Running head: Interpersonal Images of Winning & Losing Candidates

This is a working paper. Please do not cite without permission. Contact: Stephen@rpi.edu

The analyses reported in this article were designed to address two weaknesses within the literature on candidate image: (1) lack of long range longitudinal analyses permitting comparative study of images across multiple candidates and elections and (2) a dearth of research addressing the relationship to candidate preference of candidates' interpersonal communication behavior. Our analyses reveal that a small set of perceptual measures of candidate interpersonal communication behavior were consistent, powerful predictors of voter choice during the 1984, 1988, and 1992 U. S. presidential elections. Our study bolsters previous work on political image and voter choice that has concluded that image information is more predictive of voter choice than issue information and suggests that image researchers should attend to audience perceptions of candidate's interpersonal communication style.

The study of voting, especially in American presidential elections, has a long history. Early research understood voter preference in terms of audience demographics (where voter choice was believed a function of factors such as social class, ethnicity, and religion), a perspective that gave way by mid-century to an interactive, information-based model in which the voter is seen as an active agent who forms a preference for a candidate based on available information sources. In the beginning stages of this new perspective it was thought that candidate preference was an outcome of rational voter assessments of candidate positions on salient political issues (Glass, 1985). However, by the mid 1960s researchers turned their attention from rational choice models to consider the idea of candidate ``images," a concept that refers generally to a potentially manipulable set of meanings attached by voters to a seeker or holder of a political office. Although the literature attests to considerable variety in approaches to operationalizing image, candidate image has most frequently been operationalized in one of two ways: (a) as the composite of a candidate's perceived political positions on a set of campaign issues or (b) as aspects of a candidate's perceived character or personality. As we described in an earlier article (xxxx, 1988), the latter has been by far the most frequently addressed by researchers, with large numbers of studies relating voter choice and media exposure (e.g., in televised debates, in commercials, etc.) to perceived qualities of candidates.

The analyses reported in this article were designed to address two weaknesses within this body of literature: (1) lack of long range longitudinal analyses permitting comparative study of images across multiple candidates and elections and (2) a dearth of research addressing the relationship to candidate preference of candidates' interpersonal communication behavior. This article extends earlier work (xxxxx, 1988 and xxxxx, 1990) in which we examined hypotheses relating voter choice and gender to perceptions of interpersonal communication behaviors of Ronald Reagan and Walter Mondale in the 1984 U.S. presidential election.

Long Range Analysis

Researchers have often used semantic differential ratings to assess respondents' views of candidates with respect to personality traits such as ``honesty", ``anxiety", ``reliability", ``aggressiveness", or ``sincerity" (e.g., Kaid, 1995), with respect to performance-related evaluative properties of candidates reflected in such dimensions as ``informed" and ``qualified" (e.g., Powell, 1977), and with respect to dimensions of emotional warmth and relationship feelings such as ``warm", ``friendly", ``caring", and ``involved" (e.g., Pfau,
Diedrich, Larson, and Van Winkle, 1995). Likert-type rating have also been used to assess image (e.g., Anderson and Kibler, 1978 and Kendall and Yum, 1984) as have free response interviews (cf., Miller, Attenberg, and Malanchuk, 1984) and Q-sort ratings (e.g., Bass, 1997; Marr, 1973; Nimmo and Savage, 1976). Frequently, factor analysis has been employed to locate underlying image dimensions that cut across the content of a larger number of survey items. Kaid (1995) notes that there is a history of rough correspondence in studies of the dimensionality of image as measured by semantic differentials (she contends that many studies have found between 3 and 5 similar image factors), but there has been little cross-referencing or standardization of scales across studies. This is unfortunate because standardization would permit the longitudinal study of stability and change in operant images in American political process - images that cut across candidates and elections - and in the ways audiences respond to those images at the polls. With over 30 years of image research establishing the boundaries of this area of inquiry, there have been relatively few studies of this type rendering it difficult to understand image as a cross-contextual phenomenon; indeed our twelve year, three election design is among the first to take this perspective.

Some prior studies have looked at images at more than one point in time but with few exceptions, these studies have used a before-and-after design to study short term change in images of a single candidate or of a slate of candidates at various points in a campaign (commonly, before and after a public debate or primary election) or to study the relative contributions of images and issue information at various stages in a single campaign. This pervasive characteristic of studies in the area would suggest that image is conceptualized as a phenomenon attached to particular people acting in the role of political candidates. However if we take seriously the notion that image is a set of perceived rather than actual qualities of candidates, we may inquire more abstractly into the typicality of images of winning versus losing candidates, male and female candidates, candidates for local and national office, etc. The present analysis operates at this level, examining the perceived communication qualities of winners and losers across three U.S. presidential elections. Using the same procedures for assessing image in the 1984, 1988, and 1992 elections, our study represents a search for persistent perceptual markers generalizable across candidate and election that are predictive of voter choice.

Research that has been conducted across longer spans of time and more than one candidate suggests that the evaluation process is often stable and consistent. According to Miller, Wattenberg, and Malanchuk (1984, p. 200), ``The aggregate consistency in the use of particular dimensions hints at the possibility that there are enduring, general categories or criteria that people apply in the judgments of political leaders." An intensive single case study by Baas (1979, 1997) investigated Lasswell's (1960) psychodynamic theory of political image formation in changes over a fourteen year span. Baas's studies provide support for the Lasswellian theory that political images are extensional representations of structures of evaluation constructed by individuals to cope with subjective experience in the world of interpersonal and family relationships. Bass found considerable stability in political images over the fourteen year time interval but his sample was restricted to a single case. Mansfield and Nimmo (no date) studied stability of candidate images during a twelve year period spanning three U. S. Presidential elections (1976, 1980, and 1984) using Q-sort technique and a sample of eleven members of the League of Women Voters who completed measures at all three time periods. Mansfield and Nimmo found that images of the ``ideal president" were self-consistent across all three elections.

Pfau, et al. (1995) tested an alternative perspective on the longitudinal dynamics of candidate assessment, arguing that, with unknown candidates, perceptions of relational information (that is, how caring, warm, affectionate, and trustworthy a candidate is) weigh most heavily in the candidate's initial appearance to a national audience. As the campaign progresses, however, the importance of relational information declines as voters differentiate the contestants on the basis of knowledge of their relative competence. Pfau et al.'s analysis provided support for this dynamic.

One of the advantages of a Q-methodological approach is that, in the standard practice of scaling Q stimulus items in a quasi-normal distribution (more items permitted to have mid-scale ratings, fewer to have extreme ratings) and decomposing matrices of between-sort correlation (most frequently with factor analysis), raters in effect are provided with a procedure that gives much greater weight to items that raters themselves find particularly salient (Stephen, 1985). Hence, in Q approaches to image assessment, items of no relevance to raters are relegated to the fatter mid-region of the Q distribution are thereby statistically devalued in subsequent intersort correlation. Though our assessment did not use Q methodology, the items we used were obtained from an assessment procedure expressed originally as a Q sort (xxx, 1986) and in transferring the items to a Likert-type format we contrived a means of allowing respondents to declare scale items irrelevant. We felt it was of significant value to permit respondents to declare which stimulus items were useful in the description of candidates rather than to assume that we knew that information ahead of time. We will return to this point later in our discussion of our approach to measurement.

Interpersonal Communication and Image

In contrast to trait approaches, Kaid (1977) found that interpersonal communication among voters was the single best predictor of candidate choice in pre-election polling and called for studies to be conducted to discover which aspects of interpersonal communication performance influence voter choice. Some studies (e.g., Hacker, 1995; Pfau, Diedrich, Larson, and Van Winkle, 1995; Schmitt-Beck, 1994) have targeted interpersonal communication as a mediating channel of influence in much the same way that other researchers have viewed television and other mass media as influential channels that impact the kinds of images voters form of candidates (see Hellweg, Dionisopoulos, and Kugler, 1989).

However, outside of our own studies (xxx, 1988 and xxx, 1990), virtually no other research has conceptualized candidate image as a set of interpersonal communication behaviors displayed by candidates themselves. We have found this surprising since, as we have argued previously (xxxx 1988), there are good theoretical reasons for taking a behaviorally descriptive approach, rather than attempting to operate at the level of higher order inference inherent in trait descriptions. Audiences may infer traits from observed behaviors but rules of correspondence in the process of inference that mediates between behaviors and the traits they are taken to represent are hardly invariant. When an audience concludes that a political candidate is aggressive (that is, possessed of the trait aggressiveness), it is difficult to be sure of the behavioral phenomena that led to that conclusion - such an inference may be derived from any of a variety of of underlying behaviors or may in fact not be related to behavior at all. An audience may conclude that a candidate is aggressive based on a wide range of observed behavior; indeed, different audience segments may base their common attribution of aggressiveness on altogether different behaviors. Alternatively, an audience may conclude that a candidate is aggressive solely because influential opinion leaders say so.

This not only leaves the conceptual waters murky but also places applied social scientists at a substantial disadvantage when trying to translate decades of research on the relationship between image and voter choice into practical recommendations for seekers of political office - we are not sure of the value of recommending to candidates that they ``be sincere", or act in ways that are ``trustworthy". Hence our approach has been more atomistic, measuring perceptions of particular elements of candidates' interpersonal communication behavior. Our data place us in a position to relate those perceived behaviors to choices voters make at the polls. An approach that stresses perceptions of behavior (e.g., ``smiles frequently") places us in a better position to tease apart the foundations of candidate image than would be the case had we treated at the level of higher order abstraction (e.g., ``is friendly").

Thus it is important to recognize that while our approach indeed takes a perspective on interpersonal communication and candidate image, it is the interpersonal behavior of candidates, not the interpersonal behavior of voters, with which we have been concerned. This places us outside other streams of research in the candidate image literature that incorporate a focus on interpersonal communication. While other studies might well be related to traditions of research examining gatekeeping, social networks, and communication flows, as we have elaborated elsewhere (xxx, 1988), our work is perhaps better viewed in the context of person perception or parasocial interaction research.

In previous work (xxx, 1988) we examined the relative importance of perceptions of candidate communication behavior and perceptions of candidate stands on campaign issues. Our data, taken from the 1984 Reagan/Mondale election, indicated that perceptions of communication behavior was by far the stronger predictor. We subsequently used the same data set to inquire about the relative importance of issues vs. perceptions of candidate communication behavior in the prediction of voting choices for male and female respondents. That study lent strength to a conclusion derived from Chodorow's (1978) object relations theory of gender and interpersonal orientation that while both issues and interpersonal communication behavior would be consequential for both women and men, perceptions of interpersonal communication behavior would play a greater role in the evaluative processes of women than they would for men.

Our focus in this article is on the question of consistencies in perceptions of the communication behavior of winning and losing presidential candidates. In order to address this, we repeated during the 1988 and 1992 presidential elections the procedures we employed during the 1984 election. Our analysis addressed the following research questions:

1. How consistent are respondents' perceptions of the communication characteristics of winners and losers?

2. Is it possible to differentiate winners and losers on the basis of perceived communication behavior?

3. What is the dimensional structure of perceptions of winning and losing candidates' perceived communication behavior?

4. Over the course of the three elections, what was the relative importance of perceptions of political issues versus perceptions of candidates' communication behavior in the prediction of voter preference?

In addition, we present a case study of changes in perceptions of George Bush's interpersonal communication behavior since we measured perceptions of Bush as a winning candidate in the 1988 election and as a losing candidate in 1992 election.

Sample and Procedures

The sample consisted of three groups of respondents recruited at the time of the 1984, 1988, and 1992 United States presidential elections. At each election data were collected from the same three eastern U. S. universities (one additional university was included in 1984). The total sample consisted of 1,215 respondents: 401 for the 1984 election, 563 for the 1988 election, and 251 for the 1992 election. In each case the sample contained slightly more females than males, though the ratio was consistently proportional across the three samples. For the 1984 election the sample was 52% female, for the 1988 election the sample was 54% female, and for the 1992 election the sample was 55% female.

On the day of each election, respondents were asked to complete a three-part questionnaire during regularly scheduled classes set aside for this purpose. Part 1 of the questionnaire evaluated perceptions of the interpersonal behavior of the primary presidential candidates (Reagan and Mondale in 1984, Bush and Dukakis in 1988, and Clinton, Bush, and Perot in 1992). Part 2 was a survey assessing perceptions of respondents' attitudes toward election issues and their perceptions of the candidates' stands on these issues, respondents' voting intentions, and respondents' habits of information consumption.

Procedures and measures deviated little at the three points of the study. In each instance subjects were recruited from the same university populations and completed
responses the day of the election. Measures differed only in the specific issues that were represented and of course in the presidential candidates used as targets for the interpersonal perception assessments. The sets of campaign issues, culled from pre-election analysis from major national newspapers and news magazines, were in each case the top 10 salient content issues for the election. We assessed respondents' perceptions of the candidates' position on each issue as well as the respondents' opinion of the importance of each issue. Five issues were salient across all three elections: abortion, taxes, crime, defense spending, and the environment.

In our earlier work on the 1984 election, we studied image in terms of perceptions of Walter Mondale and Ronald Reagan. For the present study, however, our interest was not in the qualities of particular candidates but rather in perceived qualities of winners and
losers. Thus, while we asked respondents in 1988 to rate Bush and Dukakis, and in 1992 to rate Clinton and Bush, these ratings were recoded following each election and the combined data set consists of a set of ratings of winners (combining ratings of Reagan, Bush, Clinton) and losers (combining ratings of Mondale, Dukakis, and Bush).

Our studies have used a novel approach to investigate relationships between perceptions of interpersonal communication. Respondents were asked to rate each candidate on a 100-item inventory of verbal and nonverbal behavior typically displayed by individuals in interpersonal settings. Items in the inventory were adapted from an instrument developed to provide detailed descriptions of interpersonal communication style (xxxx, 1986). The instrument was chosen because it provided a large and diverse sample of potentially relevant aspects of interpersonal behavior that respondents might use in constructing candidate images. Representative items include ``laughs frequently", ``listens intently and carefully", ``explains by using examples, analogies, or stories", and ``controls what gets talked about".

On the day of each of the three elections, respondents were asked to assess the extent to which each of the 100 communication style items was characteristic of the behavior of each major candidate. Respondents did this by rating the candidates on a five-point scale ranging from ``least characteristic" to ``most characteristic". The procedure was designed so that all the presidential contenders could be rated simultaneously on the same item. For example, in the 1984 election, respondents were asked to write the letter ``R" (for Reagan) or ``M" (for Mondale) in spaces that represented one of the five scale points. Respondents were explicitly instructed that they could fill in the same scale value for multiple candidates if they felt both candidates deserved the equivalent rating on the item. For the 1992 election, in addition to the candidates of the major national parties, respondents rated Ross Perot.

In addition, respondents were given the opportunity to declare any item irrelevant as a behavior in terms of which they could characterize all the candidates they were rating or a particular candidate. It seemed possible that our respondents might differ in the degree to which they felt they could evaluate candidates with respect to the various aspects of communicative interaction reflected in the inventory. This might occur because of differences in exposure to and/or perception of media representations of the two candidates. Rather than force our respondents to rate Reagan and Mondale on every item, therefore, we felt it would be a more appropriate procedure to permit them to select only those descriptions about which they felt they could make valid judgments. A consequence of this procedure was that the size of our sample of rated items (i.e., items that were not declared irrelevant) varied from item to item. We therefore selected for analysis only those items for which 80% of our respondents provided a rating other than ``irrelevant".

Candidate preference was measured by asking respondents to indicate whether they had ``definitely" made up their minds about their candidate of choice, or whether they were as yet undecided. Those who had definitely made up their minds were then asked to identify their preferred candidate (any of the principal candidates in the election or someone else, who they could write in). In subsequent analyses involving the candidate preference measure, data analysis was performed using only those respondents who had definitely made a choice between the democratic and republican party candidates. Too few subjects (n = 26 or 3.1% of the total combined sample) were definitely inclined toward Ross Perot in the 1992 election to merit including these cases in our analyses.

The focus of our investigation was first on the question of cross-election consistency in candidate image and subsequently on the question of the substance of images of winners and losers. Clearly the consistency issue was prerequisite - if there was no consistency, there would be little point in speaking of generalized images of winners and losers in American presidential politics. On the other hand, to find evidence of consistency is to begin to address the question of image content. We explored the issue of consistency in two ways. First we looked for variability across the three elections in the range of interpersonal communication items deemed relevant by subjects. Finding remarkable consistency, we then turned to an exploration of the factor structure of the interpersonal ratings across the three elections. At this point we revisited an issue we examined in the 1984 election (xxx, 1988) on the relative importance of issues and images in the prediction of voter choice. Finally we considered the case of stability and change in perceptions of George Bush, a winning candidate in 1988 and a losing candidate in 1992.

Variability in item relevance

In our analysis of the 1984 election (xxxx, 1988), we reported that after applying the 80% valid cases criterion, 36 communication style items remained for analysis. This outcome was consistent across the three elections. Table 1 presents the 36 communication style items that were deemed relevant in describing both winners and losers across the three elections by 80% of the combined sample. One additional item (5. ``Likes to tell stories or anecdotes.") was deemed relevant in describing winners but not in describing losers. With only 5 exceptions, these are the same items that we reported in the earlier study.

It is instructive to note the clear and consistent differentiation with respect to characteristics of winners and losers. Generally, winners appear to respondents to communicate in a more self contained, secure, relaxed, and interpersonally functional manner. In an interesting contrast to this trend, winners are less likely to be perceived as treating the other as an equal (item 23), somewhat less likely to admit being wrong (item 41), and slightly more likely to be perceived as giving vague answers (item 59). Otherwise, however, they are textbook studies in optimal, nondefensive interpersonal communication behavior. Next to them, losers are perceived in ways that appear to be somewhat overbearing, tense, contentious, histrionic, and serious.

As a group, ratings of these 36 loser perceptions and 37 winner perceptions do an impressive job in predicting voting preference. Entering all variables as a block in a multiple regression analysis yields a multiple R of .86 (F (73,115) = 4.49, p ^< .001), accounting for 74% of the variance (adjusted R² = .58) in the dependent variable. Even accepting the diminished adjusted R² value, this is considerably better prediction than reported typically in the literature relating candidate image to voter choice. Prediction was slightly better using winner perceptions alone (R = .79, F (37,171) = 7.42, p ^< .001) over using loser perceptions alone (R = .68, F (36,172) = 4.24, p ^< .001) but the best results were obtained by combining both sets of items. There were no significant overall sex effects in the ratings: the results reported applied equally well to males and females.

Dimensionality of candidate communication

In the next stage of our analysis we used factor procedures to probe interdependencies in the set of items that had passed the 80% criterion. Factor analyses were computed separately for items describing interpersonal communication behavior of winners and items describing interpersonal communication behavior of losers. We used principal components analysis with orthogonal rotation determining the number of factors to retain based on examination of a scree plot of diminishing variance accounted for among a larger set of potential factors with eigenvalues greater than 1. The solution for winners (see Table 2) yielded 5 factors that together accounted for 45% of the variance. The solution for perceptions of losing candidates (see Table 3) yielded 5 factors together accounting for 44% of the variance.

As Tables 2 and 3 attest, there is no consequential difference between the factor solutions for perceptions of winners and perceptions of losers. The same critical dimensions are operative in the evaluation of both types of candidates. The primary dimension (factor 1 for both winners and losers) appears to describe an attentive, thoughtful, considerate, egalitarian interaction style. A separate cluster of items (factor 2 for losers and factor 3 for winners) identifies a good natured convivial quality in interaction exemplified by laughing and smiling and sharing anecdotes. A third dimension (factor 2 for winners and factor 4 for losers) consists of a set of items that reference self confidence, social presence, and assertiveness. A fourth dimension (factor 4 for winners and factor 3 for losers) consists of a set of items that suggest aggressive verbal attack. The last dimension (factor 5 for both winners and losers) suggests an ability to speak with great volume and force and compelling gesture so as to dominate interaction with others. All in all this suggests that a successful presidential candidate requires mastery of a broad interactional range, a healthy, balanced personality, and considerable oratorical facility.

Issues versus images in candidate choice

This analysis was conducted using the same procedures used in our study of the 1984 election. Four sets of variables were created: a set of perceptions of communication behaviors of winners, a set of perceptions of communication behavior of losers, ratings on the 5 issues common across the three elections (abortion, taxes, defense spending, the environment, and crime), and a set of ratings representing respondents' judgments of the importance of each of the five issues. We wished to assess the relative contributions of each of these sets of variables. Prior to doing so we reduced the two sets of communication behavior items using multiple regression. Two regression analyses were conducted, one for the set of winner perceptions and one for the set of loser perceptions. In each case the variables were entered as a group and then the group reduced through backward elimination of variables that did not account for unique portions of the variance in candidate preference, the dependent variable.

The multiple regression analyses reduced the set of winner variables to 9 items (numbers 5, 15, 23, 36, 37, 45, 53, 59, and 89 - see Table 1) accounting for 57% (R = .75, F (9,199) = 28.7, p < .001) of the variance in voter preference and reduced the set of loser variables to 6 items (numbers 13, 15, 18, 23, 72, and 84) accounting for 40% (R = .63, F (6,202) = 22.54, p < .001) of the variance in voter preference.

When all four sets of variables were thus prepared, our main analysis was conducted using the SPSS TEST procedure, which allows one to estimate which designated subsets of items within the regression equation contribute uniquely to variance in the dependent measure, voter preference. This unique contribution is indicated by the squared semipartial correlation associated with those variable subsets one may wish to evaluate (Cohen and Cohen, 1983). In the source table reported in Table 4, the semipartial statistic is identified as ``RSQ Change."

Reiterating the results of our analysis of the 1984 campaign, political items reflecting perceptions of candidates' stands on the issues made a significant but barely perceptible contribution in an otherwise powerful prediction equation. Perceptions of issue importance did not contribute significantly. Perceptions of candidates' interpersonal communication behaviors accounted for nearly all the predicted variance in voter preference.

Stability and change in perceptions of George Bush

Finally, we turned to a consideration of the unique opportunity afforded to our analysis by the fact that George Bush was a winning candidate in 1988 but a losing candidate in 1992. We explored how perceptions of Bush changed using correlational analysis. To do so, we computed the mean ratings for each of the interpersonal communication items for Bush in 1988 and for Bush in 1992. We also computed combined mean ratings on the items for the other two winners (Reagan and Clinton) and for the other two losers (Mondale and Dukakis). We next computed correlations between these various sets of ratings. The correlation coefficients yield indexes of similarity between the sets of ratings, which can usefully be expressed as the percent of overlap by squaring the coefficients (obviously, the issue of statistical significance is not relevant in this application of correlational analysis). The resulting matrix of coefficients in presented in Table 5.

The results of the correlation analysis reveal that there is some basic similarity (roughly 16% overlap) in the mean ratings of winners and losers exclusive of ratings of George Bush. Mean ratings of Bush's interpersonal communication style in 1988 was quite similar (r = .84 or 64% overlap) to the ratings of Reagan and Clinton, the other winning candidates. Nevertheless, his 1988 ratings bore a non-trivial resemblance (r = .64 or 36% overlap) to the ratings of Mondale and Dukakis, the other losing candidates. Already somewhat similar in appearance to the other losing candidates, by 1992 that similarity increased (r = .72 or 49% overlap). More remarkably, however, George Bush's resemblance to other winning candidates dropped to a minuscule level (r = .23 or 4% overlap). In 1992 George Bush looked less like a winner than other losing candidates. Overall the analysis suggests the possibility that George Bush's position in the public eye as a successful presidential figure was precarious at the moment of his election and eroded significantly by end of the 1992 campaign.

There have been a number of good summary sources treating candidate image research that have been published since we gathered our first wave of data in 1984 (e.g., Glass, 1985; Hacker, 1995; Hellweg, Dionisopoulos, and Kugler, 1989; Miller, Wattenberg, and
Malanchuk, 1984). These sources document a wealth of research on the role of image in U. S. elections and with the modest contribution of our own studies it seems reasonable to conclude that image is a powerful and robust predictor of candidate choice and may well be the predictive factor next to which little else matters. For this reason we feel justified in making a strong recommendation that image researchers embark on an effort to standardize measures of image so that we can extend our knowledge of long term and cross-candidate effects. In our view it is most unfortunate that there have been but a handful of studies spanning multiple elections. Ironically, one of the staples of the image literature is the idea that image became an important factor in voter choice as a result of the evolution of twentieth century mass media, particularly television. But tests of this assumption and others like it require long-term cross-election analyses. While the field sustains its focus on single elections, it may miss opportunities to discover broader processes. Given the reasonably mature state of knowledge in the area of image, it seems prudent to now call for such efforts.

On the other hand, although data from many studies suggest that we know that image matters a great deal in candidate choice, we do not know much about the processes of decision making that guide those for whom image is not a factor. Accounting for 60% to 75% of the variance in voter choice is hardly inconsequential, yet 25% to 40% of the variance in voter choice remains unaccounted for and we doubt strongly that voter choice is much a random or whimsical affair. Following up on our regression analysis of the relationship of the communication style items to voter choice, we conducted a post mortem on the left over residual variance (that is, the variance in voter choice unaccounted for by the communication style items). The residual variance was correlated at the .50 (p ^< .001) level with the vote variable, on which winning candidates had been coded with lower values and losing candidates had been coded with higher values. This means that the residualized scores were greater for those who voted for losing candidates. We interpret this to mean that the perceptions of communication style items did a better job of accounting for the voting behavior of those who voted for winning candidates. What then goes on in the decision processes of those voting for losing candidates? Do they perceive candidates in different terms than those whose support ultimately elects the winner?

Pfau et al.'s (1995) study suggests that interpersonal information matters more in the early evaluation of unknown candidates, but our results support the overwhelming influence of interpersonal information in the evaluation of candidates with high media exposure the very day of the election. It remains possible, however, that variations in sampling designs used in the two studies are responsible for this difference. One clear limitation of our study is its dependence on a college sample.

Another unique factor of our study that served as both a weakness and a strength was our novel use of relevance ratings in our measure of perceived interpersonal communication. These ratings unquestionably boosted both the content and predictive validity of our measure. Using our procedure to discard items about whose relevance our respondents could not reach consensus stripped away 63 of 100, items resulting in a pool of content that was extremely powerful in its ability to predict voter choice. In a sense, we allowed our respondents to write their own measure of interpersonal communication. However, since we have now employed this procedure in three elections covering twelve years and since we have
found that the items chosen are relatively invariant across time we would recommend that future applications of the item set adopt the reduced set of items and require all respondents to rate them. Ideally a smaller auxiliary sample would be recruited and given the entire 100 item set with the usual relevance ratings to serve as a check on the possibility that there may eventually be some shift in the selected content. Assuming that no such shift occurs, eliminating the relevance ratings will greatly reduce the large sample attrition we experienced as a result of the requirement of many multivariate analysis techniques (e.g., multiple regression and factor analysis) that data are present on all measures for all subjects employed in the analysis. Use of statistical techniques to handle missing values (e.g., mean substitution) are simply unsatisfactory.

In an earlier analysis (xxx, 1988), we grounded our exploration of interpersonal communication styles of candidate images in the emerging theoretical perspective in media studies of parasocial interaction. We argued that interpersonal communication behavior is perceived by audience members and used to construct fantasies of the lives and character of media figures, especially figures with significant television exposure such as political candidates. Drawing on Caughey (1984), Goffman (1959), Meyrowitz (1985), and others, we suggested that perceptions of interpersonal communication behavior of candidates constitutes the elemental building materials for the construction of more detailed accounts of candidates held by those viewing the behavior of candidates on television. These perspectives support the idea that television exposure not only foregrounds frontstage performances of candidates (e.g., in debates, press conferences, and other formal occasions) but simultaneously leaks more intimate information that becomes the focus for construction of elaborated visions of the candidates as rounded social actors. These perceptions constitute a primary basis for candidate choice.

In reviewing the literature on candidate image, Nimmo (1995) and others suggest the possibility that voters enter the evaluation process with prefabricated notions of ideal qualities for office holders. Candidates are observed in order to assess the degree to which they correspond to an ideal type. The greater the correspondence, the greater the probability the candidate will obtain an audience member's vote. Miller, Wattenberg, and Malanchuk (1984) take a similar position, arguing that audience members reduce the effort of their decision making by applying enduring structures of evaluation, referred to as schemas, in considering candidates.

The results of this study will not resolve this issue, though they provide strong support for the idea that audiences evaluate the interpersonal behavior of candidates using criteria of enduring relevance. Whether voters bring these criteria with them to the campaign or rediscover them in successive elections we cannot say. However, our data suggest quite strongly that from the standpoint of voter preference there are five principal dimensions of relevance in the interpersonal communication behavior of U.S. presidential candidates. We feel these data represent a potential practical benefit in their specificity relative to prior studies. They suggest the possibility that continuing research in this direction will provide value in helping candidates to prepare more successful presentations of self. To the extent that this comes about, we may hasten the day when candidates mount campaigns in which issues rather than images take on primary importance.

Andersen, P. A. and Kibler, R. J. (1978). Candidate valence as a predictor of voter preferences. Human Communication Research. 5, 4-14.

Bass, Larry (1979). An intensive analysis of the interpersonal sources of the symbolic meaning of the Constitution. American Journal of Political Science. 23, 101-120.

Bass, Larry. (1997). The interpersonal sources of the development of political images: An intensive, longitudinal study. Operant Subjectivity. 20, (3/4, April/July), 117-142.

Caughey, J. L. (1984). Imaginary social worlds. Lincoln, Nebraska: University of Nebraska Press.

Chodorow, Nancy. (1978). The reproduction of mothering. Berkeley: University of California Press.

Glass, David P. (1985). Evaluating presidential candidates: Who focuses on their personal attributes? Public Opinion Quarterly. 49, (4, Winter), 517-534.

Goffman, E. (1959). The presentation of self in everyday life. Garden City, NY: Doubleday Anchor.

Hacker, Kenneth L. (1995). Interpersonal communication and the construction of candidate images. In Kenneth Hacker Candidate images in presidential elections. (pp. 65-82). Westport, CT: Praeger Publishers.

Hellweg, Susan A., Dionisopoulos, George N., and Kugler, Drew E. (1989). Political candidate image: A state-of-the-art review. Progress in Communication Sciences. 9, 43-78.

Kaid, Lynda Lee. (1977). The neglected candidate: Interpersonal communication in political campaigns. Western Journal of Speech Communication. 41, (4, Fall), 245-252.

Kaid, Lynda (1995). Measuring candidate images with semantic differentials. In Kenneth Hacker (Ed.). Candidate images in presidential elections. (pp. 131-134). Westport, CT: Praeger Publishers.

Keeter, Scott (1987). The illusion of intimacy: Television and the role of candidate personal qualities in voter choice. Public Opinion Quarterly. 51, 344-358.

Kendall, K. E. and Yum, J. O. (1984). Persuading the blue-collar voters: Issues, images, and homophily. In R. N. Bostrom (Ed.). Communication yearbook 8. (pp. 707-722). Beverely Hills, CA: Sage.

Lasswell (1960). Psychopathology and politics. NY: The Viking Press.

Mansfield, M. and Nimmo, Dan. (no date). Tracking the league: The stability of candidate images across time and presidential debates, 1976, 1980, and 1984. Unpublished manuscript.

Marr, Theodore J. (1973). Q and R analyses of panel data on political candidate image and voter communication. Speech Monographs. 40, 56-65.

Meyrowitz, J. (1985). No sense of place. NY: Oxford University Press.

Miller, Arthur H., Wattenberg, Martin P., and Malanchuk, Oksana (1984). Cognitive representations of candidate assessments. In L. Kaid and D. Nimmo (Eds.). Political communication yearbook 1984. (pp. 183-210). Carbondale, Illinois: Southern Illinois University Press.

Nimmo D. (1995). The formation of candidate images during election campaigns. In Kenneth Hacker (Ed.). Candidate images in presidential elections. (pp. 131-134). Westport, CT: Praeger Publishers.

Nimmo, D. and Savage, R. L. (1976). Candidates and their images: Concepts, methods, and findings. Pacific Palisades, CA: Goodyear.

Pfau, Michael, Diedrich, Tracy, Larson, Karla M., and Van Winkle, Kim M. (1995). Influence of communication modalities on voters' perceptions of candidates during presidential primary campaigns. Journal of Communication. 45 (1, Winter), 122-133.

Powell, L. (1977). Voting intention and the complexity of political images: A pilot study. Psychological Reports. 43, 343-347.

Schmitt-Beck, Rudiger (1994). Intermediation environments of West German and East German voters: Interpersonal communication and mass communication during the first all-German election campaign. ital European Journal of Communication. 9, 381-419.

Stephen, T. (1985). Q-methodology in communication science: An introduction.
Communication Quarterly. 33, 193-208.

Item

Winners

Losers

1.

Controls what gets talked about.

3.16

3.48

2.

Dominates others in conversation.

3.13

3.43

3.

Tells jokes frequently or injects humor into

the conversation.

3.55

2.76

4.

Laughs frequently.

3.54

2.80

10.

Has a loud voice.

2.94

3.29

12.

Smiles frequently.

4.08

3.39

13.

Explains by using examples, analogies,

or stories.

3.75

3.47

15.

Overstates ideas or exaggerates them to

emphasize a point.

3.68

3.70

18.

Gestures dramatically.

3.03

3.42

19.

Shows attention by directing his/her body

towards the listener.

3.78

3.59

20.

Uses facial expressions and/or meaningful

gestures.

3.63

3.30

22.

Intellectualizes and tries to reason through

a topic.

3.39

3.54

23.

Treats the other person as an equal.

2.89

3.05

30.

Expresses ideas well, speaks easily and

smoothly.

3.62

3.40

31.

Insists that terms be carefully defined.

3.28

3.53

32.

Chooses words carefully.

3.46

3.68

35.

Is quick to challenge or object.

3.86

3.99

36.

Picks up details in others' conversations.

3.70

3.84

37.

Brings up topics in the right time and place.

3.42

3.20

39.

Behaves in a sympathetic or considerate

manner.

3.42

3.28

41.

Is the sort of person who will admit being

wrong.

2.56

2.72

44.

Makes frequent and appropriate eye contact.

4.03

3.76

45.

Appears confident and sure that he/she is

right.

4.21

4.06

53.

Listens intently and carefully.

3.63

3.62

59.

Gives vague answers - does not take a stand.

3.20

3.17

63.

Has social poise and presence; appears

socially at ease.

4.09

3.44

69.

Disagrees frequently.

3.46

3.72

72.

Takes the initiative; offers suggestions,

information, or plans.

3.54

3.44

77.

Chooses words which fit the subject and are

appropriate for the audience.

3.70

3.53

78.

Behaves assertively.

3.85

3.69

79.

Behaves in a fast-paced way; acts quickly.

3.20

3.12

84.

Is likely to blame or accuse.

3.47

3.64

87.

Is calm and relaxed in manner.

3.82

3.47

91.

Seems to be aware of the impression he/she

makes on others.

3.87

3.35

92.

Can be judgmental.

3.68

3.70

100.

Tends to be liked and accepted by others.

3.97

3.02

Factor 1

Loading

53.

Listens intently and carefully.

.71

31.

Insists that terms be carefully defined.

.68

22.

Intellectualizes and tries to reason through

.64

a topic.

72.

Takes the initiative; offers suggestions,

.59

information, or plans.

36.

Picks up details in others' conversations.

.58

39.

Behaves in a sympathetic or considerate

.56

manner.

41.

Is the sort of person who will admit being

.56

wrong.

37.

Brings up topics in the right time and place.

.55

32.

Chooses words carefully.

.53

23.

Treats the other person as an equal.

.48

59.

Gives vague answers - does not take a stand.

-.47

77.

Chooses words which fit the subject and are

.44

appropriate for the audience.

15.

Overstates ideas or exaggerates them to

-.44

emphasize a point.

Factor 2

63.

Has social poise and presence; appears

.76

socially at ease.

87.

Is calm and relaxed in manner.

.63

45.

Appears confident and sure that he/she is

.60

right.

78.

Behaves assertively.

.59

100.

Tends to be liked and accepted by others.

.51

44.

Makes frequent and appropriate eye contact.

.42

1.

Controls what gets talked about.

.40

Factor 3

4.

Laughs frequently.

.78

5.

Likes to tell stories or anecdotes.

.76

3.

Tells jokes frequently or injects humor into

.76

the conversation.

13.

Explains by using examples, analogies,

.63

or stories.

12.

Smiles frequently.

.62

Factor 4

84.

Is likely to blame or accuse.

.64

92.

Can be judgmental.

.62

69.

Disagrees frequently.

.62

35.

Is quick to challenge or object.

.51

Factor 5

18.

Gestures dramatically.

.70

10.

Has a loud voice.

.56

19.

Shows attention by directing his/her body

.53

towards the listener.

2.

Dominates others in conversation.

.42

Factor 1

Loading

32.

Chooses words carefully.

.65

31.

Insists that terms be carefully defined.

.65

22.

Intellectualizes and tries to reason through

.61

a topic.

36.

Picks up details in others' conversations.

.56

53.

Listens intently and carefully.

.55

37.

Brings up topics in the right time and place.

.54

30.

Expresses ideas well, speaks easily and

.53

smoothly.

41.

Is the sort of person who will admit being

.48

wrong.

23.

Treats the other person as an equal.

.48

72.

Takes the initiative; offers suggestions,

.46

information, or plans.

Factor 2

4.

Laughs frequently.

.71

12.

Smiles frequently.

.71

20.

Uses facial expressions and/or meaningful

.59

gestures.

3.

Tells jokes frequently or injects humor into

the conversation.

.55

13.

Explains by using examples, analogies,

.47

Factor 3

84.

Is likely to blame or accuse.

.73

92.

Can be judgmental.

.68

69.

Disagrees frequently.

.62

35.

Is quick to challenge or object.

.59

15.

Overstates ideas or exaggerates them to

.51

emphasize a point.

59.

Gives vague answers - does not take a stand.

.41

Factor 4

63.

Has social poise and presence; appears

.68

socially at ease.

45.

Appears confident and sure that he/she is

.61

right.

87.

Is calm and relaxed in manner.

.60

91.

Seems to be aware of the impression he/she

.51

makes on others.

77.

Chooses words which fit the subject and are

.45

appropriate for the audience.

100.

Tends to be liked and accepted by others.

.43

Factor 5

2.

Dominates others in conversation.

.72

1.

Controls what gets talked about.

.59

10.

Has a loud voice.

.57

18.

Gestures dramatically.

.52

Sum of

RSQ

Source

DF

Squares

Change

F

Sig F

Style items

Winners

9

12.10

.15

14.7

.001

Losers

6

9.3

.10

15.8

.001

Issues items

Issues

5

1.45

.02

2.95

.01

Importance

5

.39

.00

.79

ns

Regression

25

55.33

22.53

.001

Residual

346

33.98

Total

371

89.31

Multiple R

=

.79

R Square

=

.62

Adjusted

R Square

=

.59

Other Winners

.42

Bush 1988

.64

.84

Bush 1992

.72

.23

.59