a difference score from the two generality scores, that is, the dependent variable became

Gj' = Gimage,j - Gissue,j

where:

Gj = difference score in public office j.

Gimage,j = generality score for image attributes in public office j.

Gissue,j = generality score for issue attributes in public office j.

Since H1 was supported, Gj was correspondingly positive. The closer the value of Gj to zero, the closer the generality of image and issue attributes approach equality. The larger the value of Gj, the more general were image attributes relative to issue attributes for a given level of office. This difference score reflects the hypotheses, where it is the use of the two types of attributes relative to one another in a given situation that is of concern. The statistical analysis took the form of an analysis of variance 3 x 3 within subjects repeated measures design. The two factors consisted of political party affiliation (democrat, independent, and republican) and level of office (State Senator, U.S. Senator, U.S. President). Thus, H3, which predicted that the use of general attributes relative to specific attributes would vary by party affiliation and level of office, could be tested in the same paradigm and analysis as H2. In fact, since it was predicted that party affiliation and level of office would have significant interaction effects, the particular ANOVA paradigm described above was required.

A significant main effect for level of office was found (2, 658) = 10.053, p < .01], as well as a significant interaction effect [F (4, 658} = 3.159, p < .05 ]. [Three subjects had to be removed from the ANOVA due to data processing limitations.] The main effect for party affiliation was not significant. Figure 1 describes the nature of the level of office effect and its interaction with party affiliation.

FIGURE 1

DIFFERENCES IN GENERALITY BY PARTY AFFILIATION AND LEVEL OF OFFICE

The interaction effect between level of office and party affiliation complicated conclusions about H2. From Figure 1, it appeared that voters with a democratic affiliation were more likely to use issues as general attributes for the U.S. Senator level of office than voters with a republican or independent affiliation. Further, from Figure 1, it was evident that the graphs did not clearly suggest a decreasing function, whereas H2 predicted a monotonically decreasing function. Hence, H2 is not supported. On the other hand, H3 was partially supported since the generality scores varied with a combined (interaction) effect by party affiliation and level of office.

Since H2 was not supported, it was felt that perhaps the earlier assumption of increasing voter involvement with increasing levels of office was unwarranted. To ascertain the relationship between involvement and level of office, a simple repeated measures analysis of variance was computed for a dependent variable of "interest" and a treatment variable of levels of office. [This study is part of a larger project to study political advertising effects. The project consists of measurements over three time periods. The data used for this study came from the first questionnaire, uncontaminated by advertising treatments. However, the project allowed for several covariate measures. Hence, an "interest" variable was available for inclusion in this study.] The interest variable came from a question in the aforementioned questionnaire which asked, "How much interest do you have in the election for the following offices?" All relevant offices were listed along with a five-point response scale described by "Very Interested" to "Very Uninterested". Table 4 displays the mean interest ratings for the three offices. The analysis of variance provided some support for the earlier assumption that involvement increases with level of office [F(2, 334) = 121.92, p < .01]. Hence, it was still unclear as to why H2 was not supported.

TABLE 4

MEAN INTEREST LEVEL BY OFFICE

The final hypothesis predicted that patterns of voting and level of office could be associated with the use of general attributes. [It is evident that the theoretical approach given in this paper implies that use of general attributes describes part of the cognitive structure of the voter. The analysis approach given for H4 implies that voting behavior affects the use general attributes, where, in fact, no conclusion is possible on the causal direction. This analysis is therefore intended to be exploratory.] To test the hypothesis, a 3 x 3 analysis of variance within subjects repeated measures design was used. The two factors consisted of patterns of voting (partisan voting, crossover or non-partisan voting, and independent voting) and the same levels of office noted earlier. Both main effects and the interaction effect were significant. The voting pattern effect was significant at p < .05 [F(2, 329) = 4.184] while both the level of office [F(2, 658) = 12.686] and the interaction effect [F(4, 658 = 3.406] were significant at p < .01. Figure 2 displays the results. While H4 was supported, again, because of a significant interaction effect, conclusions were not clear. The same general shape of the curves were evident when Figure 2 is compared with Figure 1. Crossover voters perhaps tended to use less generalized image attributes at lower level offices than partisan or independent voters. As was the case for H3, the voters perhaps used more image attributes than general attributes at the U.S. Senator office level relative to the other two offices.

FIGURE 2

DIFFERENCES IN GENERALITY BY VOTING PATTERNS AND LEVEL OF OFFICE

DISCUSSION

The hypotheses tested were given partial support. It was determined that image attributes were more general than issue attributes. It was found that the level of office affected use of image attributes as general attributes, but that interaction with the voter's party affiliation left conclusions somewhat ambiguous. Party affiliation per se did not have an effect on use of general attributes. Finally, the voting pattern of the respondents was found to vary with use of generalized attributes, but that interaction with the level of office again left conclusions unclear.

While others have suggested that general attributes are more stable and enduring, it is evident that situational considerations mediate attribute generality. In this study, for example, the level of office on which a voter must make a voting decision affected use of image versus issue attributes as general attributes. Thus, the difference in past research regarding the determinance of image attributes relative to issue attributes can be explained at least partially on the basis of the office being studied. The voting pattern results also suggest that there possibly are environmental considerations (e.g., intensity of media coverage, candidate promotion, and involvement of non-partisan advocate groups) which mediate voter generalization of attributes. Herein lies a limitation to this study (and similar survey studies): a lack of control over the immediate environment. Thus, while it may be tempting to recommend a strategy of use or relative use of image attributes because of their generality, the results of this study could be biased by the peculiarities of the voting district used for sampling. Hence comparable research is needed in contrasting voting districts.

IMPLICATIONS

There has been much attention on the methodological aspects of cognitive and attribute modeling, e.g., ascertaining a summative or averaging cognitive procedure of processing attributes (Troutman and Shanteau, 1976); inclusion or exclusion of the evaluative aspect of the adapted Fishbein attitude model (Sheth and Talarzyk, 1972; Cohen and Athola, 1971); individual or group level of analysis (Nakanishi and Bettman, 1974); and inclusion of all or only determinant attributes (Alpert, 1972). [See Wilkie and Pessemier (1973) for additional issues.] Selection of object attributes to label cognitive dimensions has basically evolved to the use of direct importance ratings, indirect multidimensional scaling approaches, statistical significance of belief ratings (as in regression analyses), and/or determinance (Wilkie and Pessemier, 1973; Myers and Alpert, 1968). Thus, many researchers have been concerned with whatever attributes which have been empirically proven to have the greatest weight and/ or determinance without great attention to the nature of the attributes. Geistfeld et al. (1977) noted that in the past "nearly anything that a consumer perceives about a product may qualify as a product characteristic." As a more theoretically disciplined alternative, Geistfeld et al. suggested a hierarchy of product attributes with the hierarchy based on the degree of abstractness of a given product's attributes. Whether it is abstractness or the related attribute characteristic of generality focused on in this study, if theories are to be tested which involve object or person attributes, it seems theoretically prudent that researchers ought to pay more attention to the theoretical nature of attributes.

REFERENCES

Robert Agranoff, The Management of Election Campaigns, Boston: Holbrook Press, 1970.

Mark I. Alpert, "Identification of Determinant Attributes: A Comparison of Methods," Journal of Marketing Research, Vol. 8, May 1972, 184-191.

Angus Campbell, Philip E. Converse, Warren E. Miller, and Donald E. Stokes, The American Voter, New York: John Wiley, 1960.

Angus Campbell, Gerald Gurin, and Warren E. Miller, The Voter Decides, Evanston, Ill: Row, Peterson, 1954.

J. B. Cohen and O. T. Athola, "An Expectany X Value Analysis of the Relationship Between Consumer Attitudes and Behavior," Proceedings, Association for Consumer Research, 1971, 344-64.

Phillip E. Converse, Aage R. Clauson, and Warren E. Miller, "Electoral Myth and Reality: The 1964 Election," American Political Science Review, Vol. 59, 1965, 321-36.

Phillip E. Converse, Warren E. Miller, Jerrold G. Rusk, and Arthur C. Wolfe, "Continuity and Change in American Politics: Parties and Issues in the 1968 Election," American Political Science Review, Vol. 63, 1969, 1083-1105.

Loren V. Geistfeld, George B. Sproles and Suzanne B. Badenhop, "The Concept and Measurement of a Hierarchy of Product Characteristics," William D. Perrault (ed.), Advances in Consumer Research, Vol. IV, 1977, 302-07.

H. E. Krugman, "The Impact of Television: Learning Without Involvement," Public Opinion Quarterly, 1965, Vol. 29, 349-356.

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

Masao Nakanishi and James R. Bettman, "Attitude Models Revisited: An Individual Level Analysis," Journal of Consumer Research, Vol. 1, June, 1974.

Garret J. O'Keefe, Harold Mendelsohn, and Jenny Liu, "Voter Decision Making 1972 and 1974," Public Opinion Quarterly, Vol. 40, Fall, 1976, 320-30.

Harold M. Schroder, Michael J. Driver, and Siegfried Streufert, Human Information Processing, NY: Holt, Rinehart, and Winston, 1967.

W. A. Scott, "Cognitive Complexity and Cognitive Flexibility," Sociometry, Vol. 25, 1962, 405-414.

W. A. Scott, "Structure of Natural Cognitions," Journal of Personality and Social Psychology, Vol. 12, 1969, 261-272.

J. N. Sheth and W. Wayne Talarzyk, "Perceived Instrumentality and Value Importance as Determinants of Attitudes," Journal of Marketing Research, Vol. 9, 1972, 6-9.

C. Michael Troutman and James Shanteau, "Do Consumers Evaluate Products by Adding or Averaging Attribute Information?" Journal of Consumer Research, Vol. 3, September, 1976, 101-06.

Daniel M. Wegner, "The Generality and Articulation of Attributes in Person Perception," Paper presented at the Meeting of the Midwestern Psychological Association, Chicago, May 1975.

Daniel M. Wegner, "Attribute Generality: The Development and Articulation of Attributes in Person Perception," Journal of Research and Personality, Vol. 11, 1977, 329-39.

Daniel M. Wegner and Roger W. Buldain, "Attribute Generality and the Inference Process in Impression Formation," Personality and Social Psychology Bulletin, 3, 1977, 91-94.

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

Daniel C. Williams, Stephen J. Weber, Gordan A Haaland, Ronald H. Mueller, and Robert E. Craig, "Voter Decisionmaking in a Primary Election: An Evaluation of Three Models of Choice," American Journal of Political Science, 20 (February, 1976), 37-49.

Richard F. Yalch, "Attribution Theory and Voter Choice" in Schlinger, Mary Jane (ed.), Advances in Consumer Research, Vol. 2, 1975, 783-791.

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