Jump to ContentJump to Main Navigation
Leaders' Personalities and the Outcomes of Democratic Elections$

Anthony King

Print publication date: 2002

Print ISBN-13: 9780199253135

Published to Oxford Scholarship Online: November 2003

DOI: 10.1093/0199253137.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2017. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in OSO for personal use (for details see http://www.oxfordscholarship.com/page/privacy-policy). Subscriber: null; date: 25 February 2017

The Impact of Candidate Traits in American Presidential Elections

The Impact of Candidate Traits in American Presidential Elections

Chapter:
(p.44) 2 The Impact of Candidate Traits in American Presidential Elections
Source:
Leaders' Personalities and the Outcomes of Democratic Elections
Author(s):

Larry M. Bartels

Publisher:
Oxford University Press
DOI:10.1093/0199253137.003.0002

Abstract and Keywords

The aim of this chapter is to provide a systematic test of the conventional wisdom that personality is key in contemporary American electoral politics. Using survey data from the six most recent presidential elections, the contours are examines of the candidates’ images (traits), the bases of those images in voters’ more fundamental political predispositions, and the impact of voters’ assessments of the candidates’ personal qualities on individual voting behaviour and on aggregate election outcomes. In stark contrast with the popular conception of contemporary electoral politics as candidate–centred and image–driven, it is argued that candidates’ images are largely epiphenomenal and have only a modest impact on election outcomes. This conclusion is underlined by the analysis given of the 2000 (Bush vs. Gore) presidential election, in which the estimated impact of voters’ assessments of the candidates’ personalities was even smaller than in the previous five elections considered here, although quite probably large enough to be decisive in an election decided by a few hundred votes in a single state.

Keywords:   Bush, candidates' images, candidates' personal characteristics, candidates' personalities, candidates' traits, democratic elections, election outcomes, Gore, United States, voting behaviour

One of the most familiar themes in modern electoral analysis emphasizes the potency of televised images of political candidates to sway the voting behavior of mass electorates. Beginning with the rise of television as a political force in the 1950s, political journalists and campaign consultants have suggested that the “old politics” of socially and ideologically based party coalitions has given way to a “new politics” centering on the personal qualities of individual candidates, as conveyed, or even created, through the medium of television. Thus, by the early 1960s, one leading political scientist felt it necessary to deny that American voters were “moved by subconscious urges triggered by devilishly skillful propagandists.”1 A decade later, another prominent academic study decried what its authors referred to as “The Myth of Television Power in National Elections.”2

Joe McGinniss's The Selling of the President 1968 marked something of a watershed in popular perceptions of the modern electoral process. Its behind‐the‐scenes look at how even so apparently unattractive a candidate as Richard Nixon could be repackaged and sold to a gullible American electorate generated both fascination and consternation. Books with titles like The Image Candidates, The Political Persuaders, and Polls, Television, and the New Politics were more scholarly in tone, but reinforced the impression that the electoral process had been fundamentally altered by the rise of television.3

(p.45) The emphasis in the late 1960s and 1970s on candidates' “images” and the politics of personality was often interpreted as contradicting an older academic perspective emphasizing the importance of stable partisan loyalties—a perspective represented most prominently by The American Voter.4 Of course, the authors of The American Voter were hardly unaware of the fact that the personal popularity of Dwight Eisenhower outweighed the Democratic Party's longstanding advantage in party identification in each of the two successive presidential elections they analyzed. Their own subsequent work elaborated the role of “dynamic” short‐term forces, including the personal images of the competing candidates, in producing such “deviating elections.”5 However, the appropriate place of candidates' personalities within a broader understanding of voting behavior and electoral politics remained unclear.

Notwithstanding the somewhat uneasy coexistence of traditional voting research, political psychology, and popular political commentary, by the 1990s The Rise of Candidate‐Centered Politics (to quote the title of an influential study by Wattenberg)6 was an accepted fact among most political scientists. Television was widely viewed as the moving force behind that development. For example, Butler and Ranney concluded that “Television . . . has done more than anything else to transform electioneering.” In particular, they argued, television is “more inclined to focus on visual images and personalities. It has intensified the concentration on leaders as the sole spokesmen of their parties.”7

The impact of personality in contemporary American politics has also been accepted as a matter of faith among political observers beyond academia. As one prominent commentator, Anna Quindlen, put it in the midst of the most recent presidential campaign,

personality is key in Election 2000. . . . just as they use it to choose their friends, their spouses, the neighbors they invite over for a barbecue, the co‐workers they join for lunch, so voters use their impressions of a candidate's personality to choose a president. In the next 90 days millions of people will decide, finally, whether they think Al Gore is rigid and humorless or instead serious and diligent, whether George W. Bush is straight‐talking and sure of himself or simply arrogant and tactless. And that will matter. A lot.8

(p.46) A similar understanding of the electoral impact of personality was conveyed in a typical description in the New York Times of candidate Bush's efforts to portray himself as a common man rather than a scion of wealth: “in an election that many Republican and Democratic strategists believe will turn as much on personality as on policy, the way voters react to each candidate's efforts along these lines could be decisive.”9

My aim in this chapter is to provide a systematic test of the conventional wisdom that, as Quindlen put it, “personality is key” in contemporary American electoral politics. Using survey data from the six most recent presidential elections, I shall examine the contours of the candidates' images, the bases of those images in voters' more fundamental political predispositions, and the impact of voters' assessments of the candidates' personal qualities on individual voting behavior and on aggregate election outcomes. In stark contrast with the popular conception of contemporary electoral politics as candidate‐centered and image‐driven, I shall argue that candidates' images are largely epiphenomenal and that they have only a modest impact on election outcomes. This conclusion is underlined by my analysis of the 2000 presidential election, in which the estimated impact of voters' assessments of the candidates' personalities was even smaller than in the previous five elections considered here—though quite probably large enough to be decisive in an election decided by a few hundred votes in a single state.

Candidate Traits

The best available data on prospective voters' personal impressions of American presidential candidates come from a battery of “candidate trait” items included in recent American National Election Study (NES) surveys.10 In each presidential election year since 1980, NES has invited survey respondents to rate the competing presidential candidates on a series of character traits. The specific personal qualities tapped in these “trait ratings” have varied somewhat from year to year, but three traits have been included in all of the past six presidential election years and two more have been included in five of those six years. These five hardy‐perennial traits are:

  1. moral

  2. knowledgeable

  3. inspiring

  4. provides strong leadership

  5. really cares about people like you.

  6. (p.47)

Considerable scholarly work has gone into conceptualizing and validating these and other specific trait questions and exploring their interrelationships.11 However, for my purposes here it is only necessary to accept the five traits included regularly in the NES surveys as a fair sampling from the broader range of personal qualities that might conceivably be relevant to prospective voters.12 Indeed, even readers unhappy with this selection of relevant traits may be reassured to learn that my conclusions turn out to be rather surprisingly insensitive to exactly which traits, or how many traits, are included in the analysis.13

The standard format of the NES trait questions requires respondents to assess how well each of a series of words or phrases describes a particular candidate. For example:

Think about Bill Clinton. In your opinion, does the phrase “he really cares about people like you” describe Bill Clinton extremely well, quite well, not too well, or not well at all?

In my recoding of the survey responses, extremely well corresponds to a score of 100, quite well to a score of 66.7, not too well to a score of 33.3, and not well at all to a score of zero.14

(p.48) Table 2.1 provides a summary of the average ratings on the resulting 0–100 scale for each major‐party presidential candidate on each of the five traits included regularly in the NES surveys. (Respondents in 1980 were not asked the “really cares” questions; respondents in 2000 were not asked the “inspiring” questions.) The trait ratings come from preelection surveys conducted in the two months preceding each November election. Given my interest in the electoral impact of trait assessments, all of the data in Table 2.1 and throughout this chapter come from NES survey respondents who reported, in a post‐election reinterview, voting for one or the other of the major‐party presidential candidates; those who reported not voting or voting for a minor‐party candidate for president are excluded from my analysis.15

Table 2.1. Average Trait Ratings of U.S. Presidential Candidates, 1980–2000

1980

1984

1988

1992

1996

2000

Reagan/Carter

Reagan/Mondale

Bush/Dukakis

Bush/Clinton

Dole/Clinton

Bush/Gore

Really cares

Republican candidate

48.7

49.5

40.0

47.2

46.5

(0.9)

(0.8)

(0.8)

(0.8)

(0.9)

Democratic candidate

56.7

57.8

58.3

53.7

54.2

(0.7)

(0.7)

(0.7)

(1.0)

(0.9)

Moral

Republican candidate

64.2

72.3

66.3

68.4

70.1

64.1

(0.9)

(0.7)

(0.7)

(0.7)

(0.7)

(0.8)

Democratic candidate

72.6

66.4

62.8

46.2

38.3

64.3

(0.9)

(0.6)

(0.6)

(0.7)

(0.9)

(0.8)

Knowledgeable

Republican candidate

62.4

64.6

68.1

67.6

70.2

58.7

(1.0)

(0.8)

(0.6)

(0.6)

(0.6)

(0.8)

Democratic candidate

63.8

65.0

64.4

66.8

72.4

69.7

(0.9)

(0.6)

(0.6)

(0.6)

(0.8)

(0.7)

Inspiring

Republican candidate

51.0

58.2

43.6

44.0

44.3

(1.0)

(0.8)

(0.7)

(0.8)

(0.8)

Democratic candidate

41.5

44.4

48.1

55.0

51.0

(1.0)

(0.7)

(0.7)

(0.7)

(0.9)

Strong leader

Republican candidate

57.3

65.7

50.8

53.2

58.2

57.5

(1.0)

(0.8)

(0.8)

(0.8)

(0.8)

(0.8)

Democratic candidate

39.2

46.7

51.9

56.0

54.0

52.7

(1.0)

(0.7)

(0.7)

(0.7)

(0.9)

(0.8)

Note: Average ratings on 0–100 scale by major‐party presidential voters only (with standard errors in parentheses).

(p.49) Given the large sample sizes of the NES surveys, the mean trait ratings are estimated fairly precisely—the standard errors of the estimated means are never more than 1 point on the 100‐point scale. Thus, we can be fairly sure that the differences in mean ratings from candidate to candidate and from trait to trait that appear in Table 2.1 are real. However, it is worth noting that most of the differences among candidates for any given trait are relatively modest, especially for candidates of the same party. For example, the range of the six “really cares” ratings for Republicans is only from 40.0 to 49.5 (both, as it happens, for the elder George Bush), while the corresponding range for Democrats is only from 53.7 to 58.3 (both, as it happens, for Bill Clinton). The average difference in ratings for competing candidates, averaging over both traits and election years, is only a little more than 8 points on the 100‐point scale, and the only cases in which any pair of competing candidates' ratings on any trait differ by as much as 20 points are the two comparisons involving Bill Clinton's ratings on the “morality” trait, which were markedly lower than those of any other candidate in the six elections covered by my analysis.

Obviously, the fact that stark differences in aggregate public perceptions of the competing candidates' personal qualities are rare tends to limit the potential impact of personalities on election outcomes. Even if voters weigh candidate traits heavily in deciding whom to vote for, the net effect of “image” considerations will be modest if prospective voters do not clearly prefer one of the competing candidates to the other. One reason why they might not is that parties presumably try to avoid nominating candidates with salient character flaws. In addition, as I shall attempt to demonstrate here, the average impressions of the competing candidates reported in Table 2.1 are moderated by a strong tendency for each party's partisans to think highly of their own candidate and less highly of his opponent. The result is that most of the average ratings fall in a rather narrow, mildly positive range, from about 50 to 65 on the 100‐point scale.

Most of the differences that do appear in the various candidates' average trait ratings seem to be consistent with the assessments of contemporary observers regarding the candidates' personal strengths and weaknesses. For example, Ronald Reagan in 1984 received higher marks than any other recent presidential candidate for being “inspiring” and a “strong leader,” while Jimmy Carter in 1980 received the lowest marks on those dimensions. Democratic candidates in every election year were more likely than their Republican opponents to be perceived as “really caring” about ordinary people. And Carter in 1980 was (by a slight margin) the most highly rated candidate with respect to “morality,” while, as I have already indicated, Clinton's “morality” ratings in both 1992 and 1996 were much lower than those of any other recent presidential candidate.

(p.50) In the Eyes of the Beholders: Political Biases in Trait Ratings

The average trait ratings presented in Table 2.1 provide some evidence of real—and not unreasonable—differences in aggregate public perceptions of different candidates' personal qualities. However, those generally modest aggregate differences disguise a good deal of disagreement among survey respondents in any given year regarding the presidential candidates' personal qualities. Much of this disagreement presumably reflects idiosyncratic differences in individual responses to specific candidates, or even simple measurement error in the survey responses. At the same time, however, there are clear patterns in the individual responses suggesting that specific impressions of the candidates' traits are strongly affected by respondents' more general political attitudes.

The extent to which perceptions of political stimuli may be colored by political predispositions is vividly illustrated by the reactions of Patterson and McClure's survey respondents to television advertisements and news stories during the 1972 presidential campaign. Consider, for example, the following pairs of reactions to news stories and advertisements featuring George McGovern, each of which includes one reaction from a McGovern supporter followed by one from a Nixon supporter:16

  • He's a good, honest man. You can see that.
  • Can't believe what you see. It would be worse with him.
  • He looks so able. I think he's very able.
  • He's a weak sister. You can see that immediately.
  • He really cares what's happened to disabled vets. They told him how badly they've been treated and he listened. He will help them.
  • McGovern was talking with these disabled vets. He doesn't really care about them.
  • He's just using them to get sympathy.
  • It was honest, down‐to‐earth. People were talking and he was listening.
  • Those commercials are so phoney. He doesn't care.
  • McGovern had his coat off and his tie was hanging down. It was so relaxed, and he seemed to really be concerned with those workers.
  • He is trying hard to look like one of the boys. You know, roll up the shirt sleeves and loosen the tie. It's just too much for me to take.
As Patterson and McClure aptly observed, “Different people were watching the same George McGovern on the screen, but clearly, they were not seeing (p.51) the same man . . . Their response was not to the image, but to the politics George McGovern personified.”17

The perceptual biases that Patterson and McClure observed in their respondents' reactions to news coverage and advertisements also appear in a wide variety of other data on perceptions of political figures, including the trait ratings of presidential candidates collected by NES. The magnitude of these perceptual biases is illustrated in Figure 2.1, which shows how ratings of Bill Clinton as a “strong leader,” “moral,” and “knowledgeable” varied with the partisan loyalties of the people doing the rating in the 1992 NES survey. In each panel of the figure, the array of bars represents the percentages in each partisan group (from strong Democrats on the left to strong Republicans on the right) who said the trait in question described Clinton “extremely well” or “quite well.”

The pattern of partisan bias evident in Figure 2.1 is especially striking because the measure of party identification used to differentiate voters' assessments of Clinton's personal qualities in 1992 comes from a separate NES survey conducted with the same respondents two years earlier, before Clinton had even emerged as a salient political figure. Thus, there is no question here of voters adjusting their partisanship to comport with their assessments of

                   The Impact of Candidate Traits in American Presidential Elections

Fig. 2.1 Partisan Biases In Clinton Trait Ratings, 1992

(p.52) Clinton. Impressions of the candidate were strongly conditioned by political loyalties that clearly predated his arrival on the national political scene.

The largest partisan biases evident in Figure 2.1 are for assessments of Clinton as a “strong leader” and “moral,” in the first and second panels respectively. For each of these traits the proportion of strong Democrats who thought the label fitted Clinton “extremely well” or “quite well” exceeded the corresponding proportion of strong Republicans by approximately 50 percentage points, and even weak party identifiers differed in their assessments by about 25 percentage points. Short of arguing that Democrats and Republicans differed markedly in their definitions of what it means to be a “strong leader” or “moral”—and that Clinton just happened to exemplify the Democratic definition much better than the Republican definition in each case—it seems impossible to resist the conclusion that the trait assessments shown in Figure 2.1 reflect the biases of the voters doing the rating as much or more than they do the intrinsic qualities of the candidate being rated.

Even seemingly straightforward assessments of the candidates are subject to significant partisan biases. For example, the right‐most panel of Figure 2.1 shows the percentage of each partisan group that rated Clinton as “knowledgeable.” Even among strong Republicans, 65 percent viewed Clinton as knowledgeable—but that figure was almost 20 points lower than among Independents and 27 points lower than among strong Democrats; and strong Republicans were only one‐third as likely as strong Democrats to rate Clinton as “extremely” knowledgeable.

The impact of political biases on perceptions of the candidates' traits is documented more generally in Table 2.2, which reports the results of statistical analyses in which each relative trait rating (that is, the signed difference between the Republican and Democratic candidates' respective ratings) in each election year is regressed on voters' party identifications, political ideologies, and perceptions of recent changes in economic conditions.18 Each of these political determinants is recoded to range from 1 for an extreme pro‐Democratic response to 1 for an extreme pro‐Republican response, with a score of zero indicating a neutral position.19

(p.53)

Table 2.2. Political Determinants of Relative Trait Ratings, United States, 1980–2000

1980

1984

1988

1992

1996

2000

Really cares

Intercept

−10.0

−8.7

−4.0

−2.7

−5.5

(1.0)

(1.0)

(1.5)

(1.2)

(1.2)

Republican party identification

31.8

27.9

35.1

34.9

39.4

(1.5)

(1.5)

(1.5)

(1.7)

(1.8)

Conservative ideology

13.6

13.7

16.7

19.0

10.3

(2.5)

(2.6)

(2.4)

(2.9)

(2.4)

Economic assessment

18.8

11.4

20.7

17.8

7.9

(2.0)

(2.2)

(2.2)

(2.4)

(2.0)

Standard error of regression –

33.9

32.9

34.0

33.1

35.6

Adjusted R 2

0.46

0.38

0.50

0.53

0.47

N

1,361

1,193

1,356

1,034

1,120

Moral

Intercept

−10.5

4.3

3.0

27.8

34.0

1.0

(2.2)

(0.8)

(0.9)

(1.5)

(1.3)

(1.1)

Republican party identification

19.6

16.3

14.2

22.0

32.7

23.6

(1.8)

(1.3)

(1.3)

(1.5)

(1.8)

(1.7)

Conservative ideology

8.4

14.9

10.6

16.2

8.4

10.4

(3.2)

(2.1)

(2.2)

(2.4)

(3.1)

(2.3)

Economic assessment

5.7

9.3

8.4

7.2

3.9

7.8

(2.9)

(1.7)

(1.9)

(2.2)

(2.6)

(1.9)

Standard error of regression

33.6

28.3

27.8

34.3

35.0

34.0

Adjusted R 2

0.19

0.29

0.22

0.29

0.39

0.31

N

837

1,365

1,193

1,355

1,031

1,120

Knowledgeable

Intercept

−6.0

−1.9

3.9

7.7

0.0

−10.3

(2.2)

(0.9)

(0.8)

(1.2)

(1.0)

(1.1)

Republican party identification

21.6

19.0

15.6

16.6

17.1

20.1

(1.8)

(1.4)

(1.3)

(1.3)

(1.5)

(1.7)

Conservative ideology

15.5

15.3

5.0

11.1

6.4

9.7

(3.3)

(2.3)

(2.2)

(2.0)

(2.5)

(2.3)

Economic assessment

8.8

8.8

8.5

10.3

8.4

5.4

(2.9)

(1.9)

(1.9)

(1.8)

(2.1)

(1.9)

Standard error of regression

34.1

32.1

27.5

28.9

28.3

33.8

Adjusted R 2

0.25

0.27

0.21

0.26

0.25

0.24

N

836

1,366

1,194

1,355

1,033

1,120

Inspiring

Intercept

2.7

11.1

−4.2

−1.8

−3.2

(2.4)

(1.0)

(1.0)

(1.4)

(1.3)

Republican party identification

28.5

25.3

21.1

26.7

28.4

(2.0)

(1.6)

(1.5)

(1.5)

(1.9)

Conservative ideology

14.6

14.4

7.2

14.6

13.8

(3.6)

(2.5)

(2.6)

(2.4)

(3.2)

Economic assessment

19.9

20.7

11.4

12.8

16.2

(3.2)

(2.0)

(2.2)

(2.2)

(2.7)

Standard error of regression

37.3

34.6

32.2

33.6

37.0

Adjusted R 2

0.31

0.39

0.27

0.37

0.37

N

836

1,365

1,190

1,356

1,032

Strong leader

Intercept

8.8

16.0

−1.0

8.4

7.8

6.8

(2.5)

(1.0)

(1.0)

(1.4)

(1.3)

(1.2)

Republican party identification

34.4

27.4

25.6

26.9

29.8

35.4

(2.1)

(1.6)

(1.5)

(1.5)

(1.8)

(1.8)

Conservative

14.1

9.8

10.7

11.8

9.8

10.0

ideology

(3.7)

(2.5)

(2.7)

(2.4)

(3.1)

(2.4)

Economic assessment

19.1

26.2

13.3

16.0

13.2

8.1

(3.3)

(2.0)

(2.3)

(2.1)

(2.6)

(2.0)

Standard error of regression

38.3

34.7

33.2

33.5

35.2

35.8

Adjusted R 2

0.37

0.43

0.34

0.38

0.38

0.43

N

837

1,363

1,189

1,355

1,033

1,120

Note: Regression coefficients based upon major‐party presidential voters only (with standard errors in parentheses).

(p.54) Given this coding of the variables measuring political predispositions, the intercepts in Table 2.2 represent the relative trait ratings of the competing candidates in each election by “neutral observers”—survey respondents without partisan attachments who are ideologically moderate and who view the state of the national economy as essentially unchanged from the preceding year. If trait assessments were simple reflections of the intrinsic qualities of the candidates, we would expect these intercepts to vary markedly from trait to trait and from year to year with the personal qualities of the two candidates in each election. But we would not expect the coefficients for partisanship, ideology, and economic perceptions to reflect any strong relationship between voters' general political views and their specific assessments of the candidates' traits. In fact, however, there is much more systematic variation evident among voters in each election than there is across election years. This (p.55) pattern is most striking in the case of empathy ratings, where variations in “neutral observer” ratings from year to year amounted to only a few points, but strong partisans on each side gave their respective candidates an edge amounting to about 30 points on the 100 to +100 scale (by comparison with pure independents). For readers enamored of “statistically significant” parameter estimates, the average t‐statistic for the twenty‐eight separate Republican Party Identification coefficients in Table 2.2 is 15.8. The corresponding average t‐statistics for the Conservative Ideology and Economic Assessment coefficients are 4.6 and 5.6, respectively.

The “neutral observer” trait ratings represented by the intercepts in Table 2.2 are presented in graphical form in Figure 2.2.20 The figure differs from the table in presenting separate ratings for each presidential candidate, rather than comparative ratings of the two competing candidates in each election year. However, the basic aim is similar in both cases—to summarize public perceptions of the candidates' personal qualities purged of the biases attributable to partisanship, ideology, and retrospective economic evaluations.

The “neutral observer” ratings in Figure 2.2 differ in some significant respects from the average trait ratings presented in Table 2.1. For example, Jimmy Carter's unusually low ratings with respect to “strong leadership” in Table 2.1 appear to be largely attributable to the circumstances in which he ran for reelection; indeed, the “neutral observer” ratings of five different Democratic candidates in six different election years display very little variation in leadership assessments, suggesting that most of the variation in average leadership ratings in Table 2.1 reflects differences in political context rather than differences in the various candidates' intrinsic personal qualities.

It is striking that the major‐party presidential candidate with the least impressive public image in the twenty years of systematic readings by NES is the one who has so frequently been referred to by journalists as a master campaigner and “once‐in‐a‐lifetime political performer,” Bill Clinton.21 As a challenger in 1992, Clinton was rated only slightly higher than Walter Mondale in 1984 or Michael Dukakis in 1988 as “caring,” and “knowledgeable,” and much lower as “moral.” His higher ratings than his predecessors as “inspiring” and a “strong leader” were largely due to the fact that he ran in a period of widespread economic discontent. What is even more surprising is that, allowing for the differences in circumstances between the two years, Clinton was viewed less favorably in 1996 than in 1992 on four of the five traits: less moral, less caring, less inspiring, and a less strong leader. The only respect in which his image improved during his first term in office is that he (p.56)

                   The Impact of Candidate Traits in American Presidential Elections

Fig. 2.2 “Neutral Observer” Trait Ratings For Presidential Candidates, United States, 1980–2000

was more likely to be considered “knowledgeable”—as were Ronald Reagan in 1984 and George Bush in 1992 after four years in the White House.

In 2000, both of the competing candidates were viewed as fairly typical nominees of their respective parties, with the notable exception of George W. Bush's unusually low rating as “knowledgeable.” Al Gore was perceived as more knowledgeable (by 69 to 59 on the 100‐point scale) and caring (by 52 to (p.57) 47), while Bush was viewed as a stronger leader (by 58 to 51) and slightly more moral (by 64 to 63). Thus, as in each of the other elections considered here, the net impact of candidate traits on the election outcome was dependent on the relative weight attached to different specific traits, and mitigated by the fact that neither candidate enjoyed a more favorable public image than his opponent across the whole range of traits likely to be considered relevant by prospective voters.

The Causal Status of Trait Ratings

My claim that specific impressions of the competing candidates are powerfully shaped by more basic political predispositions is reminiscent of the “funnel of causality” framework proposed by the authors of The American Voter and subsequently elaborated in a series of articles by Miller and Shanks culminating in their book‐length study of The New American Voter.22 Like Miller and Shanks, I argue that any reasonable assessment of the causal impact of candidate traits must take systematic account of the extent to which prospective voters' trait assessments are epiphenomenal.

More specifically, I shall follow Miller and Shanks in constructing a multistage causal model of voting behavior in which basic political predispositions may have powerful indirect effects by virtue of their importance in shaping more specific political opinions and impressions. In Miller and Shanks's formulation, eight distinct categories of explanatory variables are arrayed in a causal sequence with six distinct stages leading up to the ultimate decision to vote for one candidate or another. For my purposes here, that much complexity seems both unnecessary and potentially counterproductive.23 Thus, my own analysis simply distinguishes between two categories of potential influences on voting behavior—assessments of the candidates' personal traits, (p.58) and the more basic political attitudes (party identification, ideology, and economic assessments) employed as explanatory variables in the regression analyses reported in Table 2.2.

I assume here that the strong relationships reported in Table 2.2 between trait assessments on the one hand and partisanship, ideology, and economic assessments on the other reflect the impact of political predispositions on specific impressions of the candidates. In that case, it seems reasonable to discount the apparent impact of trait assessments on voting behavior to allow for the fact that trait assessments are, in significant part, simply reflections of more basic partisan, ideological, and economic views rather than independent reactions to the candidates' intrinsic personal qualities. However, an alternative explanation for the observed relationship between political attitudes and trait ratings is that personal impressions of the candidates influence prospective voters' partisanship and (perhaps less directly) their ideological views and perceptions of economic conditions. If that interpretation is correct, then my analysis will understate the electoral impact of trait perceptions by misattributing to partisan, ideological, and economic bias some of the direct effect of trait assessments on votes. Thus, it seems prudent to determine which of these contrasting causal interpretations seems most tenable.

Making convincing inferences about causal connections between political attitudes and opinions using survey data collected at a single point in time is notoriously difficult. Fortunately, in this case it is possible to bring some additional leverage to bear by analyzing data from repeated surveys with the same respondents over a period of several months. The 1980 National Election Study included an election‐year panel survey in which the same respondents were interviewed in January or February, again in June, and again during the fall campaign. By relating their political attitudes and trait assessments in the fall interview to the answers they gave to the same questions before the first primary votes were cast in early spring, we can gauge the extent to which changes in trait assessments over the course of the election year flow from pre‐existing partisan, ideological, and economic attitudes—and, conversely, the extent to which changes in partisanship, ideology, and economic assessments flow from preexisting attitudes regarding the candidates' personal qualities.

Tables 2.3 and 2.4 present the results of parallel regression analyses in which each of the attitudes employed as explanatory variables in my analysis of vote choice in the 1980 election appear as dependent variables, with the entire set of corresponding attitudes as measured before the first primaries in early spring appearing as explanatory variables. Table 2.3 includes the results for the four fall trait ratings; Table 2.4 provides the parallel results for the fall measures of partisanship, ideology, and economic perceptions. If partisanship, ideology, and economic perceptions are causally prior to candidate trait assessments, as I have asserted, then we should see some tendency in Table 2.3 for trait assessments in the fall to be influenced by political attitudes at the (p.59) beginning of the election year, even with pre‐primary trait assessments included in the regression analyses. Conversely, if partisanship, ideology, and economic perceptions are significantly influenced by perceptions of the candidates' personalities, then we should see significant coefficients for trait ratings in Table 2.4, even with prior levels of partisanship, ideology, and economic perceptions included in the regression analyses.

The results in Table 2.3 provide very strong evidence that candidate trait ratings are affected by partisan bias and fairly consistent evidence for ideological bias as well. On average, and other things being equal, strong Republicans and strong Democrats diverged by about 28 points between January and September in their assessments of Reagan and Carter on each of the four 200‐point relative trait scales. Even with fewer than 500 respondents in the panel analysis, the average t‐statistic for the four separate partisan bias estimates is 5.9. The magnitudes of the corresponding ideological bias estimates are a good deal smaller, and the average t‐statistic for these four effects is only 1.0; nevertheless, the consistency of the estimated effects across three of the four traits makes it very unlikely that they are merely due to chance. By comparison, the estimated effects of economic assessments on subsequent changes in trait ratings are both small and inconsistent, suggesting that trait

Table 2.3. Impact of Pre‐Primary Party Identification, Ideology, Economic Assessment, and Trait Ratings on Fall Trait Ratings, United States, 1980

Moral

Knowledgeable

Inspiring

Strong leader

Intercept

4.9

0.1

6.3

14.4

(2.6)

(2.8)

(3.1)

(3.2)

Republican identification

10.1

9.1

16.6

19.5

(2.0)

(2.2)

(2.5)

(2.5)

Conservative ideology

4.1

5.9

5.0

0.7

(3.6)

(3.8)

(4.3)

(4.4)

Economic assessment

0.9

1.8

2.9

4.2

(3.4)

(3.7)

(4.1)

(4.2)

Moral

0.350

0.076

0.173

0.112

(0.050)

(0.054)

(0.060)

(0.062)

Knowledgeable

0.027

0.263

0.151

0.171

(0.048)

(0.051)

(0.057)

(0.059)

Inspiring

0.036

0.115

0.173

0.167

(0.044)

(0.047)

(0.053)

(0.055)

Strong leader

0.163

0.053

0.194

0.324

(0.045)

(0.049)

(0.055)

(0.057)

Standard error of regression

27.8

30.0

33.6

34.7

Adjusted R 2

0.30

0.27

0.39

0.45

N

478

475

476

477

Note: Regression coefficients based upon major‐party presidential voters only (with standard errors in parentheses).

(p.60)

Table 2.4. Impact of Pre‐Primary Attitudes and Trait Ratings on Fall Party Identification, Ideology, and Economic Assessment, United States, 1980

Republican identification

Conservative ideology

Economic assessment

Intercept

0.052

0.059

0.420

(0.032)

(0.027)

(0.037)

Republican identification

0.888

0.110

0.062

(0.025)

(0.022)

(0.029)

Conservative ideology

0.030

0.551

0.037

(0.044)

(0.038)

(0.052)

Economic assessment

0.061

0.037

0.261

(0.041)

(0.036)

(0.049)

Moral

0.0013

0.00054

0.00059

(0.00061)

(0.00053)

(0.00072)

Knowledgeable

0.00004

0.00004

0.00107

(0.00058)

(0.00050)

(0.00068)

Inspiring

0.00031

0.00097

0.00075

(0.00054)

(0.00046)

(0.00063)

Strong leader

0.00056

0.00005

0.00176

(0.00056)

(0.00048)

(0.00066)

Standard error of regression

0.343

0.297

0.406

Adjusted R 2

0.78

0.42

0.11

N

488

488

489

Note: Regression coefficients based upon major‐party presidential voters only (with standard errors in parentheses).

assessments (with the possible exception of leadership ratings) are insensitive to perceptions of economic conditions over the course of an election year.

The corresponding results in Table 2.4 suggest that the political attitudes treated as causally prior in my analysis are largely but not entirely unaffected by perceptions of the candidates' personal qualities. Six of the twelve relevant parameter estimates (in the bottom half of Table 2.4) have t‐statistics greater than or equal to 1.0, but these potentially significant effects are scattered among the columns and rows of the table in a way that suggests no clear pattern. Even for the most general of the traits, “strong leadership,” the effects on broader political attitudes are small and statistically uncertain (with an average t‐statistic of 1.3).

Taken together, the results presented in Tables 2.3 and 2.4 provide considerable support for treating trait assessments as partly “caused” by party identification and other more basic political attitudes that are themselves largely unaffected by personal reactions to the candidates. Since my primary interest here is in the independent electoral impact of candidates' traits, I will attempt to specify the extent to which the apparent effects of trait assessments reflect the personal qualities of the competing candidates rather than the political biases of the voters doing the assessing. While political biases are by no means (p.61) inconsequential, they seem to me (as to Miller and Shanks) to be more properly counted as effects of partisanship, ideology, or economic considerations than as evidence of the political significance of personal images.

The Impact of Trait Ratings on Voting Behavior

The “neutral observer” trait ratings in Figure 2.2 are intended to provide a summary of each recent presidential candidate's distinctive public image, shorn of positive or negative biases attributable to partisanship, ideology, and economic assessments. My aim in the remainder of this chapter is to gauge the impact of those public images on voting behavior and on the outcomes of recent presidential elections. To that end, Table 2.5 presents the results of separate probit analyses of vote choice in each of the six presidential elections from 1980 through 2000. The analyses are, once again, limited to NES survey respondents who reported voting for a major‐party presidential candidate. The explanatory variables include party identification, ideology, and economic assessments, as well as relative ratings of the competing presidential candidates on the five character traits included most regularly in recent NES surveys.

The parameter estimates for party identification, ideology, and economic assessments in Table 2.5 offer few surprises. All have the expected (positive) sign, and the magnitudes of the estimated effects are reasonably stable from year to year. (The 2000 election is something of an outlier in this respect, with party identification having a stronger impact and ideology and economic assessments having weaker effects than in other recent election years.)

The parameter estimates for the various candidate traits included in Table 2.5 are also uniformly positive, though a few are smaller (and a few more are only slightly larger) than their respective standard errors. These estimates are graphically summarized in Figure 2.3, which shows the estimated impact of each trait in each election year. The magnitudes of the effects vary from year to year, reflecting some combination of election‐specific salience effects and pure sampling error.24 However, the general tendency is for “really cares” to have the largest electoral impact, followed in descending order by “strong leader,” “moral,” “inspiring,” and “knowledgeable.”

The estimated effects of trait ratings on vote choice presented in Table 2.5 are derived from a very simple model in which the weights attached to trait assessments are allowed to vary across traits and across elections, but are assumed to be equal for both candidates and for all voters in any given election. I have also examined somewhat more complicated models in which the competing candidates in each election may be evaluated on the basis of different traits, or in which different voters may attach different weights (p.62)

Table 2.5. Impact of Relative Trait Ratings on Vote Choice, United States, 1980–2000

1980

1984

1988

1992

1996

2000

Intercept

−0.018

0.313

0.237

−0.206

−0.457

−0.027

(0.122)

(0.074)

(0.066)

(0.109)

(0.113)

(0.077)

Republican identification

0.927

0.832

0.963

1.066

0.990

1.317

(0.116)

(0.099)

(0.096)

(0.100)

(0.126)

(0.124)

Conservative ideology

0.554

0.547

0.559

0.974

0.701

0.365

(0.196)

(0.169)

(0.171)

(0.178)

(0.221)

(0.156)

Economic assessment

0.384

0.365

0.203

0.147

0.472

0.093

(0.155)

(0.127)

(0.138)

(0.136)

(0.185)

(0.123)

Really cares

0.01792

0.01198

0.01242

0.01738

0.00930

(0.00222)

(0.00232)

(0.00216)

(0.00230)

(0.00236)

Moral

0.00829

0.00690

0.01057

0.00845

0.00889

0.00890

(0.00236)

(0.00252)

(0.00266)

(0.00197)

(0.00236)

(0.00238)

Knowledgeable

0.00242

0.00585

0.00259

0.00254

0.00772

0.00285

(0.00227)

(0.00214)

(0.00266)

(0.00260)

(0.00345)

(0.00229)

Inspiring

0.00552

0.00177

0.00738

0.00309

0.00519

(0.00223)

(0.00214)

(0.00234)

(0.00228)

(0.00267)

Strong leader

0.01700

0.01100

0.01128

0.01119

0.00256

0.01476

(0.00229)

(0.00221)

(0.00232)

(0.00231)

(0.00279)

(0.00237)

Log likelihood

−236.2

−326.6

−306.0

−285.0

−187.6

−234.6

Pseudo‐R 2

0.59

0.65

0.63

0.69

0.73

0.70

N

835

1,355

1,183

1,349

1,031

1,120

Note: Probit coefficients for (Republican) presidential vote choice (with standard errors in parentheses).

(p.63)
                   The Impact of Candidate Traits in American Presidential Elections

Fig. 2.3 The Impact Of Candidate Traits On Presidential Votes, United States, 1980–2000

Note: The election years were 1980, 1984, 1988, 1992, 1996, and 2000.

to the same trait.25 However, these more elaborate models produce only modest improvements in fit over the simpler model presented in Table 2.5,26 and their implications with respect to the aggregate impact of candidate traits on election outcomes are quite similar to those reported here.27 Thus, (p.64) my conclusions seem to be surprisingly insensitive to the precise form of the candidate trait effects represented in Table 2.5.

The Impact of Trait Ratings on Election Outcomes

The parameter estimates reported in Table 2.5, and presented graphically in Figure 2.3, indicate the extent to which individual voters in each election year were swayed by their assessments of the competing presidential candidates' specific personal qualities. However, the reactions of individual voters may or may not add up to a consequential political outcome. For one thing, any given voter may attach significant weight to a variety of candidate characteristics but favor the Democratic candidate in some respects and the Republican candidate in others, with little net impact one way or the other on her eventual vote choice. Then too, many voters who do perceive a consistent personal advantage for one candidate over the other will already be very likely to support that candidate on other grounds, so that “image” considerations have little behavioral impact at the margin. Finally, and perhaps most importantly, a great deal of personal enthusiasm for each candidate is likely to be canceled out, in the electorate as a whole, by other voters' equally genuine enthusiasm for his opponent. Only when large numbers of more or less uncommitted voters are in substantial agreement regarding the personal superiority of one candidate to the other are their assessments likely to have a significant impact on the election result.

Gauging the impact of candidates' personal traits in a given election requires an assessment of how that election would have turned out if the candidates had had different personal qualities from those they actually had. The most relevant counterfactual would seem to be a (hypothetical) election in which the competing candidates were viewed equally favorably (or unfavorably) on each potentially relevant trait dimension by a “neutral observer” unaffected by partisan, ideological, or economic biases. To the extent that the actual election outcome departs from the hypothetical outcome, it seems reasonable to conclude that the candidates' personalities had a politically consequential impact. Thus, my approach here is to compare the estimated probability of casting a Republican vote for each voter in each NES survey, as estimated from the probit analyses reported in Table 2.5, with a counter‐factual probability calculated by adjusting the voter's relative trait assessments to remove the effect of differences in the competing candidates' personal qualities as judged by “neutral observers”—differences of the sort illustrated in Figure 2.2. I take the difference between these two probabilities to represent the impact of the candidates' “objective” personal strengths and weaknesses on each voter's choice, and the average difference for all the voters in a given election year to represent the net electoral impact of candidate traits on the outcome of that election.

(p.65) My estimates of the net impact of candidate traits on the outcomes of the six presidential elections considered here are presented in Table 2.6. The estimate in each case is of the net impact of trait assessments on the Republican share of the major‐party presidential vote; thus, a positive entry implies that the election outcome was more favorable to the Republican candidate than it would have been if “neutral observers” had perceived the two candidates as equally attractive on each trait dimension, while a negative entry implies a net personal advantage for the Democratic candidate.

The most striking feature of the estimates in Table 2.6 is that the net effects of candidate trait assessments are generally quite modest in magnitude. The average effect for the six elections is 1.6 percentage points, and the largest effect (in 1992) is only 3.5 percentage points. By comparison, the average margin of victory in these six elections (that is, the winning candidate's plurality of the two‐party popular vote, including George W. Bush's small negative plurality in 2000) was 8.7 percentage points, and the average inter‐election vote swing was about 5 percentage points. In three of the six elections—including the two with the largest net trait effects—the winning candidate would quite probably have won by a larger margin had personal qualities played no role in determining the election outcome. The only case in which it seems at all likely that perceptions of the candidates' personal qualities had a decisive impact on the election outcome is the 2000 election, where Bush's half‐point advantage with respect to candidate traits was probably one of many “decisive” factors contributing to his razor‐thin victory.

A second important point to note about the estimates of net electoral impact in Table 2.6 is that they comport rather poorly with conventional wisdom regarding the political and personal qualities of the various candidates

Table 2.6. Net Electoral Impact of Candidate Traits, United States, 1980–2000

Estimated Republican vote gain or loss (%)

Candidate advantaged

Winning candidate

Decisive impact?

1980

+ 1.0

Reagan

Reagan

No

(1.0)

(55.3%)

1984

+ 0.5

Reagan

Reagan

No

(0.8)

(59.2%)

1988

−1.5

Dukakis

Bush

No

(0.5)

(53.9%)

1992

+ 3.5

Bush

Clinton

No

(0.7)

(53.5%)

1996

+ 2.6

Dole

Clinton

No

(0.9)

(54.6%)

2000

+ 0.4

Bush

Bush

Probably

(0.5)

(49.7%)

Note: Estimated net impact of candidate traits on Republican share of two‐party presidential vote (jackknife estimates with standard errors in parentheses).

(p.66) in these six presidential elections. Ronald Reagan, the “great communicator” and consummate actor, had a net advantage of only 1 percentage point over Jimmy Carter in 1980 and a net advantage of less than 1 percentage point over Walter Mondale in 1984. Michael Dukakis, widely regarded as cold and politically inept, appears here to have been the only Democratic presidential candidate in the past two decades to enjoy a more favorable personal image than his Republican opponent. And Bill Clinton, “a once‐in‐a‐lifetime political performer,”28 appears to have had the worst personal image of any presidential candidate in recent American history, at least as measured by the impact of trait assessments on election outcomes. The estimates in Table 2.6 suggest that Clinton lost about 3.5 percentage points to George Bush in 1992 and about 2.5 percentage points to Bob Dole in 1996.

The assessments of electoral impact offered in Table 2.6 differ in some important details from those produced by Miller and Shanks for some of the same elections using much the same data and similar methods of analysis, and summarized in Table 1.2 of the present volume. For example, Miller and Shanks estimated that Jimmy Carter had a slight net advantage over Ronald Reagan due to trait assessments in 1980 and that the independent contribution of personal qualities was “close to invisible” in 1992.29 My estimates in Table 2.6 imply that Reagan probably had a slight net advantage over Carter in 1980 (increasing Reagan's vote share by about 1 percentage point) and that George Bush had an unusually large net advantage over Bill Clinton in 1992 (increasing Bush's vote share by about 3.5 percentage points).30

In some cases, these differences reflect differences in the specific set of candidate traits included in each analysis; whereas I focus here on the subset of traits included most consistently in the NES surveys since 1980, Miller and Shanks considered all of the traits included in each election year, eventually dropping those with statistically “insignificant” effects. The differences in results also reflect different assumptions about the impact of other political factors Miller and Shanks's analysis included a wider variety of explanatory factors and a more heroic set of assumptions regarding the causal ordering of those explanatory factors than I employ here.

Despite these differences in detail, my analysis generally tends to confirm and reinforce Miller and Shanks's central claims that “[m]ost of the statistical (p.67) relationships between vote choice and comparative evaluations of the candidates' personal qualities should be seen as misleading or spurious” and that “media emphasis on candidates' personal attributes and voters' volunteered appraisals of candidates' personalities should not be taken at face value” by electoral analysts.31 Voters' impressions of the candidates' personalities are strongly shaped by more basic political predispositions. A candidate's net advantage with respect to one personal trait is not infrequently offset by a net disadvantage with respect to some other equally important trait. And pundits' assessments of the candidates' personal qualities, and of the electoral relevance of those qualities, are likely to be significantly distorted by post hoc reasoning from observed effects to putative causes. For all of these reasons, the electoral impact of candidates' traits is likely to be less than meets the eye.

Bush Vs. Gore

If the estimated candidate trait effects in Table 2.6 are even approximately right, how could conventional wisdom about the personal qualities of recent presidential candidates—and about the impact of those qualities on election outcomes—be so wrong? The crucial point, I suspect, is that in every election conventional wisdom seems to have overestimated the relative personal qualities of the winning candidate. Ronald Reagan must have been a strong, charismatic candidate because he defeated a sitting president and then won reelection in a landslide. Michael Dukakis must have been cold and inept because he blew a double‐digit lead in the polls. Bill Clinton must have been an inspiring, empathetic political performer because he returned the Democrats to the White House after the long hiatus of the Reagan era. In each case, the post hoc assessment of the candidates' personal strengths and weaknesses simultaneously validated and was validated by the notion that, as Quindlen put it, “voters use their impressions of a candidate's personality to choose a president.”

Quindlen's own application of the “cult of personality” thesis to the 2000 presidential campaign hints at how the thesis can easily become self‐fulfilling. “In the next 90 days,” she wrote in August 2000, “millions of people will decide, finally, whether they think Al Gore is rigid and humorless or instead serious and diligent, whether George W. Bush is straight‐talking and sure of himself or simply arrogant and tactless.” How are we to know what they decided? Presumably, since “personality is key in Election 2000,” by seeing how they voted.

The potential fluidity of assessments of candidates' personal strengths and weaknesses was clearly illustrated by changing impressions of the competing candidates over the course of the 2000 campaign. Al Gore was widely (p.68) portrayed as rigid and humorless when Bill Bradley seemed likely to derail his nomination, but became serious and diligent when he surged in the polls following the Democratic convention. Conversely, George W. Bush was affable and charismatic through most of the spring and summer before becoming lazy and confused in late August and September—that is, at just the point when he began to trail Gore in the polls. As one prominent reporter admitted,

There is something of a circular phenomenon here. Mr. Bush's gaffes on the stump are nothing new, but they are being picked up more by the media because the context of the campaign has changed and the governor is no longer viewed as the towering favorite.32

In the end, of course, Bush did manage to win the election—not in the national popular vote, but in the Electoral College (probably), and, more importantly, in the U.S. Supreme Court. Thus, analysts and commentators after the election focused on “explaining” his victory and, even more, Gore's defeat. How could the nominee of the incumbent party not have triumphed easily in a period of peace and apparent economic prosperity? One common explanation focused on Gore's putative personal weaknesses—including many of the same weaknesses that had loomed large the previous spring but receded with Gore's rise in the polls in late summer and early fall. Gore was too wooden; prone to fibbing; not a strong leader. As Berke himself put it, Gore was “a flawed candidate who squandered a prime opportunity to capture the White House.”33

On the one hand, the estimate presented in Table 2.6 suggests that Gore was a less attractive candidate than Bush, all in all, but only by less than 0.5 percentage points. Of course, given the extraordinary closeness of the election, it is quite probable that that sliver of personal advantage was essential to Bush's victory. In an election “decided” by a few hundred votes in Florida, any advantage, however slight, can reasonably be considered decisive. On the other hand, given the precision of the estimate in Table 2.6, it is also quite possible that Gore rather than Bush was the “real” winner of the 2000 popularity contest. (The relevant probability is a little over 20 percent.) What is clear is that neither candidate enjoyed a major electoral advantage on the basis of personality. Even more than in the other presidential elections covered by my analysis, the importance of candidate images among the various factors shaping the outcome of the 2000 election outcome was remarkably modest.

Partisans of candidate‐centered politics may resist this conclusion on the grounds that the specific candidate traits included in Table 2.5 did not really tap the dimensions of personal evaluation that were most salient in the 2000 election. In order to test that possibility, I repeated my analysis of the 2000 outcome using a somewhat richer collection of trait assessments than in the analysis reported in Table 2.5. Respondents in the 2000 NES survey were (p.69) invited to rate the presidential candidates as “dishonest,” “intelligent,” and “out of touch” in addition to the four traits included in Table 2.5.

Perhaps unsurprisingly, given the events of the campaign, Bush had a modest (5‐point) advantage (among major‐party voters) with respect to perceived honesty, while Gore had a modest (7‐point) advantage with respect to perceived intelligence. Gore was also perceived as slightly more “in touch.” Adding voters' comparative assessments of the candidates on those traits to the set of explanatory variables included in Table 2.5 produced a significant estimated effect for being “in touch” (roughly similar in magnitude to the effects for “moral” and “really cares”), a smaller estimated effect (about half as large) for honesty, and an even smaller (indeed, negative, albeit imprecise) estimated effect for intelligence.

What is most important for my purposes here is that the net impact of these additional trait ratings on the outcome of the 2000 presidential election seems to have been very modest. Whereas the calculation reported in Table 2.6, based upon four comparative trait assessments, suggested that Bush gained about 0.4 percentage points due to voters' reactions to the candidates' personal qualities, the corresponding estimate based upon all seven comparative trait assessments is only slightly larger—about 0.6 percentage points.34 Needless to say, this estimate is statistically indistinguishable from the estimate presented in Table 2.6, and, for that matter, only roughly distinguishable from zero; nevertheless, it is precise enough to make it quite clear that the electorate was not swayed far in either direction by the candidates' personalities.

In short, my analysis suggests that the 2000 election was a virtual dead heat with respect to the candidates' personal qualities, as in almost every other respect. What is surprising is not that the electoral impact of candidate traits in 2000 was modest, since that has generally been the case in recent presidential elections. What is surprising is that in 2000 the modest effect of candidate traits was, quite probably, large enough to be decisive.

Acknowledgment

The research reported in this chapter was supported by Princeton University's Woodrow Wilson School of Public and International Affairs. Mark Fischle, Tali Mendelberg, Karen Stenner, and participants in seminars at Princeton and at the University of Washington provided helpful reactions to earlier versions of the analysis.

Notes:

(1) V. O. Key, Jr., The Responsible Electorate: Rationality in Presidential Voting 1936–1960 (New York: Vintage Books, 1966), p. 7.

(2) Thomas E. Patterson and Robert D. McClure, The Unseeing Eye: The Myth of Television Power in National Elections (New York: G. P. Putnam's Sons, 1976).

(3) Joe McGinniss, The Selling of the President 1968 (New York: Trident Press, 1969); Gene Wyckoff, The Image Candidates (New York: Macmillan, 1968); Dan Nimmo, The Political Persuaders: The Techniques of Modern Election Campaigns (Englewood Cliffs, N.J.: Prentice‐Hall, 1970); Harold Mendelsohn and Irving Crespi, Polls, Television, and the New Politics (Scranton, Penn.: Chandler, 1970).

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

(5) For example, Donald E. Stokes, “Party Loyalty and the Likelihood of Deviating Elections,” Journal of Politics, 24 (1962), 689–702; Donald E. Stokes, “Some Dynamic Elements of Contests for the Presidency,” American Political Science Review, 60 (1966), 19–28.

(6) Martin P. Wattenberg, The Rise of Candidate‐Centered Politics: Presidential Elections of the 1980s (Cambridge, Mass.: Harvard University Press, 1991).

(7) David Butler and Austin Ranney, eds., Electioneering: A Comparative Study of Continuity and Change (Oxford: Clarendon Press, 1992), p. 281.

(8) Anna Quindlen, “It's the cult of personality,” Newsweek, August 14, 2000, p. 68.

(9) Frank Bruni, “Bush insists to voters his blood is red, not blue,” New York Times, April 24, 2000, p. A19.

(10) The data utilized here, along with details of the study design, codebooks, and other relevant information, are publicly available through the NES website, www.umich.edu/∼nes.

(11) Donald R. Kinder, Mark D. Peters, Robert P. Abelson, and Susan T. Fiske, “Presidential Prototypes,” Political Behavior, 2 (1980), 315–37; Robert P. Abelson, Donald R. Kinder, Mark D. Peters, and Susan T. Fiske, “Affective and Semantic Components in Political Person Perception,” Journal of Personality and Social Psychology, 42 (1982), 619–30; Donald R. Kinder, “Presidential Character Revisited,” in Richard R. Lau and David O. Sears, eds., Political Cognition (Hillsdale, N.J.: Lawrence Erlbaum Associates, 1986); Donald R. Kinder and Susan T. Fiske,“Presidents in the Public Mind,” in Margaret G. Hermann, ed., Political Psychology: Contemporary Problems and Issues (San Francisco: Jossey‐Bass, 1986); Arthur H. Miller, Martin P. Wattenberg, and Oksana Malanchuk, “Schematic Assessment of Presidential Candidates,” American Political Science Review, 80 (1986), 521–40.

(12) Miller and Shanks worried that “one of these questions (concerning ‘strong leadership’) presents a somewhat different criterion for evaluation than a ‘personal’ or non‐governmental quality, for it involves an implicit standard that is hard to distinguish from ‘would be a good President.’ ” See Warren E. Miller and J. Merrill Shanks, The New American Voter (Cambridge, Mass.: Harvard University Press, 1996). While recognizing the logical force of that concern, I treat the “strong leader” trait in the same way as the others in the following analysis, for two reasons. First, assessments of the candidates' leadership qualities do not turn out to be markedly more strongly related to broader political views, or to vote choices, than are some of the other trait assessments considered here. Second, including assessments of the candidates' leadership qualities in my battery of trait ratings will, if anything, exaggerate the impact of trait ratings on voting behavior, which will tend to work against the thrust of my argument that candidates' personal qualities generally have quite modest electoral effects.

(13) I repeated all of the analyses reported in this chapter using random subsets of the five‐trait battery. The estimated effects of trait assessments on specific election outcomes were generally similar to those reported here, with only a very slight tendency for more extensive trait batteries to produce larger estimated effects. In the case of the 2000 election I also examined the impact of including additional traits beyond the five included regularly in previous NES surveys; the results of that exercise (detailed in due course) likewise suggest that the precise set of traits included in the analysis is relatively inconsequential.

(14) Alternative recoding schemes reflecting different assumptions about the appropriate relative distances between response categories produce results essentially similar to those reported here.

(15) My analysis also excludes approximately 10 percent of the preelection respondents in each year who, for one reason or another, could not be reinterviewed after the election.

(16) Patterson and McClure, The Unseeing Eye, pp. 66, 114.

(17) Patterson and McClure, The Unseeing Eye, p. 66.

(18) Focusing on relative trait ratings for competing candidates (as in Table 2.2) rather than on assessments of individual candidates (as in Table 2.1) simplifies the analysis and presentation, and has the additional advantage of rendering the data impervious to idiosyncratic individual differences in the overall “generosity” of trait ratings. See Henry E. Brady, “The Perils of Survey Research: Inter‐Personally Incomparable Responses,” Political Methodology, 11 (1985), 269–91. The corresponding cost is that systematic differences in the sources or electoral impact of the two candidates' images cannot be detected from an analysis focusing on differences in trait ratings—a point I return to subsequently.

(19) In the case of economic perceptions, responses are scored differently depending upon which party controls the White House in order to produce the correct partisan valences in the regression analyses reported in Table 2.2. In election years in which a Republican president was in office (1984, 1988, 1992) respondents who said the economy was “much better” get a score of +1, those who said the economy was “about the same” get a score of zero, and those who said the economy was “much worse” get a score of − 1. In election years in which a Democratic president was in office (1980, 1996, 2000) respondents who said the economy was “much better” get a score of − 1, those who said the economy was “about the same” get a score of zero, and those who said the economy was “much worse” get a score of + 1. Given this coding scheme, the expected impact of economic perceptions on relative (Republican minus Democratic) trait ratings under the hypothesis of partisan bias is positive in every election year.

(20) The “neutral observer” ratings for assessments of each candidate as “inspiring” are omitted from Figure 2.2 in order to conserve space. They generally parallel the “strong leader” ratings, except that George Bush in 1992 and Bob Dole in 1996 were each viewed as being slightly less inspiring than Bill Clinton.

(21) Caryn James, “Presenting a masterpiece in political theater, with a scene from ‘Camelot’,” New York Times, August 16, 2000, p. A26.

(22) Miller and Shanks, The New American Voter. See also Warren E. Miller and J. Merrill Shanks, “Policy Directions and Presidential Leadership: Alternative Interpretations of the 1980 Presidential Election,” British Journal of Political Science, 12 (1982), 299–356; J. Merrill Shanks and Warren E. Miller, “Policy Direction and Performance Evaluation: Complementary Explanations of the Reagan Election,” British Journal of Political Science, 20 (1990), 143–235; J. Merrill Shanks and Warren E. Miller, “Partisanship, Policy and Performance: The Reagan Legacy in the 1988 Election,” British Journal of Political Science, 21 (1991), 129–97.

(23) Miller and Shanks, The New American Voter, chap. 8. Miller and Shanks included a much richer set of explanatory variables in their analyses than I employ here, but at the cost of making a variety of potentially problematic assumptions about causal ordering. For example, on the one hand, their analytic framework treated voters' retrospective evaluations of presidential performance as potential causes, but not potential consequences, of their impressions of the candidates' personal qualities; on the other hand, voters' prospective evaluations of how the competing candidates would perform in office were treated as potential consequences, but not potential causes, of their assessments of the candidates' personal traits. In the absence of good evidence regarding the nature of the causal relationships among these various explanatory factors, it seems desirable to simplify the analysis as much as possible by limiting the number of explanatory variables and assumptions of causal priority.

(24) Since the standard errors of the probit coefficients range from about 0.002 to 0.003, many of the year‐to‐year variations in the impact of each trait evident in Figure 2.3 are too large to be plausibly attributable solely to sampling error.

(25) Each of these more elaborate model specifications included twice as many trait effects as the basic model presented in Table 2.5. The first alternative model included separate trait effects for Republican and Democratic candidates in each election. The second alternative model included interactions between each relative trait rating and voters' party identification, allowing Republican and Democratic voters to attach different weights to each trait in each election. The third alternative model included interactions between each relative rating and the strength of voters' partisanship, allowing strong partisans and independents to attach different weights to each trait in each election. I am grateful to Tali Mendelberg, Karen Stenner, and Mark Fischle for suggesting these alternative model specifications.

(26) Only three of the eighteen likelihood ratio tests (for the three alternative models applied to each of six elections) reflect “statistically significant” improvements in fit (at the conventional 0.05 level), and the average p‐values (over the six elections) range from 0.15 for the second (partyspecific) alternative specification to 0.53 for the third (strength‐of‐partisanship) model.

(27) For the second (party‐specific) and third (strength‐of‐partisanship) alternative models the estimated net effects of candidate traits are practically identical to those for the basic model, never differing by more than 0.2 percentage points. The first (candidate‐specific) alternative model produces somewhat different estimates, but even these differ from the corresponding estimates in Table 2.6 by an average of less than 1 percentage point, and the average magnitude of the estimated net effects is only 1.5 percentage points (as against 1.6 percentage points for the simpler model in Table 2.6).

(28) See James, “Presenting a Masterpiece in Political Theater.”

(29) Miller and Shanks, “Policy Directions and Presidential Leadership;” Miller and Shanks, The New American Voter, p. 480.

(30) Additional differences appear with respect to individual trait assessments in specific election years. For example, Miller and Shanks reported in The New American Voter (p. 425) that “honest,” “inspiring,” and “really cares” were “the only traits for which the comparative evaluations of the two candidates . . . exhibited a significant relationship to the vote” in 1992, net of other factors; the corresponding results in Table 2.5 show fairly strong and statistically significant effects for “really cares,” “strong leader,” and “moral” and considerably weaker effects for “inspiring” and “knowledgeable.” (Assessments of honesty are not included in my analysis because that trait did not appear consistently in the NES surveys employed here.)

(31) Miller and Shanks, The New American Voter, pp. 501–2.

(32) Richard L. Berke, “Tested and occasionally tripped, Bush may yet rue a mirror crack'd,” New York Times, September 18, 2000, p. A16.

(33) Richard L. Berke, “Many seem skeptical of Gore's future,” New York Times, December 17, 2000, p. A1.

(34) The precise estimate is 0.59 percentage points (pro‐Bush), which is in reasonably close agreement with the out‐of‐sample estimation error of 0.56 percentage points (pro‐Bush) from Bartels and Zaller's analysis of the 2000 presidential election in historical perspective. See Larry M. Bartels and John Zaller, “Presidential Vote Models: A Recount,” PS: Political Science & Politics, 34 (2001), 9–20. The agreement is, of course, coincidental; but the results presented here lend some support to Bartels and Zaller's assertion (p. 19) that there is “no need, and little warrant, to posit either unusual incompetence on Gore's part or unusual skill on Bush's part” in order to account for the outcome of the 2000 election.