Policy, Personality, and Presidential Performance John H. Kessel

Policy, Personality, and
Presidential Performance*
A Paper presented at a Panel in Honor of
John H. Kessel
by
John H. Aldrich
Paul Gronke
and
Jeff Grynaviski
Department of Political Science
Duke University
*
Prepared for delivery at the Annual Meeting of the Midwest Political Science Association, Palmer House
Hotel, Chicago, Ill., April 15-18, 1999. Copies of this paper can also be found at the websites for the Duke
University Democracy, Institutions, and Political Economy Program, http://www.poli.duke.edu/dipe, the
Political Methodology section, http://polmeth.calpoly.edu, and at Gronke’s web page,
http://www.duke.edu/~gronke. The authors would like to thank the Duke University Democracy, Institutions,
and Political Economy Program and the Duke University Arts and Sciences Council for financial support.
We are also indebted to John Brehm and Neil Carlson for their able assistance in the use and abuse of CALIS.
Abstract
Policy, Personality, and
Presidential Performance
John H. Aldrich, Paul Gronke, and Jeff Grynaviski
The importance of personality and performance assessments for candidate
evaluations and choice has been well established, most prominently in the work on
presidential prototypes by Kinder and colleagues, and the social cognitive model of vote
choice by Rahn and colleagues. This paper takes a revisionist look at the effect of
personality assessments for understanding presidential elections. Most of the
experimental and survey data were collected in a relatively brief period, particularly
between 1980 and 1984. Among the unique aspects of the 1980s was the impact of the
distinctive personality of Ronald Reagan, sometimes called the “teflon president,”
because of the degree to which the public admired him as an individual, regardless of
political events. His persona therefore might reasonably be assumed to have uniquely
influenced the times and thus the models and results. We examine this question primarily
by replicating the Rahn, et al. models using the NES surveys from 1984 through 1996,
allowing us to evaluate the structure and performance of presidential prototypes and their
role in candidate assessments over a longer period of time and greater variety of
candidates and presidents.
1
Introduction
One of the many significant contributions that John Kessel has made has been the
demonstration of the value of replication of behavioral models over time, that is over a
variety of electoral circumstances and candidate pairings (e.g., Kessel, 1992). While he
is especially noted for his work on the role of issues in politics (Kessel, 1972; 1975;
1984; Bruce, Clark, and Kessel, 1991; Clark, Bruce, Jacoby, and Kessel, 1991), his full
electoral models assess the longitudinal contributions of all three major forces – parties,
issues, and candidates – on the vote. In this paper, we take a leaf from his book and
begin a longitudinal assessment of the literature that studies impression formation about
candidates, information processing, and judgment (see Ferejohn and Kuklinski, 1990;
Lodge and McGraw, 1995).
Kinder and colleagues applied psychological explanations of how individuals
form impressions of others to politics, notably to impression formation of presidents and
presidential candidates. They also developed measures that began to be used in the
National Election Studies (NES) surveys, most fully beginning in 1984.
In that year,
but using their own survey, Rahn, et al. proposed a “social cognitive model” of voting
that had at its core an account of electoral choice based on candidate appraisal (Rahn,
Aldrich, Borgida, and Sullivan 1990; Sullivan, Aldrich, Borgida, and Rahn 1990). In this
paper, we seek to meld the work of Kinder, et al. and Rahn, et al. by using the NES
measures to develop dimensions of candidate assessments. These results are then used to
replicate the candidate-centered voting model of Rahn, et al. in 1984 (the first year in
2
which the NES measures allow for a substantial test of the model) and then in 1988,
1992, and 1996. We evaluate the lasting contribution of this work in light of our
replication and extension.
Personality and Performance in Candidate Assessment
In a seminal set of articles, Abelson, Kinder, and Fiske argued that political
person perception draws on three separate sources. First, citizens have meta-theories
(Kinder and Abelson 1981) or prototypes about what constitutes leadership:
(p)rototypes are categories people hold about the nature of the world. An
ideal presidential prototype in particular consists of the features that
citizens believe define an exemplary president (Kinder, Peters, Abelson,
and Fiske 1980).
Prototypes are evaluative rulers against which presidential candidates and presidents are
measured. Prior to widespread information about a candidate – for example, Gary Hart in
the weeks following his second place finish in the 1984 Iowa caucuses and subsequent
victory in New Hampshire – the prototype dominates and only later does “reality” intrude
(Bartels 1988). Prototypes also function as an “ideal type,” and a leader’s perceived
successes or failings may be largely due to perceptions about his ability to fulfill our
highest expectations.
A full statement of the possible contents of such a prototype can be found in
Kinder, Abelson, and Fiske (1979). This set may include traits (personality characteristic
ascribed to leaders), affective reactions (patterns of emotional responses elicited by
leaders), behavioral expectations (understandings of what actions presidents take), ideal
types (what the president should be and do), and spontaneous images (respondent
3
generated lists of expectations).1 Sullivan, Aldrich, Borgida, and Rahn (1990) identify a
somewhat different set of dimensions of personality assessment. They argue that
candidates are evaluated on three basic dimensions: altruism vs. selfishness, strength of
will vs. lack of will power, and trustworthiness vs. untrustworthiness. Miller,
Wattenberg, and Malunchuk propose yet a third set of dimensions. Based on an
exploratory factor analysis of the open-ended “likes/dislikes” responses, they suggest that
the responses fall into five categories: competence, integrity, reliability, charisma, and
personal attributes (Miller, Wattenberg, and Malunchuk 1986).
While specifics of the categories differ, in general terms these authors identify a
very similar set of standards underlying voters’ personal assessments of presidential
candidates. This should not be surprising. Experimental research has shown that
character and personality assessments are a foundation of interpersonal relations. These
assessments are not cognitively demanding to make (in part because routines have been
developed and practiced repeatedly), they can therefore be rapidly mobilized, and they
often turn out to be accurate, thus reinforcing their use. These sets of scholars draw upon
research on interpersonal relations and person perception, extending the categories to
perceptions of political actors (Kinder 1978; Miller, Wattenberg, and Malunchuk 1986;
Rahn, et al., 1990; Sullivan, et al., 1990).
An important difference between Kinder’s and Rahn’s teams’ approaches,
however, is that Rahn and colleagues employ an explicitly comparative approach. Thus,
1
The 1979 Pilot Study of the National Election Study included the most elaborate set of candidate characteristics and
evaluative scales. The list was substantially cut down for the 1980 study, extended in 1984, and was included in limited
4
candidates are advantaged to the degree by which they exceed the other candidate on one
or another dimension of evaluation, whether they are evaluated high or low in absolute
terms. This is a quite reasonable expectation, since the conclusion of the evaluation
process is a voting choice, a choice made between or among competing options. We
adopt the comparative approach in this paper. Second, while employing the language of
a “prototype,” Sullivan et al., (1990) reject a single, prototypical standard. Instead, they
argue that citizens may employ two rules, judging a candidate against a standard of
“superman” or one of “everyman.” “Everyman” is an “intuitive profile of human nature,”
a standard by which we judge each other, while “supermen” are individuals who far
exceed the performance standards that we expect of most people. The question, then,
becomes whether presidential candidates are judged most often as everymen or
supermen, and what the consequences may be for political choice.
In addtition to political and leadership standards, individuals also evaluate
political candidates based on and on actual performance as a politician. According to
some, performance serves as a parallel evaluative system, based more on actual
performance in office, or policy information that voters receive during a campaign.
Kinder and Abelson (1981), for example, believe that some changes in the trait profiles
that they observed during the 1980 campaign were a product of the campaign. Rahn,
Aldrich, Borgida, and Sullivan (1990) take a different tack, identifying a set of adjective
pairs, analogous to the “personality” items, which individuals use to ascribe professional
capabilities. What they term competence is a set of “personal characteristics that have a
form from 1988-1996.
5
clear professional component, ” standing in contrast to the more “purely personal and
character oriented” standard of evaluation, personal qualities. Their survey, conducted
by the Gallup organization in 1984, asked voters to rate the candidates on both the
“personal” and the “competence” dimensions.2
The case of Gary Hart in 1984, already mentioned, once again is a useful
illustration of the relative importance of personality versus performance. Hart was
initially able to benefit from widespread projection based on prototypical expectations:
“viable Democratic presidential contender.” Not much more was known about the
candidate at that point, even though well over 80% of the respondents in weekly polls
were willing to hazard a feeling thermometer evaluation and a series of issue placements
(Bartels, 1988; Kinder and Gronke 1985). As the campaign wore on, facts about Hart’s
political positions and personality (heavily colored by the Mondale campaign) served to
diminish his support in the electorate. Depending on your perspective, he was either
punished for his continued ambiguity (Alvarez 1997) or he could no longer benefit from
voter projection due to uncertainty (Page 1978).
However one decides to label the dimensions, their relative impact depends in
part on the contents of the prototype and in part on history. Disentangling these two is a
difficult task. As Kinder, Abelson, and Fiske (1979) conclude: “…properties of specific
candidates may shade the meaning and perhaps shift the importance of competence and
integrity.” Changing standards of evaluation over time, during the 1980 campaign
2
The survey also included items asking how “most people” fared on these dimensions. These measures provided them the
ability to assess candidates against both the “everyman” and the “superman” standards, an approach we are unable to
6
(Kinder and Abelson 1981), and varying standards across candidates, as measured in
1979 (Kinder, Abelson, and Fiske 1979) illustrate the interaction of general standards,
particular candidate arrays, and the campaign environment.
Hart, Reagan, Clinton, and Presidential Candidate Assessments
The number of articles spawned by this research agenda speaks, at least in part, to
its power in allowing us to understand and predict presidential outcomes. The
convergence in results employing different data sets and different methods is good
evidence in favor of the social-cognitive approach.3 So, the reader might ask, why are we
asking questions today? For three reasons: Bill Clinton, Gary Hart, and Ronald Reagan.
Reagan as President apparently was able to base much of his sustained support on
high approval of him as an individual. The “teflon President” was able to weather a good
deal of political difficulties in this way. Clinton appears to be nearly the opposite. His
sustained support appears to derive more directly from assessments rooted in political
judgments, such as satisfaction with economic conditions. Continued high approval
ratings came directly in the face of widespread disenchantment with his personal qualities
(Moore 1999; Schneider 1998; Berke 1998). Perhaps, then, the public’s assessment of
duplicate using the National Election Study’s data.
3
These different data and methods include an exploratory factor analysis of open ended candidate likes and dislikes items,
using three decades of NES surveys;, another using new trait and affect measures, added to the 1979 pilot and 1980
regular National Election Study; and a third using specially written items included in a Gallup poll in a multi-equation
statistical estimation model.
7
the president may be different than yet understood. There are at least two distinct
possibilities. First, the public, in the face of real or perceived “hounding” of presidential
candidates and public figures, may have learned to discount the personal failings of their
political leaders, especially where “private” life is concerned. That is to say that one
possibility is that the way in which the public assesses political figures may have changed
since the 1980s when the Kinder and Rahn teams were writing. Second, the relative
importance of personality and performance may be contingent on a particular candidate
array. We have good prior expectations, based on a reading of the literature and our own
results, that the latter process can occur.
The lasting contribution of Gary Hart – as the “monkey business” Hart of 1988,
not the “where’s the beef” Hart of 1984 – may have been to reduce the importance of
personal sexual conduct in the way that American’s evaluate their political leaders. At
least some of the authors cited above include “morality” as one aspect of character
assessments. However, candidates’ sex lives were not a central part of the social
cognitive approach. And while the public may discount personal sexual behavior
(“morality”), this was not, in the eyes of many, this was not Clinton’s cardinal sin. He
did lie to the American people in his speech in January, 1998, and that provided ample
fodder for his political opponents. “Honesty” and “trustworthiness” are not only
standards his critics claim he failed to meet, but they are also a critical portion of each set
of scholars’ descriptions of the dimensions of candidate assessments. If the political
aspects of the “personal” have undergone a fundamental shift, this argues for a
reexamination of at least that aspect of the social cognitive model. If this has disappeared
or undergone significant modification in the post-Hart era, this should be evident in
8
measurement models applied in 1992 and 1996, and not just for candidate Clinton, but for
Bush, Perot, and Dole as well.
Finally, we return to the political figure who shaped politics during the 1980s.
One Reagan biographer calls him “an American icon” (Metzger 1989) while another
scholar argues that Reagan revived the “nostalgic myth” of America as a “shining city on
the hill.” (Combs 1993). Weiler and Barnett (1992) argue that Reagan fundamentally
changed public and political discourse in America. We do not want to engage in Reagan
hagiography. Yet, it is difficult to avoid the suspicion that Reagan was the prototypical
presidential prototype. Virtually all of the Kinder, Rahn, and their colleagues’ research
was conducted examining only one survey, election year, and pair of candidates, and that
one survey often was conducted at the highpoint of Reagan’s popularity.
A Strategy for Replication and Extension
Where do all these question lead us? We do not question the power of candidate
assessments in general nor of the social cognitive model, in particular. Indeed, a
substantial part of the appeal of that model is the centrality it gives to candidate
assessment in this candidate-centered era. Instead want to examine its stability in new
contexts. Their own research provides guidance as to where we should focus our
analytical lens. Our first task, then, is to see if we can produce reasonably similar
candidate evaluation scales in 1984 and subsequent years.
9
Second, incumbents and challengers appear to be evaluated in quite different
ways. Mondale was held to more of a “superman” standard, a finding attributed to lower
confidence that voters have in the content of their evaluations (Sullivan et al., 1990).
This is similar to the claim that ambiguity by candidates may end up penalizing them in
the eyes of voters (e.g. Alvarez 1997). Similarly, conceptions of the “ideal president”
were only related to evaluations of the sitting president, not other contenders.
Conceptions of the “ideal president,” it may turn out, may be substantially influenced by
the actual occupant of the office (Kinder et al., 1979). Thus, our next hypothesis, one
that can be tested across years, is that voters, even in the face of what Alvarez deems a
“tremendous amount of information about candidates” in presidential general elections,
will be far more certain in their evaluations of incumbents than of challengers.
Third, both sets of scholars highlight the role of political sophistication in the
process of candidate evaluation, yet come to somewhat different conclusions. Kinder and
colleagues suggest that well-educated voters consider competence (i.e., assessments of
political attributes) more in their prototypical expectations, while the less well educated
are more likely to rely on likeability and morality (that is, judgements of the candidates as
persons). The well educated are looking for an “exemplary manager” while the less well
educated are looking for an “exemplary friend” (Kinder et al, 1979 [italics in original];
see also Kinder and Abelson 1981). Rahn and colleagues, in contrast, stress the stability
of their regression coefficients across sophisticated and unsophisticated voters. All voters
rely on both the “personal qualities” and “competence” ratings of presidential candidates,
10
regardless of their level of political interest. Our third hypothesis tests whether structural
relationships are stable across subgroups (defined by political sophistication).
Finally, and most centrally, we wish to examine the hypothesis implicit in Rahn,
et al. (as in the work of Kinder and colleagues), that the structure of candidate
assessments and its role is reasonably general and, hence, reasonably similar from
election to election. Rahn and colleagues present a carefully laid out causal model of
voting. Most proximate to vote choice in their model is party affiliation and “candidate
affect” -- the candidate differential on the NES feeling thermometer scales. While the
particular language may differ, their intellectual forebears include Kelley and Mirer
(1974) and Markus and Converse (1979). In the Kelley and Mirer “simple act of voting,”
a voter chooses by relying on the candidate differential (measured by the difference in
“likes” and “dislikes”). If this is not determinative, party identification serves as a
tiebreaker. Markus and Converse (1979) present a considerably more complicated model
of choice, but the causal order is identical: working backwards from the vote, the first
two “causal arrows” are from candidate evaluations. The penultimate step in the Rahn, et
al., model is that candidate affect is determined by the cognitive assessments of candidate
competence and personal qualities (the “schema triggered affect” hypothesis). These
two, in turn, are shaped by other political forces – domestic and foreign policy, ideology,
and partisan identification. The final analysis replicates and extends their causal model.
11
Candidate Evaluation: Personality, Performance, Plus …?
Our initial task is replication. A first basic judgement in the Rahn, et al., model is
that which the individual makes about the candidate’s qualities as a person. Respondents
are given a set of adjective pairs – in the Rahn, et al., data, trustworthy-untrustworthy,
selfish-unselfish, and cool and aloof-warm and friendly – and judge the candidates using
these pairs. These pairs reflect the “personal qualities” of the candidates, or as we refer
to it “personality.” We call the second standard “performance.” This time, respondents
are asked to evaluate the candidate on three adjective pairs: ineffective-effective,
incompetent-competent, and strong-weak. These are components of personality
assessment that have a “clear professional component.” Because the measurement of
candidate affect, the third dimension of candidate assessment, through use of feeling
thermometers is widely accepted, we do not focus on it in this section.
Given the typology, our replication proceeds in two steps. First, we see whether
we can produce performance and personality as latent variables using a different data set,
the 1984 National Election Study survey. In contrast to the data set used by Rahn et al.,
which was specially designed by the investigators, we are limited to the measures that the
NES chose to include. Fortunately, the close intellectual relationship between their work
and that of Kinder and colleagues meant that an at least related set of items were included
in the 1984 study, and in each NES election study since then.
12
In order to build our set of measures, we collectively determined what we thought
to be the best fit between the NES measures and categories employed by Rahn et al. The
wording of these items, with some slight variation across phrases, is as follows:
In your opinion does the phrase “Hard-Working” describe Reagan?
1) Extremely well; 2) Quite well; 3) Not too well; 4) Not well at all?
Other phrases posed to respondents included “intelligent,” “hard working,” “provides
strong leadership,” and “really cares about people like you.” In 1984, sixteen items were
asked about each candidate (see Appendix). In 1988 and subsequently, only nine were
asked about each candidate, with one different item in 1988 compared to 1992 and 1996.
What we believed to be performance-related items included such phrases as “provides
strong leadership,” “intelligent,” and “knowledgeable.” Personality items included
measures such as “compassionate,” “moral,” and “honest.” Items that we initially chose
to be unconstrained and therefore could load on both latent dimensions were “inspiring”
and “sets a good example.” All estimations proceeded using confirmatory factor
analysis.4
Our attempts to be replicate a model consistent with the two Rahn, et al.,
categories failed miserably. We began with 1984, under the assumption that if we could
reproduce their results in 1984, then we could proceed to replicate in other years. The
consequent estimated model, however, failed to provide an adequate factor structure
4
Estimation was performed in the CALIS procedure of SAS.
13
(results not reported here but available on request). Whether it was a poor fit between the
measures available in the NES compared to their measures or some other difference,
hypothesis tests based on the reported chi-square were unable to obtain anything close to
an acceptable fit of the data and the model. 5 Hayduk (1987) reports that with large
sample sizes, the chi-square is biased upwards: “…with large sample sizes even minute
differences tend to be detectable as being more than mere sample fluctuations and hence
significant.” He therefore suggests using the ratio of chi-square to degrees of freedom
and relying on the other diagnostic goodness of fit measures provided in a typical
LISREL output. An eyeball statistic for goodness of fit is the ratio of c2 to degrees of
freedom; a ratio of two or three to one is a reasonable target; in none of our early models
were we able to obtain ratios much under 10. Nonetheless, however ecumenical we
wished to be, strict devotion to the Rahn categories was unproductive. We returned to
the drawing board and re-categorized the data. Following both our own intuition and
diagnostic statistics, we finally settled upon the following four factor solution, presented
in Figure One.
(INSERT FIGURE ONE ABOUT HERE)
We agreed on four latent dimensions of candidate assessment: “character,”
“competence,” “empathy,” and “strong leader.” The reader can refer to the specific
measures that were allowed to load on each latent dimension, shown in Figure One and
5
A good fit of the model to the data in this context implies an insignificant chi-square. The chi-square statistic measures
whether the observed data matrix S varies insignificantly from the “predicted” data or data implied matrix S (the
14
also on the first page of Table One. Character represents what various authors, cited above,
describe as personal aspects of the candidate’s personality. Can they be trusted, are they
compassionate, are they moral? Strong leader represents the performance aspect of
candidate assessments – what aspects of personality evaluation relates more directly to
activities in office. Thus, does the candidate project an image of strong leadership, does he
merit our respect, are we inspired? Finally, we allowed for two other latent factors to be
estimated, one an elaboration on character, but this time representing those standards of
evaluation that indicate that the candidate understands or empathizes with the position of
the respondent. This we label, not surprisingly, “empathy.” Finally, we added a second
facet to performance as well, assuming that some parts of performance speak more to a
candidate’s ability to perform a job well, without necessarily projecting a strong or
vigorous image. We labeled this final dimension of evaluation “competence.”
This set of four characteristics provided the most satisfactory fit of the data to the
model. As shown in Table One, almost all (save three) of the factor loadings are in the
correct direction, and only the three incorrectly signed coefficients are statistically
indiscernible from zero.6 The goodness of fit measure is credible (.9204) and the ratio of
the chi-square to degrees of freedom (at or under 4.0 for each year) indicates a reasonably
good fit.
(INSERT TABLE ONE ABOUT HERE)
manifest responses predicted using the latent constructs and factor loadings).
6
In confirmatory factor analysis, one loading has to be set to 1.0. All other loadings are scaled relative to that 1.0. This is
why, for example, the unstandardized loading on “religious” for the character dimension is 1.0.
15
We estimated a wide variety of alternative models, and via likelihood ratio tests within
nested models, were able to compare the fit of these alternative models. Most relevant to
our interests here, the four-factor solution described in Figure One and Table One
outperformed a two-factor solution that was as close to the Rahn, et al., results as we
could make it (given the constraints of developing nested models). In 1984, for example,
the chi-square for the two factor solution was 4168 (d.f. = 442). We tried many
variations on a two factor solution, including single-candidate and comparative
dimensional assessments, allowing for covariance among the errors (under the
assumption that the underlying dimensions may constitute a response set), etc. None of
these smaller-dimensional measurement models provided an acceptable goodness of fit
nor outperformed the four factor solution.
Most importantly, these results also held in the 1988, 1992, and 1996 surveys.
However, there are significant complications associated with each year that merit further
examination. First, the NES reduced the number of items composing the trait battery
after 1984. This means that our measurement models for 1988, 1992, and 1996 are based
on a sparser set of manifest variables. We present in Table Two the set of factor loadings
for 1996 for comparative purposes. As with 1984, there are no real surprises in this table.
Except for the inexplicable negative loading for “decent” on the empathy dimension
(implying that the higher value on the unobserved latent measure or “empathy,” the less
well “decent” described Clinton or Dole). Even with the reduced set of indicators, we
feel fairly confident with the measurement model in additional years.
16
(INSERT TABLE TWO ABOUT HERE)
Second, we could not include estimations of Perot in 1992 or 1996. The NES did
not include the questions asking about assessments of Perot in 1992. Attempts to estimate
the model using what data were available about Perot in 1996 were not satisfactory
These results support our suspicion that Perot was evaluated differently from major-party
affiliated (and nominated) presidential candidates. If so, we erred in trying to force the
same measurement model onto the data. To the extent that Rahn, et al., are correct in
their view that evaluations of candidate traits are determined by issues, party
identification, and ideology, it seems likely that individuals either did not have sufficient
information to evaluate Perot, or they did not have cognitive routines for assessing
unaffiliated candidates.
It may not be very surprising that we were unable to replicate the Rahn, et.al.
measurement model using the NES data. After all, the NES was not designed with these
specific needs in mind. Still, even if the differences were not due to different questions,
we do not view failure to replicate as theoretical disconfirmation of the social cognitive
approach. The details of the modeling differ, but the theory underlying the Rahn et al.,
measures is quite similar to that underlying the results of Kinder, at al., and Miller, et al –
and our own. We view this as positive confirmation of our first hypothesis. We have
produced new dimensions that are replicable across four elections, suggesting that, at
least within the context of the specific questions employed by the NES, candidate
17
impression formation has commonality over time and candidates. This allows us to move
to a comparison of our ability to predict variation in the evaluative scales.
Predicting Evaluative Standards
Our first step, establishing the structure in the responses to the NES battery and
assessing its persistence over time, is complete. We next assess the structure in the scales
themselves. The first part of the task here is to answer the question of how well we can
predict variation in candidate evaluations, using a fairly limited set of demographic
predictors. While not strictly analogous to the variance based approach to uncertainty
proposed by Alvarez and Brehm (1995), we do expect to find greater predictive power in
our models of incumbent evaluations compared to challenger evaluation. For the time
being, we will interpret higher explanatory power in these models as evidence of greater
levels of respondent knowledge or certainty about the candidates. In Figure Two, we
present the R2s of the regression of the various candidate assessment measures on
demographic characteristics (education, income, gender, race, and region). Two points
are quite striking in that figure. The first is that in 1984, 1988, and 1996, the fit of the
incumbent president (or vice president in 1988) is substantially higher than the ability to
explain citizens’ assessments of the challenger. Such a result is perhaps remarkably clear
but otherwise not surprising. After all, the social cognitive account (like most theories of
voter choice) is centrally concerned about the consequences of low and differential
information acquisition and utilization. Note also that the incumbent is typically
especially advantaged on the two performance measures, as this account would
anticipate. The second conclusion is that 1992 data are quite different. While the overall
18
fit for Clinton, the challenger, differs little from those for other challengers, the fit for
Bush is much lower than for other incumbents (including himself as vice president four
years earlier). A third conclusion we draw from the detailed results in Table 3 is that the
only major change over time in the impact of individual variables is the much stronger
impact of race on assessments of Clinton. This difference might account for at least some
of the anomaly of 1992 and explain why the explained variance is so high in 1996 (albeit
only marginally more so than 1984).
(INSERT FIGURE TWO AND TABLE THREE ABOUT HERE)
Results from Voting Models
The conventional understanding of the forces shaping voter choices has not
changed much since The American Voter (Campbell, et al., 1960). Scholars agree that
some combination of partisan considerations, issue evaluations, and candidate
assessments shape the voting decision. We do not address the various structural models
(e.g., Fiorina, 1981; Jackson, 1975; Page and Jones 1979; Markus and Converse 1979)
that compete with Rahn, et al., and others for specification of the vote choice. We simply
adopt the formulation proposed Rahn, Aldrich, Borgida, and Sullivan (1990) and by
Kinder and Abelson (1981). In both cases, argued more forcefully by Rahn and
colleagues, affective evaluations of the candidates are most proximal to (and therefore the
last formed before) the vote choice (as is also true in Page and Jones, 1979). Their final
19
vote choice combined the affective evaluations and party identification, with the
cognitive evaluations of the candidates determining the affective assessments.
The interesting question, and original developments Rahn, et al., provided,
concerned the cognitive assessments. Numerous models note the importance of, and
provide estimates of, the effect of candidate thermometer ratings on the vote (see, for
example, Abramson, Aldrich, and Rohde, 1999, for illustration over recent threecandidate elections). Therefore, we focus first on estimations with the cognitive
measures (akin to a one-step reduced form estimation).
In Table Four, we report probit estimations of the effects of the four factors of
social cognitive judgements and partisan identification on the vote. In each case, the
overall goodness of fit is quite large (as is ordinarily true in estimating vote choice
models). In each case, the impact of partisan identification is significant and large. In
each case, some aspect of candidate assessments is also significant and substantial in
magnitude. There, however, the similarities end. In particular, the specific dimensions of
candidate evaluations that are consequential differ from election to election. One could
provide an account that made sense of which factors were relevant when (with the
possible exception of 1996), but as reasonable as those may be, they would be post hoc
accounts. In other words, it seems plausible and consistent with social cognitive accounts
in general (and in the specific cases) that the particular mix of candidates and political
conditions would make some dimensions relevant to choice in one election, others in
another. The theory, however, is not yet sufficiently developed to provide stronger
20
guidance. Assuming our results here persist across alternative elections and data sets,
developing such an account becomes vital.
In Tables Five and Six we report the full replication of the Rahn, et al., model
using the 1984 and 1996 NES surveys. Reported there are the estimates for the full
sample and for the sample partitioned into thirds by level of political engagement.7 We
discuss the full-sample estimates first. (The full causal model, with separate estimates for
each of the evaluative dimensions as well as the vote, are not presented here in interests
of brevity, but are available upon request.)
(TABLES FIVE AND SIX ABOUT HERE)
This attempt at replication of the Rahn, et al., voting model fared substantially
better than our earlier replication effort. Thus, for example, the overall pattern of fit of
the equations is reasonably similar to that which they report. Direct comparisons are, of
course, impossible, because we have four dimensions of cognitive appraisal to their two
(and good evidence that their two category approach fails with the NES measures). In
addition, there are differing patterns to the effect of the various variables between the two
elections in our data. That is, we can almost certainly reject the hypothesis that the
coefficients are the same in 1984 as in 1996. Still, the most important point is that the
7
We constructed a scale of political engagement using interest in politics, frequency of political discussion, newspaper and
newsmagazine readership, frequency of TV news viewing, and attentiveness to presidential campaign news for each of
these media outlets. These variables were factor analyzed (no rotation, limited to a single factor) in order to determine
weights for an additive index. We also produced a measure of political knowledge, based on the respondents education,
the interviewer’s perception of respondent’s political knowledge; and the respondents ability to place the Democratic
candidate at a more liberal position than the Republican candidate on three seven-point issue scales. We do not report
analytical breakdowns based on the knowledge measure here.
21
overall shape of the model estimates is largely similar in the two NES data sets, in
particular. This is especially so in the modeling of the four cognitive assessments. All
variables are significant in each case, and the relative magnitudes of coefficients are
similar across the two years. The same is roughly true for the affect equation, although
there is a different mix of which cognitive dimensions are significant in the two years.
Much the same can be said about the analysis of political sophistication. Rahn, et
al., concluded that the less politically engaged differed little from those more involved.
Here, we report estimates for each of the three divisions of political engagement. In
general, the differences across the three levels are less striking than is the overall
similarity. The overall fit of the equations for those included is increase somewhat with
increasing level of engagement, but those differences are reasonably slight particularly in
the case of 1996. Indeed, the differences are sufficiently small that the overall similarity
is the more striking – and the more consistent with Rahn, et al., conclusion. Perhaps even
more supportive is the pattern of coefficient estimates. Those who anticipate a major
effect of sophistication would likely expect the effect to be revealed through consistently
larger coefficient estimates for the most sophisticated respondents, compared to those less
so. While the estimates differ across levels (and elections), to be sure, there is no
consistent pattern. We therefore conclude that, at least among those who do respond to
the relevant questions, our data are consistent with the Rahn, et al., conclusion. Perhaps
surprisingly, the only major exception is the affect equation. Comparing the two NES
surveys, the first noticeable exception is that more variables are significant among the
least than among the most engaged. Issues, for example, are not significant in either year
22
for those highest on engagement, whereas (and with virtually the same n), issues are
significant in both years for the least (and the moderately) engaged. Looking across
years, it is also striking that party, ideology, and empathy drive the results for the most
engaged in 1996, whereas ideology was not significant in 1986, but the measure of strong
leadership was. While not seeking to underplay these differences, the overall pattern is of
broad similarity in the two NES and, in so far as one can judge, the Gallup data sets.
Conclusions
In this paper, we sought to examine the effect of cognitive appraisals of
presidential candidates using different measures than in previous studies but applying
those measures over the four available elections. The measures included in the 19841996 NES surveys support a four-dimensional measurement of such appraisals. Roughly
speaking, two are focus on the individual personalities of the candidates and two are
centered more on the performance of the political tasks of the presidency expected or
already observed in the candidate. The over-time continuity in this set of factors is a
major argument in behalf not just of the dimensions themselves but on the social
cognitive theory of candidate appraisal underlying them. Another major conclusion
flows from the reasonable replicability of the social cognitive model of vote choice
developed by Rahn, et al. These estimates imply that, even though based on four rather
than two cognitive dimensions, there is strong support for an information processing
23
based account that puts candidate appraisal in the center of political choice in this
candidate-centered electoral era.
There are significant tasks remaining in this field of study. First, reasonable
arguments and evidence have now been amassed in support of different numbers and
types of factors of cognitive appraisal of presidential candidates. Both theoretical and
empirical work is sorely needed to adjudicate among the known (and conceivably many
other) alternatives. Indeed, we view this as a crucial and theoretically significant next
task. In addition, we recommend treating the four-fold solution as a hypothesis rather
than a finished product. For example, the NES questions were originally derived from
the Kinder, et al., work that was based on a different set of specific factors. Revisiting
the number and content of manifest variables in light of the results here seems warranted.
First, neither our work nor Rahn, et al. tests the full social cognitive model against
alternatives, and neither tests for specific causal claims. That is, we now have sufficient
positive evidence for this model that it – and all the others in the same class (e.g., Marcus
and Converse, 1979, and Page and Jones, 1979) – should be evaluated against one
another. Second, our estimates indicate that the relative impact of the various dimensions
of candidate appraisal vary from election to election. Social cognitive theory is not,
however, sufficiently rigorously developed in the discipline to propose hypotheses to
explain such variation.
These recommendations are primarily theoretical – indeed, very difficult
theoretical work. We close with one more substantive observation. We have learned a
24
great deal from the effort to replicate studies of candidate appraisal on different data and
in different political circumstances. This is just the lesson Kessel teaches us. We also
were motivated by the empirical puzzle of how Clinton could maintain support in spite of
public perception of personal failings (and Reagan often held high levels of support
because of perceived personal strengths). Neither our work nor those who preceded us in
the study of candidate and presidential perception seem able to explain this puzzle, at
least not in post-hoc hindsight.
25
Appendix: Candidate Characteristics, 1984 – 1996
HOW WELL DOES R FEEL THE WORD "(word or phrase)" DESCRIBES
(Candidate)?
Descriptor
HARD-WORKING
DECENT
COMPASSIONATE
COMMANDS RESPECT
INTELLIGENT
MORAL
KIND
INSPIRING
KNOWLEDGEABLE
SETS A GOOD EXAMPLE
REALLY CARES ABOUT PEOPLE LIKE YOU
PROVIDES STRONG LEADERSHIP
UNDERSTANDS PEOPLE LIKE YOU
FAIR
IN TOUCH WITH ORDINARY PEOPLE
RELIGIOUS
HONEST
GETS THINGS DONE
·
1984
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
1988
1992
1996
X
X
X
X*
X
X
X
X
X*
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Questions were asked only of Clinton, so were excluded from measurement
models to increase likelihood of compatible factor scores across candidates.
26
Extract from the National Election Study Codebook (1984):
319 HOW WELL DOES R FEEL THE WORD "HARD-WORKING"
DESCRIBES RONALD REAGAN
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
HOW WELL DOES R FEEL THE WORD "DECENT" DESCRIBES RONALD REAGAN
HOW WELL DOES R FEEL THE WORD "COMPASSIONATE" DESCRIBES RONALD
REAGAN
HOW WELL DOES R FEEL THE PHRASE "COMMANDS RESPECT" DESCRIBES
RONALD REAGAN
HOW WELL DOES R FEEL THE WORD "INTELLIGENT" DESCRIBES RONALD
REAGAN
HOW WELL DOES R FEEL THE WORD "MORAL" DESCRIBES RONALD REAGAN
HOW WELL DOES R FEEL THE WORD "KIND" DESCRIBES RONALD REAGAN
HOW WELL DOES R FEEL THE WORD "INSPIRING" DESCRIBES RONALD
REAGAN
HOW WELL DOES R FEEL THE WORD "KNOWLEDGEABLE" DESCRIBES RONALD
REAGAN
HOW WELL DOES R FEEL THE PHRASE "SETS A GOOD EXAMPLE" DESCRIBES
RONALD REAGAN
HOW WELL DOES R FEEL THE PHRASE "REALLY CARES ABOUT PEOPLE LIKE
YOU" DESCRIBES RONALD REAGAN
HOW WELL DOES R FEEL THE PHRASE "PROVIDES STRONG LEADERSHIP"
DESCRIBES RONALD REAGAN
HOW WELL DOES R FEEL THE PHRASE "UNDERSTANDS PEOPLE LIKE YOU"
DESCRIBES RONALD REAGAN
HOW WELL DOES R FEEL THE WORD "FAIR" DESCRIBES RONALD REAGAN
HOW WELL DOES R FEEL THE PHRASE "IN TOUCH WITH ORDINARY PEOPLE"
DESCRIBES RONALD REAGAN
HOW WELL DOES R FEEL THE WORD "RELIGIOUS" DESCRIBES RONALD
REAGAN
27
Figure One: Measurement Model of
Candidate Evaluation, 1984
Decent
Religious
Intelligent
Moral
Hard
Worker
x
Compassionate
Kind
Fair
“Character”
x
Set
Example
“Empathy”
Knowledgeable
“Strong Leader”
x
In Touch
“Competent”
x
Understanding
Strong Leader
Inspiring
Commands
Respect
Cares
28
Table 1: Personality and Performance, Latent Variable Equations, 1984 (Reagan and Mondale)
Manifest Variable
Religious
In Touch
Fair
Understands people like me
Sets a good example
Compassionate
Knowledgable
Decent
Hard Worker
Intelligent
Commands respect
Moral
Really cares about people
Kind
Inspiring
Provides strong leadership
Character
Reagan
Mondale
1.0000 0.6070
1.0000 0.6274
0.4385
0.2662
0.5593
0.3509
0.4334
0.7188
0.2631
0.4363
0.4894
0.9975
0.3071
0.6259
1.2943
0.4048
0.7856
0.2457
1.2243
0.6579
0.7682
0.4128
1.2379
0.7696
1.2373
0.7764
0.9961
0.6046
1.1283
0.7080
Empathy
1.0000
0.7660
1.1027
0.2327
0.5051
0.8211
0.6289
0.9055
0.1911
0.4147
1.0000
0.6366
1.1170
0.2068
0.2461
0.8005
0.5096
0.8941
0.1655
0.1970
1.1221
0.2741
0.2563
0.9213
0.2250
0.2104
1.1112
0.1118
-0.0081
0.8894
0.0895
-0.0065
Leadership
Reagan
Religious
In Touch
Fair
Understands people like me
Sets a good example
Compassionate
Knowledgable
Decent
Hard Worker
Intelligent
Commands respect
Moral
Really cares about people
Kind
Inspiring
Provides strong leadership
Adjusted Goodness of Fit
Mondale
Reagan
Competence
Mondale
Reagan
Mondale
0.4671
0.3951
0.5106
0.4359
0.0459
0.0386
-0.0208
-0.0176
-0.0096
-0.0081
0.2427
0.2071
1.0168
0.8561
0.7336
0.6212
0.4795
0.4056
-0.0677
-0.0578
0.1868
1.0000
0.1573
0.842
0.3348
1.0000
0.2835
0.8468
0.8147
0.6891
0.8574
0.7319
0.7108 0.6012
0.9715 0.8292
1.0000 0.8459
1.0000 0.8536
0.9204 Chi-square
1593.29
(degrees of freedom)
416
Null Model 34726.96
(d.f.)
496
Notes: Data Source = 1984 NES. Cell entries are confirmatory factor analysis coefficients (from manifest variable equations). Standardized coefficients are
italicized. Estimated with Proc CALIS (SAS)
Table 2: Personality and Performance, Latent Variable Equations, 1988 (Bush and Clinton)
Manifest Variable
Compassionate
Knowledgable
Decent
Intelligent
Moral
Really cares about people
Honest
Inspiring
Provides strong leadership
Character
Bush
Dukakis
0.5654
0.4344
0.4270
0.3162
Empathy
Bush
Dukakis
0.4052
0.3523
0.4467
0.3958
1.1663
0.8960
1.2156
0.9002
-0.1701
-0.1479
-0.1437
-0.1273
1.0000
0.7683
1.0000
0.7406
1.0000
0.8695
1.0000
0.8861
1.0708
0.8227
1.1034
0.8171
0.2760
0.2399
0.1282
0.1136
Leadership
Bush
Compassionate
Knowledgable
Decent
Intelligent
Moral
Really cares about people
Honest
Inspiring
Provides strong leadership
Adjusted Goodness of Fit
0.0881
Competence
Dukakis
0.0778
0.2893
0.5285
0.7625
0.5986
1.0000
0.883
1.0000
0.9533 Chi-square
(degrees of freedom)
0.2456
0.6474
0.8490
432.00
99
Bush
Dukakis
0.8660
0.6943
0.6611
0.5091
1.0000
0.8017
1.0000
0.7701
Null Model
(d.f.)
15591.06
153
Notes: Data Source = 1988 NES. Cell entries are confirmatory factor analysis coefficients (from manifest variable equations). Standardized
coefficients are italicized. Estimated with Proc CALIS (SAS)
30
Figure 2: Relative Explained Variance in Evaluative Dimensions, Incumbent vs. Challenger
0.6
1988: Bush vs. Dukakis
1984: Reagan vs.
Mondale
1992: Bush vs. Clinton
1996: Clinton vs. Dole
0.4
Incumbent
Challenger
0.3
0.2
0.1
pa
th
y
er
ad
ng
Le
te
pe
om
ro
St
ra
C
ha
Em
e
nc
er
ct
th
C
Em
pa
ad
ng
Le
te
ro
St
pe
Dimension of Evaluation, by Year
31
y
er
e
nc
er
om
C
C
ha
ra
pa
ct
th
y
er
ad
Le
Em
e
ng
ro
St
om
pe
ra
C
ha
C
te
ct
nc
er
y
th
pa
ad
Le
ng
pe
te
ro
St
om
Em
e
nc
er
ct
ra
C
ha
er
0
C
Variance Explained
0.5
Table 3: Relative Explanatory Power in Evaluative Dimensions: Significant Predictors
Party ID
Ideology
Education
Income
Female
Black
South
Character
.104 ***
.078 ***
-.002
.001
.093 ***
-.238 ***
.068 *
Reagan
Strong
Competence Leader
.129 ***
.170 ***
.108 ***
.115 ***
-.040 ***
.003
.005
.004
.131 ***
.048
.049
-.153 *
.118 *
.112 *
Empathy
.179 ***
.113 ***
-.003
.003
.081 *
-.256 ***
.133 **
Mondale
Strong
Character Competence Leader
-.045 *** -.067 ***
-.108 ***
-.015
-.001
-.029 *
.005
-.006
-.033 **
-.001
-.004
-.011 **
.012
.038
.031
.021
.145 *
.283 ***
-.017
-.013
-.030
Empathy
-.091 ***
-.027
.003
-.008 *
-.022
.100
-.102 *
R-square
.407
.297
.448
.084
.156
Character
Party ID .122 ***
Ideology .029 **
Education -.008
Income
.004
Female
.054
Black
-.124 *
South
.122 ***
Strong
Competence Leader
.110 ***
.152 ***
.025 **
.036 **
-.017
-.041 **
.006 *
.004
.027
.053
-.010
.088
.029 ***
.216 ***
Empathy
.160 ***
.036 **
-.029 *
.006
.078 *
-.105
.150 ***
Dukakis
Strong
Character Competence Leader
-.062 *** -.047 ***
-.106 ***
-.004
-.000
-.001
.035 **
.020 *
.008
-.006 *
-.003
-.010
-.030
-.024
.038
.024
.028
.121 *
.003
.026
.027
Empathy
-.102 ***
-.008
.027 *
-.007 *
.002
.077
-.000
R-square
.245
.319
.093
.143
1984
1988
.428
Bush
.279
.287
.110
.056
.248
.193
Notes: Data are drawn from the 1984, 1988, 1992, and 1996 National Election Studies. As a rough indicator of statistical significance, * = the estimated
regression coefficient is more than two times its standard error. ** = more than three times. ***
32
Table 3: Relative Explanatory Power (con't)
1992
Bush
Character
Party ID .089 ***
Ideology .032 *
Education -.017
Income
.004
Female
.019
Black
-.252 **
South
.066
Strong
Competence Leader
.076 ***
.104 ***
.033 *
.034 *
.000
-.035 *
.003
-.003
.010
-.020
-.119
-.191 *
.075
.113 *
R-square
.134
.172
.171
Empathy
.127 ***
.037 *
-.036 *
.005
-.002
-.292 **
.049
Clinton
Strong
Character Competence Leader
-.084 *** -.052 ***
-.101 ***
-.033 ** -.007
-.032 *
-.014
.040 **
.001
-.007 *
-.002
-.007
.043
.052
.058
.298 *** .227 **
.296 ***
.018
.013
.019
Empathy
-.091 ***
-.033 *
-.006
-.005
.047
.328 ***
.025
.194
.228
.215
Empathy
.151 ***
.045 ***
.016
.006
-.003
-.121 *
.041
Clinton
Strong
Character Competence Leader
-.188 *** .117 ***
-.188 ***
-.031 **
.018 *
-.024 *
-.041 *** -.016
-.024 *
-.109 *** .000
-.007 *
.067 *
.027
.055
.242 *** -.141 *
.269 ***
-.010
.027
-.029
Empathy
-.195 ***
-.036 ***
-.011
-.009 **
.073 *
.265 ***
-.010
.274
.500
.471
Dole
1996
Party ID
Ideology
Education
Income
Female
Black
South
Character
.092 ***
.025 **
.042 ***
.008 **
-.033
-.217 ***
.021
Strong
Competence Leader
.077 ***
.094 ***
.024 **
.036 ***
.031 **
.034
.007 **
.006 *
-.035
-.026
-.168 ***
-.142 *
.038
.080 *
R-square
.248
.232
.206
.079
.208
.219
.463
Notes: Data are drawn from the 1984, 1988, 1992, and 1996 National Election Studies. Cell entries are a rough indicator of statistical significance. * = the
estimated regression coefficient is more than two times its standard error. ** = more than thre
33
Table 4: Voting Models, 1984 - 1996, Including Only Evaluation and Partisanship
Character
Competence
Empathy
Strong Leader
Party ID
Constant
N of Observations
Chi-square
Pseudo-R2
Character
Competence
Empathy
Strong Leader
Party ID
Constant
N of Observations
Chi-square
Pseudo-R2
Probability of Republican Presidential Vote, 1984
By Political Engagment
Low
Medium High
Full Sample
0.626
-0.432
1.447
0.394
1.145
-0.391
0.223
-0.226
0.986
0.486
-0.104
0.846
-0.211
3.57
0.485
0.860
0.752
1.718
-0.144
3.924
0.469
0.254
0.564
0.698
-0.046
6.671
0.307
0.357
0.311
0.330
-0.158
-2.545
-0.187
-0.256
-0.404
-0.508
198
175.75
0.660
805
731.91
0.669
396
353.95
0.656
198
205.47
0.760
Probability of Republican Presidential Vote, 1988
By Political Engagment
Low
Medium High
Full Sample
0.297
0.550
0.539
0.025
0.003
1.227
-0.168
0.447
-0.377
-0.179
0.339
-1.137
0.437
2.087
0.902
0.106
0.911
2.253
0.325
3.129
0.434
1.019
1.092
1.333
0.035
10.212
0.356
0.488
0.315
0.389
0.137
-3.702
-0.578
-0.506
-0.832
-0.403
286
285.72
0.723
1165
1135.5
0.705
34
574
548.2
0.695
287
297.67
0.748
Table 4: Voting Models, con't
Character
Competence
Empathy
Strong Leader
Party ID
Constant
N of Observations
Chi-square
Pseudo-R2
Character
Competence
Empathy
Strong Leader
Party ID
Constant
N of Observations
Chi-square
Pseudo-R2
Probability of Republican Presidential Vote, 1992
By Political Engagment
Low
Medium
High
Full Sample
0.561
0.498
1.126
1.837
1.016
-2.962
-0.768
0.456
-1.684
0.787
-0.564
-2.734
0.483
2.338
-0.119
3.244
1.128
1.692
0.476
3.889
1.012
1.850
2.540
4.885
0.058
5.927
0.344
0.331
0.680
0.343
0.210
-4.872
-1.024
-1.075
-1.028
-1.482
152
166.27
0.822
616
641.66
0.796
307
326.62
0.806
152
165.37
0.865
Probability of Republican Presidential Vote, 1996
By Political Engagment
Low
Medium
High
Full Sample
0.247
0.189
1.308
0.231
0.432
0.172
1.678
1.137
1.477
2.115
0.794
4.268
0.170
4.196
0.200
0.934
0.715
1.167
0.567
0.711
0.797
0.673
0.390
1.397
0.045
8.039
0.363
0.322
0.386
0.717
0.196
-7.306
-1.434
-1.308
-1.516
-2.987
209
170.44
0.602
838
854.75
0.745
418
443.67
0.773
209
256.78
0.894
Notes: Data Source, 1984 - 1996 NES. Cell entries are probit coefficients. For the full sample, the second column contains the
standard error, and the third column contains the t-statistic. For all coefficients, rough statistical significance is indica
35
Table 5: Full Comparative Model, Evaluations and Affect, 1984
Political Engagement
Medium High
Estimates
-0.309
-0.413
-0.142
0.113
0.169
0.137
1.054
0.884
1.233
0.290
0.653
0.417
0.347
0.405
0.515
225
453
226
Full Sample
Low
Character
Intercept
Party ID
Political Issues Index
Political Ideology
Adj. R-square
N
Estimate Std. Error
0.030
-0.307
0.008
0.144
0.164
1.028
0.081
0.464
0.430
923
Competence
Intercept
Party ID
Political Issues Index
Political Ideology
Adj. R-square
-10.263
-0.416
15.533
0.176
6.952
1.541
5.235
0.571
0.398
-0.159
0.129
1.522
0.201
0.297
-0.398
0.162
1.396
0.557
0.369
-0.611
0.216
1.788
0.856
0.514
Strong Leader
Intercept
Party ID
Political Issues Index
Political Ideology
Adj- R-square
-0.466
0.269
1.975
0.832
0.460
0.053
0.015
0.289
0.142
-0.415
0.219
1.736
0.669
0.399
-0.424
0.256
1.661
0.685
0.397
-0.524
0.323
2.953
0.996
0.603
Empathy
Intercept
Party ID
Political Issues Index
Political Ideology
Adj- R-square
-0.937
0.255
1.847
0.759
0.470
0.049
0.014
0.266
0.131
-0.747
0.214
2.168
0.510
0.433
-0.972
0.254
1.453
0.678
0.433
-1.002
0.279
2.214
1.064
0.548
Low
Full Model
Estimate Std. Error
0.017
Intercept
-0.840
Party ID
0.005
0.043
Political Issues Index
0.077
0.259
0.037
Political Ideology
0.156
0.037
Character
0.139
Competence
0.001
0.021
0.014
Strong Leader
0.137
Empathy
0.019
0.132
R-square
0.821
Affect Equation
-0.099
0.044
0.322
0.159
0.122
0.018
0.178
0.089
0.741
Medium High
Estimate
-0.111
-0.061
0.035
0.059
0.353
0.221
0.147
0.148
0.062
0.188
-0.018
0.028
0.133
0.103
0.135
0.160
0.821
0.856
Notes: Data source, 1984 National Election Study. Cell entries are unstandardized regression coefficients.
Sample size is entered for the first equation only, but is fixed across all estimates.
36
Table 6: Full Comparative Model, Evaluations and Affect, 1996
Full Sample
Political Engagement
Low
Medium High
Estimates
-0.412
-0.490
-0.437
0.245
0.286
0.353
0.873
1.647
0.800
0.253
0.071
0.484
0.416
0.480
0.550
295
587
294
Character
Intercept
Party ID
Political Issues Index
Political Ideology
Adj. R-square
N
Estimate Std. Error
0.043
-0.442
0.014
0.293
0.234
1.151
0.115
0.305
0.487
1179
Competence
Intercept
Party ID
Political Issues Index
Political Ideology
Adj. R-square
-0.558
0.211
0.848
0.223
0.490
0.031
0.010
0.168
0.083
-0.480
0.169
1.247
0.141
0.417
-0.579
0.219
0.583
0.286
0.485
-0.579
0.234
0.750
0.206
0.541
Strong Leader
Intercept
Party ID
Political Issues Index
Political Ideology
Adj- R-square
-0.659
0.288
1.141
0.302
0.478
0.043
0.014
0.234
0.115
-0.570
0.233
1.619
0.213
0.410
-0.684
0.298
0.820
0.387
0.471
-0.683
0.316
0.996
0.277
0.528
Empathy
Intercept
Party ID
Political Issues Index
Political Ideology
Adj- R-square
-0.112
0.336
0.459
0.483
0.513
0.048
0.015
0.264
0.130
-1.019
0.290
2.303
0.197
0.449
-1.081
0.324
1.011
0.672
0.496
-1.222
0.392
1.272
0.402
0.581
Low
Full Model
Estimate Std. Error
Intercept
0.015
-0.180
Party ID
0.004
0.040
Political Issues Index
0.062
0.311
Political Ideology
0.030
0.180
Character
0.029
0.015
Competence
0.078
0.166
Strong Leader
0.036
0.049
Empathy
0.012
0.089
R-square
0.809
Affect Equation
-0.133
0.031
0.140
0.253
0.009
0.332
-0.025
0.061
0.739
Medium High
Estimate
-0.176
-0.226
0.034
0.056
0.187
0.483
0.220
0.111
0.037
0.043
0.172
-0.061
0.034
0.136
0.081
0.129
0.812
0.850
Notes: Data source, 1996 National Election Study. Cell entries are unstandardized regression coefficients.
Sample size is entered for the first equation only, but is fixed across all estimates.
37
References
Abelson, R.P., D.R. Kinder, M.D. Peters, and S.T. Fiske. 1982. “Affective and Semantic
Components in Political Person Perception.” Journal of Personality and Social Psychology.
42,4: 619-630.
Abramson, Paul R., John H. Aldrich, and David W. Rohde. 1999. Change and Continuity in the
1996 and 1998 Elections. Washington, DC: CQ Press.
Alvarez, R. Michael. 1997. Information and Elections. Ann Arbor, MI: University of Michigan
Press.
Alvarez, R. Michael and John Brehm. 1997. “Are Americans Ambivalent Towards Racial
Policies?” American Journal of Political Science 41 (April): 345-374.
Bartels, Larry M. 1988. Presidential Primaries and the Dynamics of Public Choice. Princeton,
N.J.: Princeton University Press.
Berke, Richard L. 1998. “In Presidents, Virtues Can Be Flaws (and Vice Versa).” New York
Times, September 27, 1998.
Bruce, John M., John A. Clark, and John H. Kessel. 1991. “Advocacy Politics in Presidential
Parties.” American Political Science Review. 85(Dec.): 1089-1106.
Campbell, Angus, Philip E. Converse, Warren E. Miller, and Donald E. Stokes. 1960. The
American Voter. New York: John Wiley & Sons.
Combs, James E. 1993. The Reagan Range: The Nostalgic Myth in American Politics. Bowling
Green: The University of Bowling Green Press.
Clark, John A., John M. Bruce, William Jacoby, and John H. Kessel. 1991. “I’d Rather Switch than
Fight: Life-Long Democrats and Converts to Republicans among Campaign Activists.”
American Journal of Political Science. 35(Aug.): 577-597.
Fiorina, Morris P. 1981. Retrospective Voting in American National Elections. New Haven, CT:
Yale University Press.
Fiske, Susan T. 1993. “Cognitive Theory and the Presidency.” In Geroge W. Edwards, John H.
Kessel, and Bert A, Rockman, eds., Researching the Presidency. Pittsburgh, PA:
University of Pittsburgh Press: 233-265.
Ferejohn, John, and James Kuklinski, eds. 1990. Information and Democratic Processes. Urbana,
IL: University of Illinois Press.
Hayduk, Leslie A. 1987, Structural Equation Modeling with LISREL: Essentials and Advances.
Baltimore, MD: Johns Hopkins University Press.
Jackson, John. 1975. “Issues, Party Choices, and Presidential Votes.” American Journal of Political
Science. 19(February): 851-80.
38
Kelley, Stanley, Jr., and Thad W. Mirer. 1974. “The Simple Act of Voting.” American Political
Science Review. 68(June): 572-91.
Kessel, John H. 1972. “Comment: The Issues in Issue Voting.” American Political Science Review.
66(June): 459-65.
Kessel, John H. 1983. “The Structures of the Carter White House.” American Journal of Political
Science. 27(Aug.): 431-63.
Kessel, John H. 1992. Presidential Campaign Politics, 4th ed. Pacific Groves, CA: Brooks/Cole.
Kessel, John H. 1984. Presidential Parties. Homewood, IL: Dorsey Press.
Kinder, Donald R. 1983. “Presidential Traits: Pilot Study Reports to the 1984 NES Planning
Committee.” Ann Arbor, MI: Center for Political Studies, University of Michigan.
--------. 1978. “Political Person Perception: The Asymmetrical Influence of Sentiment and Choice
on Perceptions of Presidential Candidates.” Journal of Personality and Social Psychology.
36,8: 859-871.
Kinder, Donald R. and Robert P. Abelson. 1981. “Appraising Presidential Candidates: Personality
and Affect in the 1980 Campaign.” Paper prepared for the 1981 Annual Meeting of the
American Political Science Association, New York, NY.
Kinder, Donald R., Robert P. Abelson, and Susan T. Fiske. 1979. “Developmental Research on
Candidate Instrumentation: Results and Recommendation. Report Submitted to the NES
Board of Overseers.” Ann Arbor, MI: Center for Political Studies, University of Michigan.
Kinder, D.R., and Paul Gronke. 1985. “Character and Affect in the 1984 Presidential Election.”
Paper presented at the Annual Meeting of the Midwest Political Science Association,
Chicago, IL.
Kinder, Donald R., Mark D. Peters, Robert P. Abelson, and Susan T. Fiske. 1980. “Presidential
Prototypes.” Political Beahvior. 2:317-37.
Lodge, Milton, and Kathleen M. McGraw, eds. 1995. Political Judgment: Structure and Process.
Ann Arbor, MI: University of Michigan Press.
Marcus, Gregory B., and Philip E. Converse. 1979. “A Dynamic Simultaneous Equation Model of
Electoral Choice.” American Political Science Review. 73(Dec.): 1055-70.
Metzger, Robert P. 1989. Ronald Reagan: An American Icon. Philadelphia, PA: University of
Pennsylvania Press.
Miller, Arthur, Martin Wattenberg, and Oksana Malunchuk. 1986. “Schematic Assessments of
Presidential Candidates.” American Political Science Review 80:2 (June) 521-40.
Moore, David W. 1999. “Good Times for Clinton the President. But Personal Reputation Hits
New Low.” Gallup News Service, The Gallup Organization
(http://www.gallup.com/poll/releases/pr990123.asp), January 23, 1999.
Page, Benjamin I. 1978. Choices and Echoes in Presidential Elections: Rational Man and
39
Electoral Democracy. Chicago, IL: University of Chicago Press.
Page, Benjamin I., and Calvin C. Jones. 1979. “Reciprocal Effects of Policy Preferences, Party
Loyalties and the Vote.” American Political Science Review. 73(Dec.): 1071-89.
Rahn, W.M., J.H. Aldrich, and E. Borgida. 1994. “Individual and Contextual Variations in Political
Candidate Appraisal.” American Political Science Review. 88,1 (March): 193-199.
Rahn, W.M., J.H. Aldrich, E. Borgida, and J.L. Sullivan. 199x. “A Social Cognitive Model of
Candidate Appraisal,” in Ferejohn, J.A. and J.H. Kuklinski (eds), Information and
Democratic Processes. Champaign, IL: University of Illinois Press.
Schneider, William. 1998. “Not a President to Respect or Fear.” The National Journal
“Cloakroom” website (http://www.cloakroom.com/members/buzz/njbuzz.htm), August 21,
1998.
Sullivan, J.L., J.H. Aldrich, E. Borgida, and W.M. Rhan. 1990. “Candidate Appraisal and Human
Nature: Man and Superman in the 1984 Election.” Political Behavior. 459-484.
Weiler, Michael and W.B. Barnett. 1992. Reagan and Public Discourse in America. Tuscaloosa:
University of Alabama Press
40