Supporting Information A. Districts in Sample

Male Republican
Female Democrat / Female Republican
Male Democrat
9 Open Seats
10 Rep Incumbents
6 Dem Incumbents / 5 Open Seats
8 Rep Incumbents
2 Dem Incumbents
AZ08 (Graf/Giffords)
CA22 (McCarthy/Beery)
CA50 (Bilbray/Busby)
CT04 (Shays/Farrell)
FL09 (Bilirakis/Busansky)
FL11 (Adams/Castor)
FL13 (Buchanan/Jennings)
IL06 (Roskam/Duckworth)
IL08 (McSweeney/Bean)
KS02 (Ryun/Boyda)
MN03 (Ramstad/Wilde)
MN04 (Sium/McCollum)
NH01 (Bradley/Shea-Porter)
NJ07 (Ferguson/Stender)
NV02 (Heller/Derby)
NV03 (Porter/Hafen)
NY11 (Finger/Clarke)
NY20 (Sweeney/Gillibrand)
OH13 (Foltin/Sutton)
PA06 (Gerlach/Murphy)
PA13 (Bhakta/Schwartz)
TX18 (Hassan/Jackson Lee)
WA08 (Reichert/Burner)
WI02 (Magnum/Baldwin)
WI04 (Rivera/Moore) / CA45 (Bono-Mack/Roth)
CT05 (Johnson/Murphy)
IL04 (Melichar/Gutierrez)
IL13 (Biggert/Shannon)
IL17 (Zinga/Hare)
KY03 (Northup/Yarmuth)
ME01 (Curley/Allen)
NY19 (Kelly/Hall)
OH18 (Padgett/Space)
OK05 (Fallin/Hunter)
PA04 (Hart/Altmire)
TX22 (Sekula-Gibbs/Lampson)
VA01 (Davis/O’Donnell)
VA02 (Drake/Kellam)
VTAL (Rainville/Welch)

Supporting Information B. An Alternative Measure of Valence

Because women are expected to be more honest and collaborative, stereotypes about women may bias the valence measure in favor of women candidates by emphasizing characteristics that women stereotypically possess. To address this concern, I construct a second valence measure, excluding the “integrity” and “ability to work well with others” items. This second valence measure concentrates on items where neither sex ought to have a distinct advantage (e.g.: “qualifications to hold office,” “overall strength as a public servant”), and items where male candidates might be expected to have a natural advantage due to stereotypes about instrumentality (e.g.: “competence,” “grasp of issues,” “ability to find solutions to problems”). To the extent that these items are tilted more towards a male advantage, this second measure poses a strict test of the hypotheses that female candidates are valence-advantaged.

Like the original measure, the alternative valence measure produces a valence advantage for females equal to 0.657 (p<0.01, two-tailed). This finding corroborates Hypothesis 1a and prior research reporting a gender gap in candidate valence.

Table B1 shows that once the gender gap in valence is taken into account, there is a direct and negative effect of sex on vote-share. This alternative measure of valence produces a vote disadvantage of 3.6% for female candidates, which is in line with the results reported in the paper. All other coefficients are similarly-sized and significant.

Table B1. Candidate’s Aggregate Vote-Share with Alternative Valence Measure

District Analysis
Model B1: Controls for Quality
(Excludes Gendered Items)
Female Candidate / -3.621**
(1.647)
Incumbent Party Candidate / 4.204
(3.319)
Open Seat / -0.293
(1.791)
Challenger Office-Holding Experience / 1.349
(1.736)
District Partisanship (Candidate-Party Previous Presidential Vote) / 0.529***
(0.120)
Congressional Quarterly Key Race / 1.284
(1.932)
Candidate’s Relative Spending Advantage (Log) / 0.595*
(0.338)
Candidate’s Previous Vote-Share / 0.106
(0.090)
Candidate’s Relative Personal Quality Advantage (Excludes Gendered Items) / 2.801***
(0.656)
Constant / 23.110***
(6.562)
F-test / 25.05***
Adjusted R-squared / 0.851
Number of Observations / 39

Analysis excludes uncontested and same-sex races. Cell entries represent regression coefficients with standard errors in parentheses. Two-tailed significance tests: * p0.10, **p0.05, ***p0.01.

To see whether gender matters to partisans and/or independents, Table B2 replicates Tables 2 and 3 from the paper, using the alternative measure of valence. Table B2shows that Democrats and Republicans are equally likely to support a female candidate, even after holding valence constant. But, for independents, candidate sex emerges as a significant and negative predictor of the vote when valence is controlled for. Male independents are predicted to have a 22.5% reduced likelihood of supporting a female candidate. The size of this effect is consistent with Table 3 from the main paper.

Although some of the coefficient sizes and significances of the control variables depart from the main models presented in Tables 2 and 3, the variables of theoretical interest (i.e., candidate sex and valence), yield results identical in direction, size and significance.

Table B2. Likelihood of Voting for a Candidate with Alternative Valence Measure

Democrats / Republicans / Independents
Model B2a: Controls for Quality
(Excludes Gendered Items) / Model B2b: Controls for Quality
(Excludes Gendered Items) / Model B2c:
Controls for
Quality
(Excludes Gendered Items)
Female Candidate / -0.372
(0.830) / -0.636
(0.596) / -0.969**
(0.485)
Female Respondent / 0.590
(0.556) / 0.251
(0.901) / 0.042
(0.575)
Female Candidate x Female Respondent / 0.339
(0.857) / -0.082
(1.053) / 0.207
(0.691)
Incumbent Party Candidate / 2.932***
(0.961) / 0.497
(0.751) / 1.127*
(0.591)
Open Seat / -0.868**
(0.400) / -0.534
(0.553) / -0.579*
(0.300)
Challenger Office-Holding Experience / 0.558
(0.615) / 1.168**
(0.525) / 0.344
(0.328)
District Partisanship (Candidate-Party Previous Presidential Vote) / -0.116**
(0.051) / -0.008
(0.032) / -0.063**
(0.030)
Congressional Quarterly Key Race / 1.332*
(0.692) / -0.612
(0.722) / 0.995**
(0.481)
Candidate’s Relative Spending Advantage (Log) / 0.196*
(0.116) / 0.259***
(0.099) / 0.134**
(0.062)
Candidate’s Previous Vote-Share / -0.039
(0.027) / 0.011
(0.027) / 0.002
(0.011)
Respondent’s External Distance to Candidate / -0.450
(0.290) / 0.178
(0.375) / 0.168
(0.241)
Respondent’s External Distance to Opponent / 0.320
(0.396) / -0.133
(0.319) / -0.291
(0.215)
Respondent’s Ideology / 0.302
(0.432) / -0.348
(0.425) / -0.755**
(0.332)
Respondent’s Presidential Approval / -1.219***
(0.334) / -1.161***
(0.210) / -1.288***
(0.225)
Respondent’s Political Knowledge / 1.791**
(0.781) / -1.380
(0.965) / 1.114
(0.700)
Respondent’s Education / -0.146
(0.121) / 0.234
(0.167) / -0.295***
(0.110)
Respondent’s Age / -0.019
(0.013) / -0.004
(0.022) / 0.014
(0.012)
Respondent’s Issue Conservatism / -1.424***
(0.401) / -1.717***
(0.468) / -1.983***
(0.365)
Respondent’s Income / -0.046
(0.064) / -0.032
(0.072) / 0.063
(0.038)
Candidate’s Relative Personal Quality Advantage (Excludes Gendered Items) / 0.511*
(0.286) / -0.073
(0.175) / 0.279*
(0.164)
Constant / 9.455*
(5.442) / 3.804
(3.337) / 7.941***
(2.786)
F-test / 5.76*** / 7.13*** / 12.69***
Number of Respondents / 669 / 719 / 734
Number of Districts / 39 / 39 / 39

Analysis excludes uncontested and same-sex races. Cell entries represent logit coefficients with standard errors in parentheses. Two-tailed significance tests: * p0.10, **p0.05, ***p0.01.

Supporting InformationC. Principal Components Factor Analysis

The principal components analysis shown in Table C1 confirms that the seven variables form a single dimension. 78% of the variance in the seven measures is accounted for by the first dimension, while the second dimension only accounts for 7.6% of the variance. Moreover, only the first dimension has an Eigenvalue greater than one, therefore only one factor is retained.

Table C1. Principal Components Factor Analysis

Factors / Eigenvalue / Difference / Proportion / Cumulative
Factor 1 / 5.461 / 4.929 / 0.780 / 0.780
Factor 2 / 0.532 / 0.231 / 0.076 / 0.856
Factor 3 / 0.301 / 0.049 / 0.043 / 0.899
Factor 4 / 0.252 / 0.039 / 0.036 / 0.935
Factor 5 / 0.213 / 0.073 / 0.031 / 0.966
Factor 6 / 0.140 / 0.040 / 0.020 / 0.986
Factor 7 / 0.100 / . / 0.014 / 1.000

Table C2. Factor Loadings and Unique Variances

Variables / Factor 1 / Uniqueness
Personal Integrity / 0.735 / 0.460
Ability to Work Well with Others / 0.883 / 0.220
Ability to Find Solutions to Problems / 0.912 / 0.168
Competence / 0.914 / 0.164
Grasp of Issues / 0.897 / 0.195
Qualifications to Hold Office / 0.899 / 0.191
Strength as Public Servant / 0.928 / 0.139

As shown in Table C2, each variable correlates highly with the first dimension, with the average correlation equal to 0.881. Of all the items in the index, Personal Integrity correlates the worst with the first factor, at 0.735, but is still quite strong by any standard. In contrast, the Ability to Find Solutions to Problems and Competence correlate the greatest, at 0.91. Overall, the results validate reducing the data into a single personal quality index.

Supporting InformationD. Macro-Analysis Variables

Variable / Coding / Mean & (SD)
Candidate’s Vote-Share / 0 to 100 / 54.466
(13.229)
Female Candidate / 0 = Male Democrat
1 = Female Democrat / 0.625
(0.490)
Incumbent Party Candidate / 0 = No
1 = Yes / 0.325
(0.474)
Open Seat / 0 = No
1 = Yes / 0.35
(0.483)
Challenger Office-Holding Experience / 0 = No Experience
1 = Experienced / 0.500
(0.506)
District Partisanship (Candidate-Party Previous Presidential Vote) / 20.994 to 87.562 / 50.624
(11.503)
Congressional Quarterly Key Race / 0 = Uncompetitive
1 = Rated as a “Toss Up” / 0.200
(0.405)
Candidate’s Relative Spending Advantage (Log) / Log ($1 to $8,112,752) / 0.575
(4.428)
Candidate’s Previous Vote-Share / 0 to 100 / 44.992
(22.248)
Candidate’s Relative Personal Quality Advantage / -6 = Republican Strongly Advantaged
6 = Democrat Strongly Advantaged / 0.072
(1.259)

Supporting InformationD. Micro-Analysis Variables

Variable / Coding / Mean & (SD)
Vote for Candidate / 0 = Vote Democratic
1 = Vote Republican / 0.519
(0.500)
Respondent Partisanship / -1 = Strong or Somewhat Strong Democrat
0 = Lean Partisan or Independent
1 = Strong or Somewhat Strong Republican / 0.013
(0.799)
Female Candidate / 0 = Male Democrat
1 = Female Democrat / 0.656
(0.475)
Female Respondent / 0 = Male
1 = Female / 0.511
(0.500)
Female Candidate* Female Respondent / 0 = Male Democratic Candidate & Male Respondent
1 = Female Democratic Candidate & Female Respondent / 0.333
(0.471)
Incumbent Party Candidate / 0 = No
1 = Yes / 0.295
(0.456)
Open Seat / 0 = No
1 = Yes / 0.375
(0.484)
Challenger Office-Holding Experience / 0 = No Experience
1 = Experienced / 0.500
(0.500)
District Partisanship (Candidate-Party Previous Presidential Vote) / 20.994 to 87.562 / 49.351
(10.099)
Congressional Quarterly Key Race / 0 = Uncompetitive
1 = Rated as a “Toss Up” / 0.206
(0.405)
Candidate’s Relative Spending Advantage (Log) / Log ($1 to $8,112,752) / 0.239
(3.606)
Candidate’s Previous Vote-Share / 0 to 100 / 43.301
(20.395)
External Ideological Distance to Candidate (Absolute Value) / 0 = Respondent Places on Democrat’s Position
6 = Respondent Distant from Democrat’s Position / 2.221
(1.459)
External Ideological Distance to Opponent (Absolute Value) / 0 = Respondent Places on Republican’s Position
6 = Respondent Distant from Republican’s Position / 1.847
(1.349)
Respondent’s Ideology / 1 = Liberal
7 = Conservative / 4.381
(1.644)
Respondent’s Presidential Approval / 1 = Strongly Disapprove
4 = Strongly Approve / 2.105
(1.204)
Respondent’s Political Knowledge – Average Correctly Identifies Partisan Identification of Member of Congress, Governor, Senator 1, Senator 2 / 0 = Incorrectly Identifies All Officers
1 = Correctly Identifies All Officers / 0.838
(0.282)
Respondent’s Education / 1 = No High School
6 = Post-Graduate Degree / 3.397
(1.394)
Respondent’s Age / 18 to 95 / 49.233
(15.383)
Respondent’s Issue Conservatism – Factor Analysis of Voter’s Attitudes Towards Stem Cell Research, the Iraq War, Minimum Wage, Abortion, the Environment, Immigration, Social Security, Affirmative Action, Taxes and Free Trade / -1.927 (Liberal Views) to 1.924 (Conservative Views) / 0.004
(0.957)
Respondent Income / 1 = <$10K
14 = $150K+ / 8.607
(3.358)
Candidate’s Relative Personal Quality Advantage / -6 = Republican Strongly Advantaged
6 = Democrat Strongly Advantaged / 0.132
(1.260)

Supporting InformationE. Multi-Level Logit

Because my models include independent variables at both the district- and individual-level, I investigate whether my results are sensitive to my modeling approach. When the correlation between observations is neglected, the standard errors of the estimates can be biased downward, leading to invalid significance tests. To avoid these problems, I clustered at the district-level to account for correlations between observations within districts in the original models presented in the paper.

Another approach to dealing with correlated data is to estimate a multi-level model. Both clustering and multi-level modeling are valid approaches to dealing with correlated data, but the techniques can yield slightly different results. Below, I test whether my results persist using multi-level modeling. Because my dependent variable is binary, I do not use a hierarchical linear modeling technique, but rather re-estimate the models using a multi-level logit. Below, I replicate Tables 2 and 3 from my paper.

Table E shows the results of my multi-level logit for Democrats, Republicans and Independents. The results of the multi-level logit are consistent with the clustered models reported in the paper. All of the coefficients from the multi-level logit estimation are similarly-sized, signed and significant. As expected, the Female Candidate coefficient is insignificant for partisan identifiers.

At the bottom of the table, the intra-class latent correlation (rho) reports the proportion of the variance that is attributable to differences across districts, while the likelihood ratio tests the null hypothesis of no cross-district variation in the vote. The p-value on the likelihood ratio exceeds 0.10, meaning that the proportion of variance in the vote that is attributable to the district is not significantly different from zero.

Table E. Multi-Level Model of Likelihood of Voting for a Candidate

Democrats / Republicans / Independents
Model E1: Controls for Quality / Model E2: Controls for Quality / Model E3:
Controls for
Quality
Female Candidate / -0.468
(0.608) / -0.280
(0.530) / -0.893**
(0.460)
Female Respondent / 0.455
(0.674) / 0.406
(0.611) / 0.260
(0.476)
Female Candidate x Female Respondent / 0.040
(0.817) / -0.139
(0.748) / -0.025
(0.582)
Incumbent Party Candidate / 1.866*
(1.037) / 0.283
(0.756) / 1.094*
(0.673)
Open Seat / -1.029*
(0.569) / 0.031
(0.447) / -0.465
(0.412)
Challenger Office-Holding Experience / 0.594
(0.522) / 0.733
(0.460) / 0.085
(0.391)
District Partisanship (Candidate-Party Previous Presidential Vote) / -0.091**
(0.042) / -0.005
(0.031) / -0.062**
(0.032)
Congressional Quarterly Key Race / 1.323*
(0.707) / -0.731
(0.517) / 0.850*
(0.444)
Candidate’s Relative Spending Advantage (Log) / 0.097
(0.114) / 0.135
(0.093) / 0.101
(0.075)
Candidate’s Previous Vote-Share / -0.014
(0.026) / 0.023
(0.021) / 0.009
(0.016)
Respondent’s External Distance to Candidate / -0.152
(0.279) / 0.089
(0.300) / 0.095
(0.230)
Respondent’s External Distance to Opponent / 0.215
(0.285) / 0.141
(0.245) / -0.159
(0.223)
Respondent’s Ideology / 0.238
(0.379) / -0.155
(0.353) / -0.430
(0.302)
Respondent’s Presidential Approval / -1.221***
(0.277) / -1.274***
(0.217) / -1.120***
(0.172)
Respondent’s Political Knowledge / 0.938
(0.731) / -0.830
(0.734) / 0.730
(0.546)
Respondent’s Education / -0.042
(0.162) / 0.070
(0.143) / -0.266**
(0.111)
Respondent’s Age / -0.015
(0.014) / -0.016
(0.013) / 0.009
(0.010)
Respondent’s Issue Conservatism / -1.544***
(0.509) / -1.459***
(0.434) / -2.188***
(0.372)
Respondent’s Income / 0.002
(0.065) / 0.029
(0.060) / 0.087**
(0.043)
Candidate’s Relative Personal Quality Advantage / 0.514***
(0.191) / 0.034
(0.180) / 0.281*
(0.149)
Constant / 7.211*
(3.777) / 2.132
(2.616) / 6.149**
(2.606)
rho / 0.000 / 0.000 / 0.052
Likelihood Ratio / 0.000 / 0.000 / 0.95
Wald Chi-Squared / 79.00*** / 104.95*** / 177.23***
Number of Respondents / 669 / 719 / 734
Number of Districts / 39 / 39 / 39

Analysis excludes uncontested and same-sex races. Cell entries represent multi-level logit coefficients with standard errors in parentheses. Two-tailed significance tests: * p0.10, **p0.05, ***p0.01.

Table E also shows that Independents in the multi-level model behave differently from their partisan counterparts. Consistent with the clustered models presented in the paper, the Female Candidate coefficient emerges as significant once Personal Quality is controlled for.

The results from the multi-level logit confirm the findings from my original clustered models: the likelihood of voting for women candidates does not depend on quality for partisans. However, once Personal Quality is controlled for, Female Candidate emerges as a significant and negative predictor of the vote for independents.

Importantly, in Model E3only the coefficient for Female Candidate is significant. The combined coefficient for Female Candidate + Female Respondent + Female Candidate x Female Respondent is insignificant. This is consistent with Model 8 from the main paper, and indicates that the presence of a female candidate reduces the propensity to support for male independents, but not for female independents.

Supporting Information F. Comparison of Means: Male Independents and Sample

Sample Mean / Male Independents
Ideology / 4.338 / 4.360
Presidential Approval / 2.130*** / 1.944
Political Knowledge / 0.759 / 0.845***
Political Interest / 2.429 / 2.662***
Education / 3.069 / 3.419***
Age / 44.214 / 44.132
Issue Conservatism / -0.001 / 0.007
Employed Full-Time / 0.518 / 0.660***
Homeownership / 0.698 / 0.699
Income / 8.043 / 8.558***
White / 0.751*** / 0.717
Married / 0.607 / 0.603
Protestant / 0.306 / 0.289
Church Attendance / 2.158*** / 1.996
Religious Importance / 0.700*** / 0.584
Born Again Christian / 0.399 / 0.378
Children in Household / 0.344 / 0.346

Cell entries represent means. Two-tailed significance tests: * p0.10, **p0.05, ***p0.01.