The Contingent Effects of Candidate Sex on Voter Choice Appendix

Appendix The Contingent Effects of Candidate Sex on Voter Choice The purpose of this appendix is to provide additional information about topics that could not be fully addressed in the manuscript. 1. Sampling Methodology The sample has been stratified by the region of residence (Northeast, Midwest, South, and West), sex (male and female), race/ethnicity (white, black, Hispanic, Asian, and others), and age groups, based on the 2010 U.S. Census data. In collecting data, we first asked demographic screening questions to 3,152 people in the SSI panel and invited 1,733 people among them according to fixed quotas so that the total sample matched the adult U.S. Census population. A total of 1,583 respondents completed the conjoint experiment tasks in our survey. Detailed descriptive statistics on our sample are shown in Table A1. 1 Table A1. Sample Descriptive Statistics Number Percentage Sex Male Female 766 817 48.4% 51.6% Education level No BA degree 888 57.1% 666 42.9% Age group 18-29 years old 305 19.3% 30-50 years old 678 42.8% 51-65 years old 66 years old or older 450 150 28.4% 9.5% Social class Lower class 481 31.0% Middle class Upper class 812 261 52.3% 16.8% 569 35.9% 1014 64.1% 1055 66.7% Black 190 12.0% Hispanic Other 226 112 14.3% 7.1% Partisanship Democratic Party 614 41.0% Republican Party Independent 467 414 31.2% 27.7% BA degree Region South Nonsouth Race/Ethnicity White Note: DK/NA respondents have been excluded. 2 Table A2 displays how our sample compares to the Census. Each cell reports the deviation between the Census and our sample for a particular group. An example is black females aged 65 and over in the Midwest where the share our sample differs from the Census by just -.10 percentage points. In most cases the deviations are extremely small, always below one percentage point, and are always within the margin of error. The results show that our samples quite accurately reflect the entire U.S. population and weighting would not change the results materially. We also checked to see if the response rate (the ratio of missing data) for our conjoint experiment question varies across groups, and found no much variation that might introduce significant bias into our results. Table A2. Deviations of Key Demographic Characteristics from U.S. Census Hispanic Region Northeast Midwest South West Age Excluding Hispanic White Black Asian Other Male Female Male Female Male Female Male Female Male Female 18-24 -0.01 -0.05 -0.35 -0.27 0.04 -0.15 -0.01 0.05 -0.02 0.04 25-44 0.09 0.35 0.95 0.49 0.23 0.06 0.03 0.07 -0.04 0.08 45-64 0.17 0.02 0.99 0.27 -0.04 0.09 -0.09 -0.10 0.03 0.09 65-over -0.07 -0.11 -0.47 -0.14 -0.10 -0.16 -0.05 -0.06 -0.01 0.05 18-24 -0.07 -0.11 -0.47 -0.63 -0.11 0.08 -0.05 -0.04 -0.04 0.02 25-44 0.05 -0.17 -0.18 0.35 -0.12 0.10 0.06 -0.14 0.06 -0.07 45-64 -0.09 0.11 0.08 0.89 0.06 0.20 0.12 -0.02 -0.05 0.08 65-over -0.03 -0.04 -0.59 -1.14 0.02 -0.10 -0.02 -0.03 0.05 -0.02 18-24 0.01 0.02 -0.87 -0.14 0.17 -0.10 -0.07 0.00 0.07 0.13 25-44 -0.08 -0.22 0.49 0.14 -0.12 0.22 -0.04 -0.01 -0.06 0.18 45-64 0.09 0.45 -0.06 1.10 -0.09 -0.14 0.17 -0.05 0.03 0.21 65-over 0.05 -0.21 -0.46 -1.62 -0.06 -0.04 -0.04 -0.06 -0.03 -0.04 18-24 0.19 -0.20 -0.36 0.07 -0.02 -0.01 -0.08 -0.13 -0.03 -0.09 25-44 -0.05 0.40 0.65 0.19 0.05 0.00 0.22 -0.09 0.07 0.07 45-64 -0.11 -0.06 0.45 0.81 0.07 0.27 -0.02 -0.16 0.00 0.05 65-over -0.07 -0.20 -0.11 -0.50 0.07 -0.07 -0.08 0.06 -0.04 -0.05 3 2. Question Wording To test the theory of gender-office congruency, we manipulated question wording and asked our respondents to evaluate candidates both for president and for Congress. In our conjoint experiment, we presented a total of ten pairs of candidates, and changed the introductory script shown before asking respondents to review candidate profiles for the first five pairs and for the last five pairs, respectively. We prepared for the two types of introductory script as follows (one is for asking respondents to assume a presidential election, and the other is for asking respondents to assume a congressional election): Introductory script for a presidential election Suppose you were asked to choose between two candidates for U.S. president. In the following questions, we will show you five pairs of potential candidates running for the U.S. presidential election. For each pair, please indicate which of the two candidates you would like to vote for. Introductory script for a congressional election Now, suppose you were asked to choose between two candidates for U.S. Congress. In the following questions, we will show you five pairs of potential candidates running for the U.S. House of Representatives. For each pair, please indicate which of the two candidates you would like to vote for. In addition to manipulating the introductory script, for every pair, we presented a pair of candidate profiles with the following question wording so as to remind our respondents that they are evaluating candidates running for either president or Congress: 4 Question about a presidential election Please carefully review the two potential candidates, running in the U.S. presidential election, detailed below. If you had to choose between them, which of these candidates would you vote to be U.S. President? Question about a congressional election Please carefully review the two potential candidates running for election to the U.S. House of Representatives, detailed below. If you had to choose between them, which of these candidates would you vote to be a member of the U.S. House of Representatives? 3. Effects of Electoral Competition Context In our conjoint experiment, we ask individual subjects to choose between a pair of candidates after reviewing their profiles. The candidate profiles shown to our subjects are generated randomly from the set of characteristics. Thus, the combination of candidate profiles is set at random and varies across pairs of candidates. For instance, some cases ask subjects to review two candidates competing between different sexes (male vs. female candidates), while others ask subjects to choose between a pair of same-sex candidates (male vs. male candidates or female vs. female candidates). Similarly, some cases ask subjects to evaluate between a pair of candidates running from the same party (Democrat vs. Democrat candidates or Republican vs. Republican candidates), while others ask subjects to evaluate those who are running from different parties (Democrat vs. Republican candidates). This section explores the effects of such different contexts of electoral competition on voter decisions. First, we examine the cases when electoral competition is between two candidates with different sexes (male vs. female candidates) to identify the effect of candidate sex on vote choice more precisely. Our original findings based on all the candidate pairings in our dataset shown in 5 Figure 2 suggest that voters overall are 1.3 percentage points less likely to vote for a female candidate. However, there is a concern that this effect may have been underestimated because this is the average effect including cases that same-sex candidates are competing. Among 15,840 evaluated pairings in our data, 7,930 pairings (50.1%) indeed have the identical sex between the two candidates.1 Thus, we reanalyze data by eliminating these same-sex pairings to focus exclusively on the cases where a pair of candidates shown to subjects have different sexes. Figure A1 shows the results based on different-sex candidate parings. The dots denote point estimates for the AMCEs, which indicate the average effect of each attribute on the probability that the candidate will be chosen. The horizontal bars show 95% confidence intervals. The findings suggest that, when the competition for electoral office is held between male and female candidates, voters are 2.5 percentage points less likely to choose a female candidate simply because the candidate is a woman. In short, the bias against female candidates almost doubles under such circumstances. The effects of other candidate attributes remain almost the same as the average ones in Figure 2, but the favorability rating among the public seems to be slightly less important when voters evaluate different-sex candidate pairings. 1 Whether a pair of candidates have the same sex in their profiles is determined by chance. 6 Figure A1. Marginal Effect of Candidate Attributes on Voting Decisions (different-sex parings) All Respondents (N = 1583) Sex: (Baseline = Male) Female Age: (Baseline = 36 years old) 44 years old 52 years old 60 years old 68 years old 76 years old Race/Ethnicity: (Baseline = White) Black Hispanic Asian American Family: (Baseline = Single (never married)) Single (divorced) Married (no child) Married (tw o children) Experience in public office: (Baseline = No experience) 4 years 8 years 12 years Salient personal characteristics: (Baseline = Provides strong leadership) Really cares about people like you Honest Know ledgeable Compassionate Intelligent Party affiliation: (Baseline = Democratic Party) Republican Party Policy area of expertise: (Baseline = Environmental issues) Foreign policy Public safety (crime) Economic policy Health care Education Position on national security: (Baseline = Cut military budget) Maintain strong defense Position on immigrants: (Baseline = Favors giving guest w orker status) Opposes giving guest w orker status Position on abortion: (Baseline = No opinion (neutral)) Pro-choice Pro-life Position on government deficit: (Baseline = Don't reduce deficit now ) Reduce deficit through tax increase Reduce deficit through spending cuts Favorability rating among the public: (Baseline = 34%) 43% 52% 61% 70% -.1 0 Change in Pr(Preferred Candidate) .1 Note: Plots show the estimated effects of the randomly assigned candidate attributes on the probability of being supported by voters. Bars represent 95% confidence intervals. Only different-sex candidate pairings (15,820 evaluated profiles) have been included for estimating the results. 7 Second, we compare the results of competitions between different-party candidate pairings (Democrat vs. Republican candidates) and same-party candidate parings (Democrat vs. Democrat candidates or Republican vs. Republican candidates).2 In our data, among 15,840 evaluated pairings, 7,809 pairings (49.3%) have the identical party label between the two candidates, and the rest of them (8,031 pairings) have different party labels between the two candidates. Figures A2 and A3 illustrate the results for different-party pairings and same-party pairings, respectively. The results show that, when electoral competition is between two candidates from different parties, independents display a substantial negative bias against female candidates of 3.6 percentage points while both Democrats and Republicans show no hostility toward female candidates (0.6 and 0.8 percentage points, respectively).3 In contrast, when electoral competition is between the two candidates running from the same party, independents and Republicans show biases against female candidates of 2.9 and 2.8 percentage points, respectively, while Democrats remain to have no bias against female candidates (1.2 percentage points reward to female candidates, but it is not statistically significant at the 10% level). 2 Our experiment permitted competition between opposite party candidates and candidates running from the same party. Thus, we are able to examine the difference between these distinct contexts, which roughly approximate general elections and primary elections, respectively. 3 Those partisan voters mostly determine their vote choice by a candidate’s party label. 8 Figure A2. Marginal Effects of Candidate Attributes on Voting Decisions by Respondent Party Identification (different-party parings) Conditional on Partisanship = Democrat Conditional on Partisanship = Independent -.3 -.2 -.1 0 .1 .2 .3 -.3 -.2 -.1 0 .1 .2 .3 Conditional on Partisanship = Republican Sex: (Baseline = Male) Female Age: (Baseline = 36 years old) 44 years old 52 years old 60 years old 68 years old 76 years old Family: (Baseline = Single (never married)) Married (no child) Married (two children) Single (divorced) Race: (Baseline = White) Asian American Black Hispanic Experience in public office: (Baseline = No experience) 12 years 4 years 8 years Salient personal characteristics: (Baseline = Provides strong leadership) Compassionate Honest Intelligent Knowledgeable Really cares about people like you Party affiliation: (Baseline = Democratic Party) Republican Party Policy area of expertise: (Baseline = Environmental issues) Economic policy Education Foreign policy Health care Public safety (crime) Position on national security: (Baseline = Cut military budget) Maintain strong defense Position on immigrants: (Baseline = Favors giving guest worker status) Opposes giving guest worker status Position on abortion: (Baseline = No opinion (neutral)) Pro-choice Pro-life Position on government deficit: (Baseline = Don't reduce deficit now) Reduce deficit through spending cuts Reduce deficit through tax increase Favorability rating among the public: (Baseline = 34%) 43% 52% 61% 70% -.3 -.2 -.1 0 .1 .2 .3 Change in Pr(Preferred Candidate) Note: Plots show the estimated effects of the randomly assigned candidate attributes on the probability of being preferred to vote in elections. They are separately estimated for the group of respondents who are Democrats, Independents, and Republicans, respectively. Bars represent 95% confidence intervals. Only different-party candidate pairings (16,062 evaluated profiles) have been included for estimating the results. Note that the range of x axis differs from Figure A3 and other figures because the effect of a candidate’s party affiliation is so large among partisan voters relative to the effects of other candidate attributes. 9 Figure A3. Marginal Effects of Candidate Attributes on Voting Decisions by Respondent Party Identification (same-party parings) Conditional on Partisanship = Democrat Conditional on Partisanship = Independent Conditional on Partisanship = Republican Sex: (Baseline = Male) Female Age: (Baseline = 36 years old) 44 years old 52 years old 60 years old 68 years old 76 years old Family: (Baseline = Single (never married)) Married (no child) Married (two children) Single (divorced) Race: (Baseline = White) Asian American Black Hispanic Experience in public office: (Baseline = No experience) 12 years 4 years 8 years Salient personal characteristics: (Baseline = Provides strong leadership) Compassionate Honest Intelligent Knowledgeable Really cares about people like you Party affiliation: (Baseline = Democratic Party) Republican Party Policy area of expertise: (Baseline = Environmental issues) Economic policy Education Foreign policy Health care Public safety (crime) Position on national security: (Baseline = Cut military budget) Maintain strong defense Position on immigrants: (Baseline = Favors giving guest worker status) Opposes giving guest worker status Position on abortion: (Baseline = No opinion (neutral)) Pro-choice Pro-life Position on government deficit: (Baseline = Don't reduce deficit now) Reduce deficit through spending cuts Reduce deficit through tax increase Favorability rating among the public: (Baseline = 34%) 43% 52% 61% 70% -.1 0 .1 .2 -.1 0 .1 .2 -.1 0 .1 .2 Change in Pr(Preferred Candidate) Note: Plots show the estimated effects of the randomly assigned candidate attributes on the probability of being preferred to vote in elections. They are separately estimated for the group of respondents who are Democrats, Independents, and Republicans, respectively. Bars represent 95% confidence intervals. Only same-party candidate pairings (15,618 evaluated profiles) have been included for estimating the results. 10 4. Party Identification and Attitudes toward Parties We measure party identification as well as attitudes toward each party using the following questions. The order of the parties in the question has been randomized to minimize bias. Question about party identification Generally speaking, do you usually think of yourself as a Republican (Democrat), a Democrat (Republican), an Independent, or something else? Question about attitude toward each party What do you think about each of the following political parties? Please rate it on a scale from 0 to 10, where 0 means you strongly dislike that party and 10 means you strongly like that party. If you feel you do not know enough about the party, just choose “don’t know.” Among independents, we take the difference between their rates for Democratic Party and for Republican Party to see if they have any leaning attitude toward either party. Figure A4 shows the number of independents for each category of relative preference. 11 Figure A4. Relative party preference among independents 180 160 140 120 100 80 60 40 20 0 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 Note: A negative value on the x-axis indicates the respondent is leaning toward the Republican Party, while a positive value indicates the respondent is leaning toward the Democratic Party. 5. Political Interest and Bias against Female Candidates We asked respondents to report the extent to which they are interested in politics on a four-point scale from not at all interested (1) to very interested (4). Figure A5 summarizes the results, which indicate the number of respondents for each category. As shown in this figure, most of our respondents (85.7%) report they are either somewhat or very interested in politics. 12 Figure A5. Political interest among respondents 800 700 600 500 400 300 200 100 0 Not at all interested Not very interested Somewhat interested Very interested To examine whether the bias against female candidates varies depending on the extent to which respondents are interested in politics, we test for heterogeneous treatment effects by including the interaction between candidates’ attributes and respondents’ political interest. Figure A6 shows the results of voter bias against female candidates. While respondents who are more interested appear to have a slightly larger bias against female candidates, we find no statistically significant difference between more and less interested respondents (p > .10). Thus, we believe there is no much difference in their bias against female candidates between actual voters and our samples matched to the adult population. 13 Figure A6. Marginal Effects of Candidate Sex on Voting Decision by Political Interest Female Candidate Disadvantage By Respondent Political Interest Not at all interested Not very interested Somew hat interested Very interested -.1 -.05 0 Change in Pr(Preferred Candidate) .05 .1 Note: Plots show the estimated effects of the randomly assigned candidate sex (female) on the probability of being supported by voters, conditional on the extent of respondent political interests. Bars represent 95% confidence intervals. 6. Implausible Combinations of Candidate Attributes The conjoint experiment creates many different combinations of candidate attributes. Almost most combinations will seem plausible to respondents, some candidate profiles will not. For example, political parties represent distinct political positions that make it likely that Republican candidates will have conservative views and Democratic candidates will have liberal positions. In our conjoint experiment, however, candidates are randomly assigned to take a particular position in four policy areas: national security, immigration, abortion, and the federal deficit. For each policy issue, there are both liberal and conservative positions. Hence, some combinations of candidate profiles may appear less credible and cause respondents not to take the exercise seriously. To examine whether implausible combinations introduce any biases to the results by leading our subjects to make artificial judgments without much cognitive effort, we focus 14 attention to some of the least plausible combinations of candidate characteristics to see if the results differ in thee settings. In particular, we highlight instances where candidates have contradictory positions from their parties on all of the four policy domains simultaneously. These would be of two types: (1) Democratic candidates who are hawkish on military, hostile towards undocumented immigrants, pro-life, and supportive to cut government spending and (2) Republican candidates who are dovish on military, welcoming to undocumented immigrants, pro-choice, and supportive to tax increase. In our data, the number of such implausible candidates is quite small: only 856 out of 15,830 pairs (5.4%) feature at least one candidate of this type. First, we investigate whether our subjects make artificial judgments without paying much cognitive effort when they are exposed to such pairs in our conjoint experiment. We follow other research that uses response times to measure effort exerted by respondents (Berinsky, Margolis, and Sances 2013; Malhotra 2008). In the full sample our subjects on average spent 38.5 seconds to complete each task. While it took only 35.9 seconds for subjects to complete the task when they are asked to choose from a pair of plausible candidates, it took 82.3 seconds to complete the task when they evaluate a pair of candidates with at least one implausible combination. The difference of response time between them is statistically significant (p < .01).4 Thus, we find no evidence to suggest that our subjects 4 There are some extreme values in our response time data. We suspect that this is because some subjects interrupted their conjoint tasks in our survey. To avoid biases causes by such extreme cases, we reanalyze the data after eliminating observations in top percentile. The results show that subjects who were exposed to implausible candidates still spent 1.7 seconds more time on the evaluation task on average than those without such exposures (p < .10). 15 make artificial judgments to save cognitive effort when they are exposed to candidate pairs with implausible profile combinations. Rather, the failure of a candidate to fit expectations results in a more effortful evaluation. Second, we exclude pairs with these implausible combinations to see if voters still punish female candidates when they are not exposed to such combinations. Figure A7 shows the results after excluding these pairs. The results remain similar to the initial results shown in Figure 2. Respondents are still less likely to vote for a female candidate, and the negative bias against female candidates is, on average, about 1.2 percentage points. 16 Figure A7. Marginal Effect of Candidate Attributes on Voting Decision after Excluding Candidate Pairs with Implausible Combinations All Respondents (N = 1583) Sex: (Baseline = Male) Female Age: (Baseline = 36 years old) 44 years old 52 years old 60 years old 68 years old 76 years old Race/Ethnicity: (Baseline = White) Black Hispanic Asian American Family: Single (divorced) Married (no child) Married (tw o children) (Baseline = Single (never married)) Experience in public office: (Baseline = No experience) 4 years 8 years 12 years Salient personal characteristics: (Baseline = Provides strong leadership) Really cares about people like you Honest Know ledgeable Compassionate Intelligent Party affiliation: (Baseline = Democratic Party) Republican Party Policy area of expertise: (Baseline = Environmental issues) Foreign policy Public safety (crime) Economic policy Health care Education Position on national security: (Baseline = Cut military budget) Maintain strong defense Position on immigrants: (Baseline = Favors giving guest w orker status) Opposes giving guest w orker status Position on abortion: (Baseline = No opinion (neutral)) Pro-choice Pro-life Position on government deficit: (Baseline = Don't reduce deficit now ) Reduce deficit through spending cuts Reduce deficit through tax increase Favorability rating among the public: (Baseline = 34%) 43% 52% 61% 70% -.1 -.05 0 .05 Change in Pr(Preferred Candidate) .1 Note: Plots show the estimated effects of the randomly assigned candidate attributes on the probability of being supported by voters. Bars represent 95% confidence intervals. 17 7. “Hillary Effect” on the Office Congruency We examine whether attitudes toward Hillary Clinton underlie views about a female president by using a measure of the degree to which respondents like or dislike Clinton. We draw on the following item, which is contained in a battery of questions about various political figures in our survey. The distribution of respondents on their favorability toward Hillary Clinton is shown below the question. What do you think about each of the following politicians? Please rate it on a scale from 0 to 10, where 0 means you strongly dislike that politician and 10 means that you strongly like that politician. If you have not heard of, or you feel you do not know enough about the politician, just choose “don’t know.” Figure A8. Favorability toward Hillary Clinton 0 5 10 15 20 25 0 (strongly dislike) 1 2 3 4 5 (neutral) 6 7 8 9 10 (strongly like) DK NA Note: Bars indicate percentages of respondents in each category. 18 30 As discussed in the text, we found that voters punish female candidates running for president while they do not in congressional elections. If this is because attitudes toward Hillary Clinton underlie views about a female president, we should not observe such a pattern of attitude change among those who are neutral or like Clinton. Figure A9 shows the results after excluding those who dislike Clinton. We find that attitudes toward Clinton do not necessarily lead respondents to show a greater bias against a female president. Even among the respondents who indicate they are neutral or like Clinton, there still exists a tendency that female candidates face a greater challenge when they run for executive office than when they run for legislative office. The AMCE estimate of a female candidate is -.0140 (p < .10) in presidential elections and .0036 in congressional elections, respectively, though the difference between the two is not statistically significant. 19 Figure A9. Marginal Effect of Candidate Attributes on Voting Decision among Respondents who are Neutral or Like Hillary Clinton (by Office Types) Conditional on Office = Congress Conditional on Office = President Sex: (Baseline = Male) Female Age: (Baseline = 36 years old) 44 years old 52 years old 60 years old 68 years old 76 years old Family: (Baseline = Single (never married)) Married (no child) Married (two children) Single (divorced) Race: (Baseline = White) Asian American Black Hispanic Experience in public office: (Baseline = No experience) 12 years 4 years 8 years Salient personal characteristics: (Baseline = Provides strong leadership) Compassionate Honest Intelligent Knowledgeable Really cares about people like you Party affiliation: (Baseline = Democratic Party) Republican Party Policy area of expertise: (Baseline = Environmental issues) Economic policy Education Foreign policy Health care Public safety (crime) Position on national security: (Baseline = Cut military budget) Maintain strong defense Position on immigrants: (Baseline = Favors giving guest worker status) Opposes giving guest worker status Position on abortion: (Baseline = No opinion (neutral)) Pro-choice Pro-life Position on government deficit: (Baseline = Don't reduce deficit now) Reduce deficit through spending cuts Reduce deficit through tax increase Favorability rating among the public: (Baseline = 34%) 43% 52% 61% 70% -.2 -.1 0 .1 .2 -.2 -.1 0 .1 .2 Change in Pr(Preferred Candidate) Note: Plots show the estimated effects of the randomly assigned candidate attributes on the probability of being supported by voters. Bars represent 95% confidence intervals. 20