The Compassion Strategy: Race and the Gender Gap in American Politics

Please Do Not Quote or Cite Without Permission

Vincent L. Hutchings

Nicholas A. Valentino

Tasha S. Philpot

Ismail K. White

University of Michigan

Vincent L. Hutchings is assistant professor of Political Science, and faculty associate in the

Center for Political Studies, at the University of Michigan. He can be contacted at (734) 764-6591

or via email at vincenth@umich.edu. Mailing address is Institute for Social Research, University

of Michigan, Center for Political Studies, 426 Thompson Street, P.O. Box 1248, Ann Arbor, MI

48106-1248.

Nicholas Valentino is assistant professor of Communications Studies, and faculty associate in the

Center for Political Studies, at the University of Michigan.

Tasha Philpot is a Ph. D. candidate in the Political Science Department at the University of

Michigan.

Ismail White is a Ph. D candidate in the Political Science Department at the University of

Michigan.

This research would not have been possible without the efforts of several graduate research

assistants including Dmitri Williams, Lara Rusch, Matthew Beckmann and Dara Faris.

2

Abstract

Several studies have shown that social “compassion” issues, and not those directly linked

to women’s interests, seem to drive the gender gap in recent American presidential elections.

Scholars have also found that many of these compassion issues are often associated with the

plight of racial minorities. We explore whether traditional partisan appeals to African Americans

may help to explain the gender gap. We hypothesize that this gap is driven by the different

reactions that white men and women have to partisan efforts to address racial inequities. In an

experiment, we manipulate news information regarding George W. Bush’s commitment to blacks

versus women. We find, as expected, that the gender gap is maximized when Bush takes the

traditional Republican stance on race, but not when he adopts more conservative positions on

women’s issues. Moreover, the gap is significantly reduced when Bush espouses a more moderate

position on race. Finally, results from survey analyses of the impact of exposure to the 2000

Republican National Convention suggest that Bush’s efforts to highlight the party’s commitment

to racial diversity was most effective among white women.

3

Since 1980, political pundits have recognized the presence of a “gender gap” in presidential

elections (Erikson and Tedin 1995). That is, women are typically more supportive than men of the

Democratic nominee for president. The magnitude of this gap has varied from about 6-percentage points

in 1980 to about 15-percentage points in 1996 (Norrander 1999a). Women have also demonstrated

greater allegiance to the Democratic Party. This gender gap in partisanship reached double-digits after

1994 (Norrander 1999a). Despite the stability of these gender gaps in recent decades, there is not yet

consensus as to their principal cause. Though we will review the literature on the gender gap below,

suffice it to say that no study to date has emphasized the divergent positions of the parties on race-

relevant issues as an important causal factor. In this paper, we explore whether the relationship between

these partisan cues and the gender gap has been overlooked.

Our speculation that partisan racial cues contribute to the gender gap derives from four generally

uncontested facts. First, the gender gap initially emerges when the major parties began to diverge on

questions of race. Secondly, this divergence prompts, or at least precedes, the widespread defection from

the Democratic Party of white men, especially southerners, which disproportionately contributes to the

rise of the gender gap (Miller and Shanks 1996; Norrander 1999b). Third, on most questions concerning

racial policies, women are generally more liberal than men (Norrander 1999a; Schuman, Steeh, Bobo, and

Krysan 1997). And finally, one of the chief sources of the gender gap derives from the greater support

among women for so-called “compassion issues” such as health care, education, efforts to alleviate

poverty, and aid to minorities (Chaney, Alvarez and Nagler 1998; Kaufmann and Petrocik 1999;

Norrander 1999a). These compassion issues---even ostensibly non-racial ones---have become linked with

attitudes about racial minorities in the minds of many Americans (Edsall and Edsall 1991; Gilens 1999;

Kinder and Sanders 1996; Mendelberg 2001; Valentino, Hutchings, and White 2002). Thus, women’s

greater concern for the disadvantaged, and the obvious racial implications of this concern, may explain

why Democrats consistently outperform Republicans when it comes to mobilizing this constituency.

Our thesis is that the salience of race in American politics produces an opportunity for candidates

to effectively convey a trait that resonates with women: compassion. Candidates can effectively influence

4

their support among white women by courting, or failing to court, African American voters. This

argument can be used to understand the efforts of George W. Bush during the 2000 presidential campaign

to appear more racially inclusive: His “compassionate conservatism” was aimed most directly at courting

white women, not people of color.1

This explanation for the emergence and maintenance of the gender gap has important

implications for the study of public opinion and voting behavior. If, as we argue, the gender gap owes

much of its force to the parties long-standing reputations on matters related to race, then this provides

strong support for scholars who have argued that this issue permeates contemporary American politics

(Bobo 2000; Carmines and Stimson 1989; Edsall and Edsall 1991; Kinder and Sanders 1996; Sears, van

Laar, Carrillo, and Kosterman 1997; Valentino and Sears 2001; Valentino, Hutchings, and White 2002).

Since party platforms are constrained by their supporters, actual policy stands are not likely to change

substantively in the foreseeable future. Consequently, the gender gap will likely remain a feature of

American politics for some time. Indeed, as a younger cohort enters the electorate in an era where the

Republican Party is increasingly dominated by white southern men and the Democratic Party by women

and racial minorities, this gap may grow.

We begin our analyses by reviewing the literature on public opinion with an eye towards

explaining why men and women might view racial matters differently. Next, we introduce an

experimental design that tests our hypotheses about partisan differences on racial issues and the

development of the gender gap. We explore the impact of our experimental manipulations on a range of

evaluative dimensions, including perceptions of which candidate is most compassionate, responsive to the

interests of blacks and women, and, ultimately, worthy of electoral support. Then we turn to an analysis

1 A number of observers of the 2000 election also believed that racial minorities were not necessarily the target of the GOP’s outreach efforts. For example, The New York Post noted that Bush was seeking to broaden the appeal of his party because “…many white suburban voters, especially women, want a candidate who will reach out to minorities.” Orin, Deborah. “Bush Hopes Latino Kin’s Spots Ad Up to Votes.” The New York Post Saturday, June 10, 2000. Similarly, David Bositis, senior analyst at the Joint Center for Political and Economic Studies, suggested that “[the Republicans were] after the moderate voters, the independents, the whites who are turned off by the party’s intolerant and insensitive image.”

5

of survey data on the 2000 presidential campaign, where we explore the effectiveness of the Republicans’

message of racial inclusion on white men versus white women.

Racial Issues And The Gender Gap

In order to examine the role that racial issues play in the gender gap we must begin with a review

of the evidence regarding the broader political significance of race in America. Arguably the most

prominent cleavages in contemporary American politics center around policies designed to assist racial

minorities. On issues such as affirmative action, black and white opinions can differ by as much as 30-to-

50 percentage points (Dawson 1994; Kinder and Sanders 1996; Kinder and Winter 2001; Schuman, et al.

1997). On racialized social policies such as criminal justice, welfare, and the minimum wage, racial

differences are almost as large. African Americans and whites typically identify with different political

parties, and vote for different candidates (Mendelberg 2001; Pomper 2001; Tate 1993). This divide is so

far reaching that some scholars regard it as fundamental to understanding historical and contemporary

party realignments (Carmines and Stimson 1989; Kinder and Sanders 1996; Valentino and Sears 2001).

The gaps between men and women on these issues are smaller, but persistent (Frankovic 1982;

Kaufman and Petrocik 1999; Norrander 1999a, 1999b; Wirls 1986). Women are more likely than men to

identify as Democrats and liberals, and more likely to support the Democratic candidate for president.

Analysts first noticed this phenomenon in the presidential contest between Ronald Reagan and Jimmy

Carter in 1980. Since that election, gender gaps in presidential vote choice and party identification have

fluctuated, but have remained consistently large (i.e. 10% or greater) in each of the last three presidential

elections.

There is little consensus about the principal source of these gender gaps, and at least three general

sets of explanations have been advanced. During the Reagan era, most explanations focused on gender

differences in the use of military force. Given the prominence of this issue during the 1980’s, it is not

Rodriguez, Lori. “Republican Convention Philadelphia 2000; Parade of diversity no sign GOP has changed foes say; Supporter say it’s a step in the right direction.” Houston Chronicle Friday, August 4, 2000.

6

surprising that this issue accounted for a significant portion of the gender gap (Conover and Sapiro 1993;

Frankovic 1982; Gilens 1988; Smith 1984). More recently, however, this dimension has been less

important (Chaney, Alvarez, and Nagler 1998). Another body of research has focused on women’s

issues, such as abortion or the Equal Rights Amendment, as motivating the gender gap (Conover 1988;

Mansbridge 1985), although subsequent analyses have found that these attitudes influence men as well as

women (Cook and Wilcox 1991; Kaufmann and Petrocik 1999). Finally, some studies emphasize gender

differences in attitudes about the economy or social welfare issues (Chaney et al. 1998; Welch and

Hibbing 1992). These “compassion” issues have been found to be the most powerful explanation for the

gender gap in recent elections (Kaufmann and Petrocik 1999).

Another issue domain that divides Democrats and Republicans as well as men and women is race.

The major parties have diverged sharply on policies designed to assist African Americans since 1964

(Carmines and Stimson 1989; Kinder and Sanders 1996; Mendelberg 2001). Moreover, since the 1970’s,

women have generally been more liberal than men on a number of indicators of racial attitudes (Shapiro

and Majahan 1986; Norrander 1999a). This is particularly true for questions on structural explanations

for racial inequality and support for affirmative action policies, but less so for questions tapping support

for intermarriage or integrated schools (Schuman, et al. 1997). This latter result may, however, be less

true of younger cohorts of women (Johnson and Mooney 1998).

To date, researchers have not seriously considered the possibility that race may be implicated in

the gender gap through so called “compassion” issues. For example, Kaufmann and Petrocik (1999) note

that issues such as social spending in general, the proper role of government, spending on the poor, and

aid to blacks contribute, collectively, much more to the gender gap in 1992 and 1996 than any other issue.

They do not, however, highlight the fact that race is intimately linked with each of these issues. Gilens

(1999) and Mendelberg (2001), for example, have reported that issues of welfare and poverty have

become racialized among both elites and the mass public. Similarly, Kinder and Sanders (1996) find that

attitudes about blacks influence policy positions on a number of issues including government spending,

7

family leave, and childcare. Finally, Edsall and Edsall (1991) argue that, since the Reagan era,

conservative rhetoric has linked the discussion of taxes and government spending on social services to

unpopular minority groups. Recent work by Valentino and his colleagues (2002) demonstrates that

strategic politicians can successfully prime racial attitudes simply by invoking these issues.

Consequently, the influence of social welfare attitudes on the gender gap may at bottom, derive from the

contrasting positions each party takes on matters of race.

There are additional reasons to expect that party reputations on racial issues might contribute to

the gender gap. These stem from when the gender gap first emerged and which voters disproportionately

contribute to it. First, contrary to popular perceptions, the gender gap did not begin in 1980. In fact, it

may have appeared as far back as the 1950’s (Kaufmann and Petrocik 1999; Norrander 1999a). However,

the earliest surveys indicate that it was men, not women, who preferred the Democrats. This gender gap

reverses with the 1964 presidential contest. For the first time, women were slightly more attracted to the

Democratic candidate than were men. This is significant because the election of 1964 also signified the

moment when the contemporary parties began to diverge on matters of race (Carmines and Stimson 1989;

Kinder and Sanders 1996). This co-occurrence is not, of course, proof that the gender gap emerges

because of race. However, subsequent elections provide further circumstantial evidence. For example,

the traditional gender gap emerges yet again in 1968, but only for the candidacy of the segregationist,

George Wallace. The gender gap is also present in 1972, just as Nixon hones his pivotal “southern

strategy.” It all but disappears in 1976, in the contest between a moderate Midwestern Republican

(Gerald Ford) and a Democratic governor from the Deep South (Jimmy Carter), neither of whom made

race central in the first election following Watergate. Finally it reemerges in 1980, with the nomination of

the “law-and-order” candidate, Ronald Reagan.

Second, although the gender gap is frequently interpreted as women’s greater support for the

Democrats, men’s support for the Republican Party also plays a central role (Kaufmann and Petrocik

1999; Norrander 1999a, 1999b; Wirls 1986). Moreover, it is white southern men who have become

especially attracted to the Republican message over the last few decades (Norrander 1999b; Miller 1991;

8

Miller and Shanks 1996). Although scholars disagree as to the precise role that the partys’ contrasting

stands on racial policy contributed to this secular realignment, it seems clear that it had some influence

(Abramowitz 1994; Abramowitz and Saunders 1998; Carmines and Stimson 1989; Valentino and Sears

2001). Consequently, it would appear that, as the Republican Party grew more racially conservative, this

transformation appealed more to white men than white women.

Hypotheses Section

The previous discussion leads us to propose a series of hypotheses amenable to empirical testing.

First, we speculate that the Democrat’s advantage among women springs, significantly, from the party’s

perceived advantage on matters of race. And the reason perceived racial differences translate into

electoral support is not necessarily because women perceive their political interests as overlapping with

those of minorities. Instead, these differences are cues to women about a candidate characteristic that

they value more than men: compassion.

If so, then women should be especially sensitive to campaign cues involving the candidates’

support for black interests. In other words, when candidates diverge on racial policies, women should be

especially likely to see the Democratic candidate as supportive of black interests. This perception should

be accompanied by the belief among women that the Democrat is also more compassionate than the

Republican. Since this quality resonates with women more than men, the traditional Democratic stand on

race should also lead women to conclude that the Democrat will be more responsive to women’s interests.

The result should be that traditional differences in the party platforms on racial issues should boost

support for the Democratic candidate among women, thereby producing the familiar gender gap. The

traditional Republican stance on race, on the other hand, should produce more favorable perceptions and

greater support among men.

Our framework may also help us understand previous finding suggesting that women’s issues are

not the primary cause of the gender gap. Our speculation is that women’s issues such as abortion rights,

set-asides for female entrepreneurs, or equal pay are not as successful at driving men and women apart

9

because they do not convey the same cues about concern for the disadvantaged that distinguish the parties

on racial issues. In other words, women’s issues do not resonate as uniformly on the dimension of

compassion, and therefore do not divide men and women as cleanly as many racial issues do.

Third, if our notion about the roots of the gender gap is correct, the Republican Party might be

able to reduce its disadvantage among women by adopting a message of racial inclusion. Reaching out to

African Americans should resonate more with women than with men, again because of the implications it

carries about how compassionate the parties are. Indeed, since men are not motivated by this particular

characteristic, they might react quite negatively to a Republican candidate making a moderate appeal on

racial issues. The implication of this prediction is not that a racially tolerant Republican candidate will

come to dominate the Democrat, who already maintains a reputation for being sympathetic to minorities,

among women. We would predict, however, that the gender gap in support for the Democratic candidate

should be significantly minimized when the Republican makes an effort to moderate his traditional

position on racial issues.

Finally, the strategy employed by George W. Bush and the Republican Party in 2000 lends itself

to a natural test of the hypotheses we have proposed. If our hypotheses about the impact of racial cues on

the gender gap are correct, exposure to Bush’s compassionate conservatism message during the

Republican National Convention should also prove more effective among women rather than men. Thus,

women who watched the convention should view Bush more favorably with regard to race relations, and

they should view him as a more appealing candidate in general. At best, men should react with

indifference to such cues. Exposure to the convention of 2000, in other words, should reduce the gender

gap.

Methods and Procedures

Given our interest in the impact of racial appeals, an experimental design is best suited to test our

hypotheses. The chief virtue of experiments is that they allow the researcher to isolate and manipulate the

factors that are expected to produce changes in attitudes or behavior (Kinder and Palfrey 1993). This

10

method has become increasingly popular in the study of political communications (see for example

Gilliam and Iyengar 1999; Iyengar and Kinder 1987; Nelson, Clawson and Oxley 1997; Miller and

Krosnick 2000; Valentino 1999). While surveys are far more effective at estimating population means and

trends, they are weaker at producing valid causal inferences about the impact of particular mass media

content. For these reasons, we develop an experimental design to examine how race versus gender cues

affect the gender gap in support for presidential candidates.

The data used to test our hypotheses are drawn from a convenience sample of 237 adult, non-

student residents from the area surrounding a large Midwestern university. Blacks and other non-whites

constituted just under 40% of our sample. Because of our interest in examining the gender gap among

whites, these subjects are not included in the following analyses. As a result, we are left with 145 white

subjects in this study, 49% are females and 51% are males. The study was conducted in July of 2000 in a

university computer lab.2 Subjects were recruited individually with flyers distributed to local businesses,

university office buildings and in a downtown area near campus. Each was told she would receive fifteen

dollars for answering questions about “current events.” As subjects entered the lab they were randomly

assigned to one of three experimental conditions or a control group, and then escorted to a computer

terminal. In order to minimize interviewer biases, subjects interacted solely with the computer throughout

the session. Once the subjects were finished answering a pre-test questionnaire about the type of radio and

television programs they preferred, the computer instructed them to read a series of short newspaper

articles. Each subject in the treatment groups viewed two non-political articles and one political story.3

2 In spite of our reliance on adult subjects rather than students, our sample is not representative of the national population (although it compares quite well to the local population). For example, a disproportionate number of our respondents have a college degree (54%), are Democrats (60%), and are ideologically liberal (54%). 3 All of the stories were presented so that they would appear authentic to the subjects. Thus, the masthead of the newspaper was superimposed over each article, and the byline was clearly visible at the beginning of each story.

11

Those assigned to the control group read only non-political articles.4 Following exposure to these articles,

subjects responded to a number of questions regarding their policy preferences and views of the major

party political candidates.

The three versions of the campaign story were written by the researchers based on actual

coverage published in major news outlets during the period of the study. The first version of the story,

which we refer to as the “Traditional Race Frame”, begins with the headline “Gore, Bush Differ on Black

Issues” and is accompanied by two contrasting color photographs of the candidates (see Appendix 3).

Gore is shown interacting amicably with NAACP head Julian Bond and Bush is shown amidst a crowd of

white supporters at Bob Jones University.5 The article focused on appearances by each candidate before

the annual convention of the civil rights organization. This version of the story begins with the reporter

observing that the candidates have tried to distinguish themselves from one another throughout the

campaign and that, “…on issues affecting African Americans the candidates have adopted dramatically

different positions.” The article goes on to note that Gore and Bush adopt starkly different positions on

issues such as support for stronger enforcement of civil rights laws, affirmative action, racially diverse

administrations, improvements in public education, and expanded access to health insurance.

An alternative version of this story, which we call the “Compassionate Race Frame”, carries very

similar content except that it presents the candidates as equally sympathetic to issues of concern to

African Americans. For example, the headline for this version reads, “Gore, Bush Similar on Black

Issues” and is accompanied by color photos of both Gore and Bush interacting comfortably with Julian

Bond (See Appendix 3). The text of this article begins much like the previous version, except that the

writer concludes, “…on issues affecting African Americans the candidates are taking surprisingly similar

positions.” For example, in this version, Bush is praised for his speech before the NAACP with many in

4 The non-political stories dealt with wildfires throughout the US, the new subway line in Los Angeles, and recent medical breakthroughs in restoring vision. The latter story was only viewed in the control group. 5 At the time of Bush’s visit, this southern university had a much-publicized policy of banning inter-racial dating.

12

attendance indicating admiration for his courage in addressing a skeptical crowd. The article goes on to

note that Gore and Bush both support stronger enforcement of civil rights laws, racially diverse

administrations, improvements in public education, and expanded access to health insurance.

The story on gender issues, which we will call the “Traditional Gender Frame”, follows the

pattern described above with the traditional race frame. The headline reads, “Gore, Bush Differ on

Women’s Issues” and notes that the candidates oppose one another on issues such as abortion rights and

women’s health care. Additionally, Bush is portrayed as less than enthusiastic about encouraging greater

business opportunities for women, or adding women to his staff.

Although the articles present vastly different images of Bush, we believe that each version is

credible. The candidates’ websites were carefully reviewed so as to summarize accurately each

candidate’s policy positions. Additionally, our stories drew upon actual news accounts of each

candidate’s speech before the NACCP, or on the campaign trail. In the end, it was possible to realistically

convey different perspectives of Bush’s political positions because the candidate’s record on compassion

issues or women’s issues can plausibly be characterized as either moderate or conservative. For example,

Bush actually did declare at the NAACP convention: “Strong civil rights enforcement will be a

cornerstone of my administration.” Additionally, his interest in improving education and his intention to

nominate Colin Powell as secretary of state were well known. However, it is also true that Bush visited

Bob Jones University, adopted a strong pro-life position on abortion, and opposed most affirmative action

policies during the campaign. If the candidates’ positions vis-à-vis blacks contribute to the gender gap,

then the traditional partisan stands on this issue should polarize men and women. Alternatively, Bush’s

efforts to appear racially moderate should appeal to women but fall flat among men.

Four dependent variables were selected for the experimental portion of our analyses. The first

two represent something of a manipulation check, and assess which candidate is most likely to Represent

the Interests of African Americans and Which Candidate is Most Compassionate (see Appendix 1 for all

question wording). We also examine whether or not our political stories affect subjects’ perceptions about

13

which candidate will best Represent the Interests of Women. Finally, we examine the impact of these

three frames on overall Candidate Preferences.

In order to examine the effects of viewing the Republican Convention, we analyze survey data

from a national Gallop poll that went into the field directly following the convention on the 4th and 5th of

August, 2000. The survey contained 1,051 respondents, of which 890 were white, and included a variety

of questions that allow us to examine the effects of race-based appeals on the gender gap. In addition to

asking How Often Respondents Watched the Republican Convention, the survey also asks How Likely

Respondents are to Vote for George W. Bush, whether or not they Approve of Bush, and Which

Candidate Would Best Handle Race Relations and a variety of other issues. Our expectation is that

increased exposure to the convention will lead to more favorable views of Bush among women,

especially in the area of race relations, and greater overall support. Men who watched the convention

will, at best, be unaffected by the show of racial diversity. At worst, they will hold it against the

Republican candidate. The functional form of this model is as follows:

Bush Evaluations=B1(Watched Convention) + B2(Female) + B3(Convention*Female) + B4(Party

Identification) + B5(Ideology) + B6(Education) + B7(Household Income) + B8(Age) + B9(Union

Membership) + B10(Likelihood of Voting) + Constant.

According to our hypotheses, B3 will be positive and significant, indicating more favorable views of Bush

among women who devoted a significant amount of time to the convention. On the other hand the B1

coefficient, representing men who watched the convention, should be negative, indicating disapproval of

Bush’s appeal to racial inclusion. In both experimental and survey analyses, the dependent variables are

coded on a 0-1 scale to ease interpretation of the OLS regression models. All of the independent

variables of interest (e.g. how much of the convention respondents watched) are also coded onto this

scale.

14

Results

Racial Cues and the Gender Gap

Our first set of hypotheses involves the power of the traditional race frame and its relationship to

perceptions of candidate responsiveness to blacks and to evaluations of candidate compassion. Our

expectation is that the Democrats’ traditional stand on race will be viewed as an indication of

responsiveness to black interests, particularly among women. Similarly, women should view outreach

efforts to blacks as a signal of a more generally compassionate orientation to politics. We would expect

that men, on the other hand, should regard the traditional Republican stand on race as more desirable, and

move toward Bush on this and any other dimension. We test these hypotheses in figures 1A and 1B.

[FIGURE 1 ABOUT HERE]

In Figure 1A, we see that in the absence of a story about the political campaign, men and women

both feel that Gore will do the best job at representing the interests of Blacks. The difference between

men and women in this condition is statistically indistinguishable from zero.6 Among those exposed to the

traditional race frame, however, a large and significant gender gap appears on this dimension. Here we

find that women are significantly more likely to view Gore as more responsive to African Americans,

whereas men become somewhat less confident in Gore’s ability to represent black interests. This first

finding suggests that the respondents in our sample were in fact sensitive to the racial cues in the

traditional race frame, and reacted in predictable ways.

We expect that the gender differences in group representation produced by the traditional race

frame would be accompanied by a gender difference in evaluations of candidate compassion. This

hypothesis is examined in Figure 1B. As in the previous figure, the gender gap is statistically

insignificant, relatively mild, and in the “wrong” direction in the control group. This suggests that the

Democratic advantage on this dimension among women is not chronically salient, even within our

6 These figures are based on the OLS regression analyses shown in tables 1A and 2A of Appendix 2 . The predicted values are estimated by manipulating a hypothetical subject’s gender and the article that he or she reads, while holding all other variables in the model at the population mean or median.

15

relatively Democratic convenience sample. As expected, however, when the candidates diverge on racial

issues the gender gap assumes its more familiar character. Women move some 15-points towards Gore,

and men move almost as sharply towards Bush. In other words, the traditional race cue sends much

different signals to men and women about how compassionate the candidates are.

[FIGURE 2 ABOUT HERE]

If our hypotheses regarding the special linkage between race and compassion cues are correct,

then traditionally women’s issues---which do not implicate these concerns as strongly---should be a less

effective trigger of the gender gap. This hypothesis is tested in Figures 2A and 2B. We find that our

gender story has no impact on the degree to which men or women feel these candidates will serve the

interests of blacks. Gore is viewed, almost equally among men and women, as the more effective

candidate on this dimension, and these perceptions are unaffected by the stimulus. We uncover similar

results for the compassion measure. In both the control condition and the gender condition, men are

slightly, but not significantly, more likely to view Gore as the more compassionate candidate.7 Thus,

whatever information the traditional gender cues convey, they apparently do not affect perceptions of the

candidates’ general orientation on compassion issues.

Our next tests involve the impact of race versus gender cues on perceptions of candidate

responsiveness to women’s interests. If our hypothesis about the impact of racial cues is correct, then

men and women should also diverge in evaluations of how well the candidates will represent women’s

interests in the presence of the cue. We examine this hypothesis in figure 3A.

[FIGURE 3 ABOUT HERE]

Our expectations regarding the relationship between partisan race cues and perceptions of

responsiveness to women’s issues are strongly confirmed. In the control group, both men and women

7 This counterintuitive difference in the control cells is due to statistical controls for party identification, ideology, education, and other variables that contribute to the baseline political differences between men and women. With these controlled, the men in our control group are actually slightly more supportive of Gore than are the women.

16

lean heavily towards Gore overall, but women view Gore as slightly less effective than do men. When

exposed to the traditional race frame however, women regard Gore as far more responsive to their

interests whereas men move sharply in the opposite direction. It is striking how powerfully men and

women diverge without any overt reference to gender or women’s issues.

In Figure 3B, we examine whether the traditional gender frame also affects perceptions about

which candidate will best represent women’s interests. Our expectation is that this cue will be less

effective than the traditional race frame. This hypothesis is partially confirmed. As with the race frame,

exposure to the gender frame also causes women to view Gore as the best representative of their interests.

However, this result does not achieve conventional levels of statistical significance (see Table 1A in

Appendix 2). Additionally, men are entirely unaffected by this frame. Thus, while there is some

indication that these cues provide information to women their inability to influence men suggests that the

traditional gender frame does not significantly contribute to the gender gap.

The bottom line for all our analyses is the impact of racial cues on the gender gap in support for

Democratic versus Republican candidates. We expect, given what we have already found, that traditional

racial cues should be causally implicated in that gap. We examine this hypothesis in Figure 4A. As in

our other analyses, men and women differ only mildly, and not significantly, with regard to their

candidate preference in the control group. However, as anticipated, exposure to the traditional race

frame produces a substantial gender gap. Women become about 15-points more supportive of Gore

whereas the Democratic candidate loses about 28-points among men. This 34-point gender gap is

considerably larger than the others we have reported.

[FIGURE 4 ABOUT HERE]

If, as we suspect, the gender gap owes far less to partisan differences on women’s issues, then we

should find that candidate preferences are only weakly affected by our gender frame. This hypothesis is

examined in Figure 4B. As anticipated, the traditional gender frame does not produce a gender gap in

candidate preference. Differences across gender are not statistically significant in the control condition

17

nor are they in the gender condition. Thus, it would seem that the most effective way to exacerbate the

gender gap is for candidates to emphasize the traditional partisan positions on race.

Our results thus far provide strong support for the notion that partisan race cues contribute

significantly to the gender gap. When Bush appears hostile to black interests, women move strongly

toward the Democrat even as men view the Republican more favorably. One explanation for these

contrasting reactions may be that these cues raise different considerations for men and women. We have

argued throughout this paper that women are likely to make more global inferences about the candidate’s

commitment to the disadvantaged when exposed to the traditional race frame. In short, the Democratic

position resonates with women’s preference for a more equitable society. Men, on the other hand, may

conclude that Democratic appeals to blacks violate, rather than uphold, the principle of equality. There is

some precedent for this speculation. Kinder and Sanders (1996) report that egalitarian values can either

contribute to or undermine support for affirmative action, depending upon the manner in which the policy

is framed. When opposition is framed in terms of granting blacks advantages they have not earned,

supporters of equality typically endorse affirmative action. When opposition is framed in terms of

reverse discrimination, supporters of equality generally oppose the policy. Thus, it is possible that men

support Bush in the traditional race frame because of their belief that Democratic efforts to advance black

interests are inconsistent with the principle of equality.

[FIGURE 5 ABOUT HERE]

To test these ideas, we interact gender with both the traditional race frame and a measure of

subjects’ support for egalitarianism.8 If our suspicions are correct then, among women in the traditional

race frame, egalitarianism will be associated with greater support for the Democratic candidate but,

among men, it will lead to greater support for the Republican. Each expectation is confirmed, as shown

in table 2A (see appendix 2).9 Because of the difficulty involved in interpreting the magnitude of the

8 The equality scale is made up from two items as described in the appendix. The alpha on this scale is 63. 9 We also examined whether the traditional gender frame primed attitudes about equality. As expected the results of these analyses fell well short of statistical significance.

18

triple-interaction, the results from these analyses are converted into predicted values as shown in figures

5A and 5B. The top panel of figure 5 shows the size of the gender gap in candidate preferences among

subjects scoring high on the egalitarian scale.10 As anticipated, men in the experimental condition prefer

Bush to Gore by a more than 5-to-1 margin.11 However women who score high on the egalitarian scale,

already predisposed to support Gore, are even more supportive of his candidacy in the traditional race

frame. This 61-point gender gap is considerably smaller among subjects scoring relatively low on the

egalitarian scale.

Compassionate Conservatism and the Reduction of the Gender Gap

Up until this point, our experimental results have indicated that the gender gap is strongly

associated with candidate differences on questions of race. If so, then it might also be possible for

candidates to alter how voters perceive them on race matters, and thereby influence the magnitude of the

gender gap. Given the Republican disadvantage among women, this strategy would seem especially

attractive. In effect, Republican presidential hopefuls are presented with two alternatives. They can

adopt their party’s traditionally conservative position on issues of race, and risk alienating women voters,

or they can seek to transform their party’s image so that they are viewed as more racially inclusive, and

attract white women. This is essentially the strategy that George Bush and the Republicans adopted

during the 2000 Presidential campaign, so it makes for a particularly interesting test here.

There are, of course, some potential drawbacks to this strategy. One is that, in gaining support

among women, Republicans might also lose support among men. Another problem is that, after almost

10 Given the liberal bias in our sample, it is not surprising that scores on the egalitarian scale are highly skewed. The mean is about .78 on a 0-1 scale, with a standard deviation of about .22. As a result, we identify “high” scores as one standard deviation above the mean and “low” scores as one standard deviation below the mean. 11 We also find that men who regard blacks as competitive threats (see Bobo and Hutchings 1996) and who believe that blacks have too much influence in society are also more likely to support Bush in the experimental condition. Women with such views are unaffected by the stimulus. Results among men are consistent with Kinder and Sanders (1996) finding that, when juxtaposed with reverse discrimination cues, egalitarian principles can lead to racially conservative outcomes.

19

forty-years of preaching racial conservatism, Republican candidates may simply lack credibility on this

issue and thus fail to alter voter perceptions. In short, a compassionate race frame may produce results not

unlike those of the traditional race frame. Our analyses in this section examine whether a strategy of

“compassionate conservatism” on race can effectively diminish the Democrat’s relative advantage among

women.

[FIGURE 6 ABOUT HERE]

Figures 6, which again examines candidate preferences, takes as our baseline the traditional race

frame. As indicated above, in most previous elections both Democrats and Republicans could be counted

on to adopt positions consistent with this political story. In Figure 6, however, we examine whether

Republicans can reduce the gender gap associated with the traditional frame by assuming a more

compassionate position on race. Ideally, from the Republican vantage point, Bush would succeed in

reducing the Democrats advantage among women without undermining his greater support among men.

We see, however, that this strategy is only partially successful. Bush’s compassion offensive does

diminish Gore’s advantage among women but it also cuts into his advantage among men. The

interaction, once again, is highly significant. Although these results underscore the perils inherent in the

Republican’s efforts to reach out to minority voters, it also highlights the extent to which the gender gap

is associated with the issue of race. When the candidates sharply diverge on racial issues the gender gap

becomes a veritable chasm. When the candidates are largely in agreement on race, the gender gap

virtually disappears.12

The Gender Gap and the 2000 Republican Convention

Our experimental findings provide strong support for our contention that racial cues can influence

the magnitude and character of the gender gap. Nevertheless, as with all laboratory experiments, our

12 We also examined the effects of the compassionate race frame on our other dependent variables. As expected, we found that when Bush moderates his positions on race, women’s views of the Republican become significantly more positive whereas men become significantly more negative.

20

analyses are vulnerable to the criticism that they lack external validity. For this reason, we also examine

the impact of racial cues among a more representative population. A CNN/USA Today Gallop telephone

survey, fielded in August of 2000, provides an excellent opportunity to examine our hypotheses outside

the rarified atmosphere and convenient sample present in the lab.

The subject of this survey was the recently concluded Republican National Convention, as well as

the presidential campaign more generally. In this convention, Republicans made a concerted effort to

showcase their commitment to racial inclusiveness and racial tolerance. In the words of Lori Rodriguez

of The Houston Chronicle:

From the opening night speech Monday by Colin Powell to a key role Thursday for Houston’s Kirbyjon Caldwell, the Republican Party strained mightily to put its best and most diverse face forward this week. There was the retired Army general chiding his party for its anti-affirmative action position; the chief national security advisor, Condolezza Rice, speaking glowingly about the Republican presidential nominee, George W. Bush….There were black singers, black dancers, black children, black preachers, a veritable army of black speakers.13

In short, the 2000 Republican National Convention sought to convey an impression of candidate Bush

that was very much like our compassionate race frame. If our experimental results are robust, then we

should find that white women responded much more favorably to this display than did white men.

Table 1 presents the ordered logit results for perceptions of which candidate would best handle a

series of important public policy issues. The first column presents the results for the candidates’ ability to

handle race relations. As anticipated, women respond quite differently than men as a result of exposure to

the convention. The interaction of gender and exposure to the convention is positive and statistically

significant, suggesting that women who watched the convention were more likely than those who did not

to view Bush as effective on this issue. Alternatively, in spite of the convention’s emphasis on diversity,

men who were exposed to this message perceived Bush as slightly less effective on race although this

13 Rodriguez, Lori. “Republican Convention Philadelphia 2000; Parade of diversity no sign GOP has changed foes say; Supporter say it’s a step in the right direction.” Houston Chronicle Friday, August 4, 2000.

21

effect is not statistically significant. This interaction stands despite controls for many variables that could

have produced biases in the viewing audience to begin with, including party, ideology, income, education,

age, union membership and previous turnout.

[TABLE 1 ABOUT HERE]

Converting the ordered logit coefficients into predicted probabilities reveals just how large these

effects are among women. For example, a political independent, who is at the mean or median on all

other variables in the model, has a .31 probability of indicating that Bush would best handle race relations

if she did not watch any coverage of the Republican convention. A respondent with the same

characteristics, who watched “a great deal” of the convention, has a .46 probability of selecting Bush on

this issue dimension. Among men, this probability declines by about 6-points.

Columns 2 and 3 present results for two other issues frequently touted as responsible for the

gender gap: education and abortion. As one of Bush’s signature issues was his relatively progressive

stance on education, and his pledge to “leave no child behind,” it is conceivable that exposure to the

convention would also impress upon women his strength on this issue. Similarly, although clearly pro-

choice, Bush’s decision to de-emphasize this contentious issue throughout his campaign may have also

provided some reassurance to women viewing the convention. We see in Table 1, however that on

neither issue does Bush gain a statistically significant advantage relative to Gore. Moreover, men again

seem to, if anything, find Bush less effective on these issues if they tuned into the convention. In general,

these results suggest that for women, the most prominent message conveyed by the Republican National

Convention, perhaps by design, involved race.

In our final set of tests, we test whether exposure to the GOP convention reduced the gender gap

in support for Bush versus Gore. In column one of Table 2, we examine the impact of the convention on

whether respondents held a favorable or unfavorable opinion of the candidate. As hypothesized, women

who watched the convention were significantly more likely to approve of Bush relative to women who

did not watch the convention, as indicated by the positive and statistically significant interaction

22

coefficient. Among men, the effects of viewing the convention were mildly negative and statistically

insignificant.

[TABLE 2 ABOUT HERE]

Converting these results into predicted probabilities, we find that a political independent with

characteristics at the mean on all other variables has a .73 probability of expressing a favorable opinion of

George W. Bush. This probability increases by 13-points among women who watched a great deal of the

convention. Among women who identify strongly with the Democratic Party, the increase is 21-points.

The last column of Table 6 presents our results for the likelihood that respondents will vote for

Bush in the November election. Once again, we find that women who viewed the convention respond

positively to Bush whereas men were largely unaffected. Since the dependent variable is coded on a 0-1

scale, the coefficient of .10 roughly translates into a 10-point increase in support for Bush among women

who paid a great deal of attention to the Republican convention. This, of course, represents the best

possible scenario for the Bush team for, unlike our lab results, men appear more disinterested than turned-

off by the compassion strategy. One explanation for this result may be that the convention

simultaneously highlighted the Republicans commitment to diversity as well as their commitment to

traditionally conservative issues such as tax cuts, increased defense spending, and family values. Thus, at

least in the short run, the potential contradictions of appearing both compassionate and conservative were

not addressed.

Conclusion

Before summarizing the results of our analyses, it is important to clarify what we are not arguing.

We do not maintain that the gender gap arises solely because of gender differences on racial policy

opinions, although these differences clearly do exist. Other scholars have found, and we find no reason to

dispute, that other issues have also contributed to this gap. Instead, we argue that the unmistakable, and

sharply contrasting stands that the two major parties have taken over the past forty years on racial policies

have contributed significantly to the gender gap. Women are less sympathetic than men to the traditional

23

Republican position on race in part because of their somewhat more racially liberal policy positions and

in part because of their greater predisposition to support compassion issues more broadly. Men, on the

other hand, respond negatively to the Democratic stand on race because of their view that it violates the

principle of equality and because the Republican message successfully primes their negative racial

attitudes (see note 11). In our view, no previous work has recognized nor appreciated the extent to which

compassion issues, and their racial implications, contribute to the gender gap.

The multi-methodological approach we adopt uncovers an important source of the gender gap. In

the lab, manipulating Bush’s stance on racial policy has a dramatic effect on the magnitude of the gender

gap. When Bush and Gore adopt traditional race stands, white women are more likely to regard Gore as

compassionate, and a far more effective representative of both blacks’ and women’s interests. Not

surprisingly, this frame also maximizes the gender gap in support for Bush. On the other hand, when both

candidates appeared sympathetic to black interests, Gore’s advantage among women evaporates and

women are much more likely to view Bush in a favorable light. As hypothesized, white men respond in a

completely opposite fashion. When exposed to the traditional frame, men are less likely to view Gore as

compassionate. Moreover, they perceive Bush as the more appropriate representative of women’s

interests when he is merely conservative, and not compassionate on race. In the end, white men are also

more likely to vote for Bush when he does not reach out to racial minorities.

These results also hold outside the lab. When exposed to the Republican show of racial diversity

at the 2000 GOP convention, white women were more reassured in Bush’s ability to handle race relations.

Men, however, were generally unaffected. Similarly, women who watched significant portions of the

convention were more likely to view Bush favorably and ultimately to indicate that they would vote for

him. Again, men who watched the convention were unmoved. In light of these results, we conclude that

the gender gap is, at least in part, a consequence of the party’s realignment on matters of race.

In addition to its implications for understanding the roots of the gender gap, our results also

underscore the extent to which the current party system is significantly influenced by lingering problem

of race. Scholars and political observers have long been aware that, at the national level and increasingly

24

at the local level, Democrats are disadvantaged among whites because of their more racially inclusive

reputation. Our results, and indeed candidate Bush’s behavior during the 2000 campaign, suggest that

Republicans are now beginning to recognize that they too pay a price for their position on the race issue.

If scholars such as Carmines and Stimson (1989) are correct, and the parties are not likely to forsake their

positions on these issues anytime soon, we can expect a long and healthy lifespan for the gender gap.

One final question worth asking is how effective was the Republican compassion strategy in the

2000 elections. We have demonstrated that they had considerable short-term success, at least among

women, but how durable were these results? Of course, among blacks they were famously unsuccessful,

as Bush received the lowest percentage of African-American support of any Republican candidate since

Ronald Reagan and Barry Goldwater (Pomper 2001). Among women, the results are more mixed.

According to the NES, the gender gap on election day 2000 was 10%, which represents a small reduction

from the 15% of 1996. However, this gap is about the same size that it was in 1992, when the candidates

adopted more or less traditional positions on matters of race.

Why didn’t the gender gap shrink further? A definitive answer to this question is beyond the

scope of this paper but we can offer a likely explanation. In spite of Bush’s sincere desire to change the

image of the GOP, he faced two powerful obstacles. One was the historical reputation of the Republican

Party on race, which Gore and the Democrats played up at every opportunity. The second was Bush’s

own mixed messages on compassion issues heavily associated with race. As indicated above, although

one of his signature issues was education, he also registered only tepid support for hate crimes legislation,

and supported a massive tax cut that, at least according to his critics, disproportionately benefited the

wealthy and indirectly damaged the social safety-net. In the end, these conflicting messages probably

undermined his efforts to recast the Republican Party’s image on race and compassion.

25

APPPENDIX 1: Scale/index construction.

Candidate Preference was based on a 5-item sequence with skip patterns.

1. “So far as you know now, do you expect to vote in the national election this coming November

or not?” (1=Yes, 5=No, 8=Don’t Know).

2. (If R plans to vote): “We all know the election is some time away and people are not certain at

this point who they will vote for. Still, who do you think you will vote for in the election for

President?” (Bush, Gore, Nader, other, don’t know, undecided).

3. “Would you say that your preference for (candidate specified in item #2) is strong or not

strong?”

4. (If R plans not to vote) “If you were going to vote, who do you think you would vote for in the

election for president?” (same choices as in item #2).

5. (For those who answered item #4) “Would you say that your preference for (candidate

specified in item #4) is strong or not strong?”

An index was constructed, running from 0 (support for Gore) to 1 (support for Bush). Those

preferring a third party candidate (Nader, Buchanan, or Other) were placed in the middle of the

index at .5.

Candidate Traits were measured with the following items:

1. “In your opinion, which candidate is more compassionate?”

Response options ranged from Gore, “Much more compassionate” (coded “0”) to Bush, “Much

more compassionate” (coded “1”).

The Group Representation questions were based on a 3-item sequence with skip patterns.

26

1. “Now we would like to know which presidential candidate you think would do a better

job representing the interests of particular groups in society. Who do you think would do

a better job representing the interests of [blacks or women]?”

Response options ranged from Gore, “Much better job” (coded “0”) to Bush, “Much better jop”

(coded “1”).

The Equality Scale is composed of the following two items:

1. Our society should do whatever is necessary to make sure that everyone has an equal

opportunity to succeed.

2. If people were treated more equally in this country we would have many fewer problems.

Response options ranged from “strongly disagree”(coded “0”) to “strongly agree” (coded “1”).

Appendix 2: Regression Analyses for Figures 1 through 6

The results from figures 1 through 4 are derived from the analyses presented below. Formally,

our model can be represented as follows:

Candidate Evaluations=B(Traditional Gender Frame) + B2(Traditional Race Frame) + B3(Female) +

B4(Gender Frame*Female) + B5(Traditional Frame*Female) + B6(Controls) + Constant.

Across each dependent variable, we expect that the b2 coefficient will be substantively large and positive,

indicating that men’s view of Bush grows more favorable when he adopts a traditional Republican stance

towards blacks. Alternatively, the net effect of the b2 and b5 interaction, indicating women’s response to

the race frame, should be of roughly equal magnitude yet negative, indicating increased support for Gore.

The coefficients representing the gender frame should typically be indistinguishable from zero. To guard

against the possibility that differences in the distribution of socio-demographic or political variables

across cells of the design might account for differences we observe, controls for ideology, partisanship,

age, and education are included.

27

Table 1A. Regression Models Predicting Effects of Traditional Gender Frame and Traditional Race Frame on Perceptions of Candidate Traits

Rep. Blacks Compassion Rep. Women Vote Choice Intercept .43***

(.09) .76***

(.08) .59***

(.08) 1.01***

(.10) Gender Frame -.04

(.08) -.03

(.07) -.00

(.07) -.01

(.09) Traditional Race Frame

.09 (.08)

.14* (.07)

.20** (.07)

.28*** (.09)

Female .06

(.07) .09

(.07) .11*

(.07) .09

(.08) Gender Frame * Female

.03 (.11)

.04 (.10)

-.12 (.10)

.05 (.13)

Race Frame * Female

-.22* (.11)

-.29** (.10)

-.32** (.10)

-.43*** (.13)

Adj. R-Squared .12 .33 .30 .45 N 116 116 116 116

Notes: * p < .05; ** p < .01; *** p < .001 for One-tailed test. Controls in each analysis, not shown here, include party identification, political ideology, age and educational attainment. Sample sizes for each cell were: Control=50; Traditional Gender cell=34; Traditional Race cell=32.

The results for figure 5 are derived from the analyses presented below in table 2A. Formally, our

model can be represented as follows:

Candidate Evaluations=B(Traditional Race Frame) + B2(Female) + B3(Egalitarian Scale) + B4(Race

Frame*Female) + B5(Race Frame*Equality) + B6 (Female* Equality) + B7 (Race Frame*Female*

Equality) + B8(Controls) + Constant.

Our expectation is that the net effect of the B and B5 coefficients will be positive, indicating that

men scoring high on the egalitarianism scale are strongly predisposed to support Bush in the experimental

condition. Women should respond in the opposite fashion, as indicated by the net effect of the triple

interaction and all the attending 2-way interactions.

(Continued on Next Page)

28

Appendix 2 (Continued From Previous Page)

Table 2A. Regression Model Predicting Effects of Attitudes about Equality and

Experimental Frames on Candidate Preferences Vote Choice Intercept 1.14***

(.21)

Traditional Race Frame -1.17**

(.53)

Female -.10

(.33)

Attitudes about Equality -.27

(.24)

Interactions Race Frame * Female 1.09*

(.64)

Race Frame * Equality 1.76**

(.63)

Female * Equality .25

(.40)

Race Frame * Equality * Female

-1.85** (.76)

Adj. R-Squared .44

N 82 Notes: * p < .05; ** p < .01; *** p < .001 for One-tailed test. Controls in the analysis, not shown here, include party identification, political ideology, age and educational attainment. Interactions were also included, although not shown, for “Black Influence” by gender, and each experimental cell by gender. Sample sizes for each cell were: Control=50; Traditional Race cell=32.

(Continued on Next Page)

Appendix 2

29

(Continued From Previous Page)

Results for figure 6 are derived from the analyses in table 3A. Formally, this model can be

represented as follows:

Candidate Evaluations=B(Compassion Frame) + B2 (Female) + B3(Compassion Frame*Female) + B4(Controls) + Constant. The traditional frame represents the excluded category in these analyses. When both candidates

appeal to blacks, as in the “compassionate frame”, we expect the b1 coefficient to be negative,

indicating men’s movement away from Bush, and the b3 coefficient to be positive, indicating

female movement towards Bush.

Table 3A. Regression Models Predicting Effects of Compassion Frame on Candidate Preferences

Vote Choice Intercept 1.26***

(.13)

Compassionate Race Frame

-.23* (.11)

Female -.34***

(.11)

Compassion Frame * Female

.35** (.16)

Adj. R-Squared .43

N 61 Notes: * p < .05; ** p < .01; *** p < .001 for One-tailed test. Controls in each analysis, not shown here, include party identification, political ideology, age and educational attainment. Sample sizes for each cell were: Traditional Race cell=32; Compassionate Race cell=29.

App

endi

x 3:

Tra

ditio

nal R

ace

Fram

e

31

App

endi

x 3:

Tra

ditio

nal G

ende

r Fr

ame

32

App

endi

x 3:

Com

pass

iona

te R

ace

Fram

e

33

References

Abramowitz, Alan I. 1994. “Issue Evolution Reconsidered: Racial Attitudes and Partisanship in the U.S.

Electorate.” American Journal of Political Science 39:1-24.

Abramowitz, Alan I. and Saunders, Kyle L. 1998. “Ideological Alignment in the U. S. Electorate.”

Journal of Politics 60:634-652.

Adams, Greg D. 1997. “Abortion: Evidence of an Issue Evolution.” American Journal of Political Science

41(3): 718-737.

Berelson, Bernard, Paul Lazarsfeld and William McPhee. 1954. Voting. Chicago: Chicago

University Press.

Bobo, Lawrence. 2000. “Race and beliefs about affirmative action: Assessing the effects of interests,

group threat, ideology, and racism.” In Racialized Politics, edited by David O. Sears, Jim

Sidanius, and Lawrence Bobo, 137-164. Chicago: University of Chicago Press.

Bobo, Lawrence and Vincent L. Hutchings 1996. “Perceptions of Racial Group Competition: Extending

Blumer’s Theory of Group Position to a Multiracial Social Context.” American Sociological

Review 61(6):951-972.

Campbell, Angus, Phillip E. Converse, Warren E. Miller, and Donald E. Stokes. 1960. The American

Voter. New York: Wiley.

Carmines, Edward G., and James A. Stimson. 1989. Issue Evolution: Race and the Transformation of

American Politics. Princeton: Princeton University Press.

Chaney, Carole K., Michael R. Alvarez, and Jonathan Nagler. 1998. “Explaining the Gender Gap in US

Presidential Elections, 1980-1992.” Political Research Quarterly 51(2): 311-339.

Conover, Pamela J. 1988. “Feminists and the Gender Gap.” Journal of Politics 50:985-1010.

Conover, Pamela J., and Virginia Sapiro. 1993. “Gender, Gender Consciousness, and War.” American

Journal of Political Science 37: 1079-99.

34

Cook, Elizabeth Adell, and Clyde Wilcox. 1991. “Feminism and the Gender Gap---a Second Look.”

Journal of Politics 53:1111-22.

Edsall, Thomas B., and Mary D. Edsall. 1991. Chain Reaction: The Impact of Race, Rights, and Taxes

on American Politcs. New York: Norton.

Erikson, Robert S. & Kent. L. Tedin. 1995. American Public Opinion. (5th Edition). New York:

Addison Wesley Longman.

Iyengar, Shanto and Donald R. Kinder. 1987. News that Matters: Television and American Opinion.

University of Chicago: Chicago.

Johnson, Monica K., and Margaret M. Mooney. 1998. “Bridging the Racial Divide in the United States:

The Effects of Gender.” Social Psychology Quarterly 61(3): 247-258.

Kaufmann, Karen M. and John R. Petrocik. 1999. “The Changing Politics of American Men:

Understanding the Sources of the Gender Gap.” American Journal of Political Science 43(3):

864-887.

Kinder, Donald R. and Thomas R. Palfrey. 1993. “On Behalf of an Experimental Political Science.” In

Experimental Foundations of Political Science edited by Donald R. Kinder and Thomas R.

Palfrey.

Kinder, Donald R. and Nicholas Winter. 2001. “Exploring the Racial Divide: Blacks, Whites, and

Opinion on National Policy.” American Journal of Political Science 45(2): 439-456.

Lipset, Seymour M., and Stein Rokkan. 1967. “Cleavage Structures, Party Systems, and Voter

Alignments: An Introduction.”In Party Systems and Voter Alignments, ed. Seymour M. Lipset

and Stein Rokkan. New York: Free Press.

Mansbridge, Jane E. 1985. “Myth and Reality: the ERA and the Gender Gap in the 1980 Election.”

Public Opinion Quarterly 49:164-78.

Miller, Joanne and Krosnick, Jon. 2000. “News Media Impact on The Ingredients of Presidential

Evaluations: Politically Knowledgeable Citizens and Guided by Trusted Sources” American

Journal of Political Science. 44(2): 301-15.

35

Miller, Warren E. 1991. “Party Identification, Realignment, and Party Voting: Back to the Basics.”

American Political Science Review 85(2): 557-68.

Miller, Warren E., and J. Merrill Shanks. 1996. The New American Voter. Cambridge, MA: Harvard

University Press.

Nelson, Thomas E. and Donald R. Kinder. 1996. “Issue Frames and Group-Centrism in American Public

Opinion.” The Journal of Politics 58(4):1055-78.

Nelson, Thomas., Rosalee, Clawson and Zoe, Oxley. 1997. “Media Framing of a Civil Liberties Conflict

and Its Effect on Tolerance.” American Political Science Review. 91:567-83.

Norrander, Barbara. 1999a. “Is the Gender Gap Growing?” In Reelection 1996: How Americans Voted,

ed. Herbert F. Weisberg and Janet M. Box-Seffenmeier, pp. 145-161. New York: Chatham

House.

Norrander, Barbara. 1999b. “The Evolution of the Gender Gap.” Public Opinion Quarterly 63(4): 566-

576.

Paolino, Phillip. 1995. “Group Salient Issues and Group Representation: Support for Women Candidates

in the 1992 Senate Elections.” American Journal of Political Science 39 (2):294-313.

Pomper, Gerald (ed.) 2001. The Election of 2000: Reports and Interpretations. Chatham New Jersey:

Chatham House Publishers.

Sapiro, Virginia and Pamela Johnston Conover. 1997. “The Variable Gender Basis of Electoral Politics:

Gender and Context in the 1992 US Election.” British Journal of Political Science 27 (4): 497-

523.

Schuman, Howard, Charlotte Steeh, Lawrence Bobo, and Maria Krysan. 1997. Racial Attitudes in

America: Trends and Interpretations. Revised Edition. Cambridge, MA: Harvard University

Press.

Scott, Jacqueline and Howard Schuman. 1988. “Attitude Strength and Social Action in the Abortion

Dispute.” American Sociological Review 53(5):785-793.

36

Sears, David O., C. van Laar, M. Carrillo, and R. Kosterman. 1997. “Is It Really Racism? The Origins of

White Americans Opposition to Race-Targeted Policies.” Public Opinion Quarterly 61: 16-53.

Smith, Tom W. 1984. “The Polls : Gender and Attitudes Toward Violence.” Public Opinion Quarterly 48:

384-96.

Stanley, Harold W., and Richard Niemi. 1991. “Partisanship and Group Support, 1952-1988.” American

Politics Quarterly 19:189-210.

Tate, Katherine. 1993. From Protest to Politics: The New Black Voters in American Elections.

Cambridge: Harvard University Press.

Valentino, Nicholas A. 1999. “Crime News and the Priming of Racial Attitudes during Evaluations of the

President.” Public Opinion Quarterly 63 (3):293-320.

Valentino, Nicholas A. and David O. Sears. 2001. “Old Times There Are Not Forgotten: Racism,

Sectional Conflict, and Partisan Realignment in the Contemporary Era.” Presented at the

American Political Science Association annual meeting, San Francisco, September, 2001.

Valentino, Nicholas A., Vincent L. Hutchings, Ismail K. White. 2002. “Cues That Matter: How Political

Ads Prime Racial Attitudes During Campaigns.” American Political Science Review Vol. 96 No.

1 (March).

Welch, Susan, and John Hibbing. 1992. “Financial Conditions, Gender, and Voting in American National

Elections.” Journal of Politics 54(1): 194-213.

Wirls, Daniel. 1986. “Reinterpreting the Gender Gap.” Public Opinion Quarterly 50(3): 316-30.

37

Figure 1: Effects of Traditional Race Frame on Which Candidate Best Represents Black Interests and Which Candidate is Most Compassionate

Figure 1A: Which Candidate Best Represents Black Interests?

0.21 0.15

0.310.27

0

0.1

0.2

0.3

0.4

0.5

0.6

Control Traditional RaceFrame

Experimental Conditions

WomenMen

Figure 1B: Which Candidate Is Most Compassionate?

0.260.21

0.400.36

0

0.1

0.2

0.3

0.4

0.5

0.6

Control Traditional RaceFrame

Experimental Conditions

WomenMen

38

Figure 2: Effects of Traditional Gender Frame on Perceptions of Which Candidate Best Represents Black Interests and Which Candidate Is Compassionate

Figure 2A: Which Candidate Best Represents Black Interests?

0.21

0.26

0.17

0.27

0

0.1

0.2

0.3

0.4

0.5

0.6

Control Traditional GenderFrame

Experimental Conditions

WomenMen

Figure 2B: Which Candidate Is Most Compassionate?

0.26

0.37

0.24

0.36

0

0.1

0.2

0.3

0.4

0.5

0.6

Control Traditional GenderFrame

Experimental Conditions

WomenMen

39

Figure 3: Effects of Traditional Race Frame and Traditional Gender Frame on Perceptions of Which Candidate Best Represents Women’s Issues

Figure 3A: Which Candidate Best Represents Women's Interests?

0.18 0.17

0.380.29

0

0.1

0.2

0.3

0.4

0.5

0.6

Control Traditional RaceFrame

Experimental Conditions

WomenMen

Figure 3B: Which Candidate Best Represents Women's Interests?

0.18 0.16

0.18

0.29

0

0.1

0.2

0.3

0.4

0.5

0.6

Control Traditional GenderFrame

Experimental Conditions

WomenMen

40

Figure 4: Effects of Traditional Race Frame and Traditional Gender Frame on Candidate Preferences

Figure 4A: Candidate Preference

0.320.26

0.60

0.41

0

0.1

0.2

0.3

0.4

0.5

0.6

Control Traditional RaceFrame

Experimental Conditions

WomenMen

Figure 4B: Candidate Preference

0.32

0.45

0.31

0.41

0

0.1

0.2

0.3

0.4

0.5

0.6

Control TraditionalGender Frame

Experimental Conditions

Gor

e

B

ush

WomenMen

41

Figure 5: Effects of Traditional Race Frame and Attitudes About Equality on Candidate Preferences

Figure 5A: Subjects Scoring High on Egalitarian Scale

0.26 0.23

0.84

0.40

00.10.20.30.40.50.60.70.8

Control Traditional RaceFrame

Experimental Conditions

WomenMen

Figure 5B: Subjects Scoring Low on Egalitarian Scale

0.290.40

0.42

0.100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Control Traditional RaceFrame

Experimental Conditions

WomenMen

Note: Subjects scoring one standard deviation above the mean on the equality scale are classified as “High” whereas those scoring one standard deviation below the mean are classified as “Low.”

42

Figure 6: Effects of Compassionate Race Frame on Candidate Preferences

0.23

0.350.35

0.57

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Traditional RaceFrame

CompassionateRace Frame

Experimental Conditions

WomenMen

43

Table 1. Ordered Logit Models Predicting Effects of Watching GOP Convention on Perception of Candidate’s Ability to Handle Race Relations,

Education, and Abortion (White Respondents Only)

Race Relations Education Abortion Female -.26

(.27) .23

(.30) -.29

(.27) Watch GOP Convention

-.28 (.33)

-.16 (.37)

-.28 (.34)

Female * Watch Convention

.94* (.44)

.67 (.50)

.63 (.44)

Control Variables Party Identification -.54***

(.05) -.77***

(.06) -.53***

(.05) Ideology -.30***

(.10) .41***

(.11) -.62***

(.10) Education -.08

(.05) -.13*

(.06) -.18***

(.05) Household Income -.09

(.06) .04

(.06) -.04 (.05)

Age .01*

(.00) .00

(.01) -.00

(.00)

(Continued on Next Page)

44

Table 1. Ordered Logit Models Predicting Effects of Watching GOP Convention on Perception of Candidate’s Ability to Handle Race Relations, Education, and

Abortion (Continued from Previous Page)

Race Relations Education Abortion Union Membership .07

(.24) -.27

(.26) .14

(.24) Voting Habits -.02

(.07) .10

(.08) .01

(.08) Cut 1 -3.09***

(.49) -3.70***

(.54) -4.66***

(.52) Cut 2 -2.46***

(.48) -3.23***

(.53) -3.93***

(.51)

Log likelihood -666.47 -533.07 -651.20 N 800 800 800

Pseudo R2 .14 .25 .18 Notes: * p < .05; ** p < .01; *** p < .001 for one-tailed test, except constant. The dependant variable in column 1 is “Next, regardless of which presidential candidate you support, please tell me if you think Al Gore or George W. Bush would better handle each of the following issues. How about Race Relations?” Response options range from Gore to “Same” or DK to Bush. Higher values indicate that Bush is more effective. The dependant variable in columns 2 and 3 refer to “education” and “abortion” respectively.

45

Table 2. Regression Models Predicting Effects of Watching GOP Convention on Approval and Likelihood of Voting for George W. Bush

(White Respondents Only) Bush Approval

(Logistic Model) Likelihood of Vote for Bush

(OLS Model) Intercept 5.37***

(.69) 1.21***

(.06) Female -.13

(.37) -.00

(.04) Watch GOP Convention

-.39 (.47)

-.02 (.05)

Female * Convention 1.25*

(.62) .10*

(.06) Control Variables

Party Identification -.80*** (.07)

-.14*** (.01)

Ideology -.66***

(.13) -.09***

(.01) Education -.08

(.07) -.00 (.00)

Household Income .29***

(.07) .03***

(.01) Age -.01

(.01) -.00

(.00) Union Membership -.58*

(.30) -.07*

(.03) Voting Habits -.14

(.10) .00

(.01) Chi Square 316.39*** -----

Log Likelihood -295.20 ----- Adjusted R2 ----- .50

N 777 801 Notes: * p < .05; ** p < .01; *** p < .001 for one-tailed test, except constant. The dependant variable in column 1 is “Next, we’d like to get your overall opinion of some people in the news. As I read each name, please say if you have a favorable or unfavorable opinion of this person – or if you have never heard of him or her. How about George W. Bush?” Response options range from “unfavorable” to “favorable.” The dependent variable in column 2 is “Next, if each of the following candidates were on the ballot for president this November, please say how likely it is you would vote for each – very likely, somewhat likely, not too likely, not at all likely, or if you don’t know enough about that person to say. How about George W. Bush?” Respondents indicating that they are very likely to support Bush are coded high.