Women’s Electoral Success and Future Political Participation (Figures)

Figure 1: Preferences for Women Politicians -1 Female-Male vote margin -.5 0 .5 1 A. Vote Share of Women Politicians Across Different Victory Margins 0 .2 .4 .6 Vote share of all female candidates .8 1 Notes: Constituency-year observations for the sample of elections with at least one woman candidate. B. No Discontinuity in Preferences for Women Politicians 0 .2 .4 .6 .8 Mixed-gender election with female winner having margin > 0 -50 0 Victory Margin female vote share Fitted values (female) 50 Fitted values (male) Victory Margin Aggregated into 2.5% Bins Notes: Constituency-year observations for the sample of elections with at least one woman candidate. Female vote share is the vote share of all female candidates in the constituency; victory margin is the difference in vote share between the top two vote winners of opposite genders.The fitted values are a lowess line. Figure 2: Continuity of the Victory Margin between Female and Male Candidates 1.5 0 .5 1 Density 2 2.5 A. Density of the victory margin -1 -.5 0 margin of victory .5 1 0 .5 1 1.5 2 2.5 B. Testing for density discontinuities at zero (McCrary Test) -1 -.5 0 .5 1 Notes: Sample restricted to those constituencies where there was at least one woman candidate. Discontinuity estimate in Figure B (log difference in height): .0065 ( s.e. = 0.1053) Figure 3: Characteristics of Candidate Pool (Winners and Runners-up) in Mixed-Gender Elections A. Any criminal charges filed against candidate B. Any serious criminal charges filed against candidate C. Candidate has less than high school education D. Log (net worth) E. Candidate is the incumbent F. Candidate belongs to a woman-headed party G. Candidate belongs to INC party H. Candidate belongs to BJP party Figure 4: Demographic Characteristics and Prior Election Outcomes A: Population fraction of women B: Literacy rate C: Population share of Scheduled Castes and Scheduled Tribes D: Male-female literacy differential E: Number of female candidates (major parties) in previous election F: Female voter turnout in previous election G: Total votes polled in previous election Figure 5: Regression Discontinuity Estimates for Women's Political Candidacy 0 .2 .4 .6 .8 A: At least one woman candidate from a major party −50 0 Victory margin between female and male candidate Any female candidate (major party) Fitted values (female) 50 Fitted values (male) Victory Margin Aggregated into 2% Bins 0 .1 .2 .3 .4 .5 B: Female share of major party candidates −50 0 Victory margin between female and male candidate Female share of major party candidates Fitted values (female) Victory Margin Aggregated into 2% Bins 50 Fitted values (male) Table 1 Women's Electoral Success and Future Political Candidacy: Regression Discontinuity Estimates At least one Female share of female major major party party candidates candidate # of female major party candidates # of major party candidates 3 4 5 6 7 0.015 [0.078] 0.060 *** [0.017] 0.013 [0.012] -0.002 [0.034] 5881 5881 5504 5881 Panel B: Local linear regressions with Imbens-Kalyanaraman optimal bandwidth Woman won 0.172*** 0.091 *** 0.176*** 0.033 previous election [0.051] [0.021] [0.061] [0.090] Optimal bw (IK) 0.1658 0.24421 0.1775 0.2092 Observations 1350 1877 1423 1638 0.052** [0.021] 0.1654 1360 0.008 [0.013] 0.2102 1521 -0.027 [0.038] 0.2766 2128 Panel C: Local linear regressions with Calonico-Cattaneo-Titiunik optimal bandwidth Woman won 0.174*** 0.088 *** 0.178 *** 0.025 previous election [0.050] [0.024] [0.061] [0.109] Optimal bw (CCT) 0.1718 0.1774 0.1805 0.1509 Observations 1388 1419 1446 1253 0.055** [0.022] 0.1679 1348 0.01 [0.016] 0.1757 1305 -0.114** [0.055] 0.1222 1044 1 2 Panel A: Controlling for quadratic polynomials in the running variable Woman won 0.185 *** 0.085 *** 0.166 *** previous election [0.038] [0.019] [0.046] R-squared Observations 5881 5881 5881 Female share Female share of of competitive independent candidates candidates Woman wins election Standard errors in brackets, clustered at the district level. ***, **, * indicate significance at 1%, 5% and 10% level respectively. Sample includes all races with any female candidates as in Meyersson (2014). Optimal bandwidths are determined by the algorithms suggested in Imbens and Kalyanaraman 2011 (IK) and Calonico, Cattaneo and Titiunik 2014 (CCT). Table 2 Impact of Women's Electoral Success on New vs Incumbent Candidates New female share New male share of of major party major party candidates candidates 1 2 Panel A: Controlling for quadratic polynomials in the running variable Woman won previous election -0.023 ** -0.025 [0.010] [0.024] Observations 5874 5874 Mean of dependent variable 0.04735 0.5338 Incumbent runs for reelection Incumbent runs for re- Incumbent runs for election from the re-election from a same party different party 3 4 5 0.065 * [0.038] 5881 0.6226 0.149*** [0.038] 5881 0.5092 -0.084*** [0.024] 5881 0.1134 Panel B: Local linear regressions with Imbens-Kalyanaraman optimal bandwidth Woman won previous election -0.024** -0.021 0.071 [0.011] [0.025] [0.043] Optimal bw (IK) 0.2731 0.2566 0.2319 Observations 2105 1976 1795 0.159*** [0.044] 0.2352 1818 -0.091 *** [0.022] 0.3489 2889 Panel C: Local linear regressions with Calonico-Cattaneo-Titiunik optimal bandwidth Woman won previous election -0.01 -0.008 0.036 [0.017] [0.032] [0.054] Optimal bw (CCT) 0.1247 0.1647 0.1475 Observations 1059 1344 1229 0.162*** [0.053] 0.1511 1254 -0.094 *** [0.030] 0.2112 1648 Standard errors in brackets, clustered at the district level. ***, **, * indicate significance at 1%, 5% and 10% level respectively. New candidates are defined as candidates who did not run in the previous election. They were identified by manual matching of candidate names across election years. Sample includes all races with any female candidates as in Meyersson (2014). Optimal bandwidths are determined by the algorithms suggested in Imbens and Kalyanaraman 2011 (IK) and Calonico, Cattaneo and Titiunik 2014 (CCT). Table 3 Candidacy Effects Across Parties and Across Geographies Female share of major party candidates Incumbent party Non-incumbent in nearby constituencies (other has a female parties have a constituencies in same district) candidate female candidate All major parties Incumbent party (dummy) (dummy) 1 2 3 4 Panel A: Controlling for quadratic polynomials in the running variable Woman won previous election 0.535*** -0.142*** 0.001 -0.002 [0.031] [0.043] [0.005] [0.008] Observations 5881 5881 5867 5874 Mean of dependent variable Panel B: Local linear regressions with Imbens-Kalyanaraman optimal bandwidth Woman won previous election 0.541 *** -0.161*** -0.006 -0.004 [0.031] [0.051] [0.005] [0.008] Optimal bw (IK) 0.2725 0.2063 0.2666 0.2737 Observations 2104 1617 2064 2113 Panel C: Local linear regressions with Calonico-Cattaneo-Titiunik optimal bandwidth Woman won previous election 0.536 *** -0.178 *** -0.007 [0.044] [0.053] [0.007] Optimal bw (CCT) 0.133 0.1885 0.1195 Observations 1116 1493 1018 -0.011 [0.011] 0.1371 1146 Female share of major party candidates in nearby constituencies (other constituencies in same parliamentary constituency) All major parties Incumbent party 5 6 0.003 [0.005] 5874 -0.007 [0.008] 5881 -0.003 [0.005] 0.2807 2166 -0.011 [0.009] 0.229 1773 -0.007 [0.007] 0.1217 1041 -0.020 [0.013] 0.1199 1025 Standard errors in brackets, clustered at the district level. ***, **, * indicate significance at 1%, 5% and 10% level respectively. Sample includes all races with any female candidates as in Meyersson (2014). Optimal bandwidths are determined by the algorithms suggested in Imbens and Kalyanaraman 2011 (IK) and Calonico, Cattaneo and Titiunik 2014 (CCT). Table 4 Women's Electoral Success and Political Candidacy: Heterogeneity by State Level Gender Prejudice Female share of major party candidates Female population share Low High 1 2 Panel A: Controlling for quadratic polynomials in the running variable Woman won previous election 0.052** 0.139*** [0.023] [0.031] Observations 3714 2160 New female share of major party candidates Female population share Low High 3 4 -0.032** [0.012] 3714 -0.002 [0.020] 2160 Non-new (incumbent + losers) female share of major party candidates Female population share Low High 5 6 0.085*** [0.020] 3714 0.141*** [0.029] 2160 0.005 [0.019] 867 0.3043 0.069*** [0.025] 887 0.1885 0.160*** [0.031] 825 0.2877 0.022 [0.029] 480 0.141 0.054* [0.030] 648 0.1291 0.053 [0.049] 342 0.0973 Panel B: Local linear regressions with Imbens-Kalyanaraman optimal bandwidth Woman won previous election Observations Optimal bw (IK) 0.041 [0.028] 929 0.1982 0.176*** [0.028] 1039 0.3643 -0.031** [0.014] 1195 0.254 Panel C: Local linear regressions with Calonico-Cattaneo-Titiunik optimal bandwidth Woman won previous election Observations Optimal bw (CCT) 0.030 [0.031] 781 0.1625 0.121** [0.049] 347 0.0992 -0.035* [0.019] 648 0.129 Standard errors in brackets, clustered at the district level. ***, **, * indicate significance at 1%, 5% and 10% level respectively. Sample includes all races with any female candidates as in Meyersson (2014). Optimal bandwidths are determined by the algorithms suggested in Imbens and Kalyanaraman 2011 (IK) and Calonico, Cattaneo and Titiunik 2014 (CCT). Low and high female population share refer to states with female population share above or below the national average. Table 5 Women's Electoral Success and Voter Behavior Female population share Low High 1 2 Female population share Low High 3 4 Panel A: Voter turnout in subsequent election Female Voter Turnout Woman won previous election -0.009 0.055*** [0.013] [0.019] Observations 3713 2158 Male Voter Turnout 0.003 0.052*** [0.011] [0.014] 3713 2158 Panel B: Preferences for women candidates Whether a woman wins the next election Woman won previous election -0.033 0.046 [0.044] [0.054] Observations 3714 2167 Vote share of all women candidates 0.028 0.073** [0.021] [0.032] 3714 2167 Panel C: Female share of major party candidates in later elections Second lag Woman won previous election 0.020 0.072* [0.025] [0.039] Observations 2939 1649 -0.002 [0.024] 2171 Third lag -0.019 [0.051] 1120 Standard errors in brackets, clustered at the district level. ***, **, * indicate significance at 1%, 5% and 10% level respectively. Sample includes all races with any female candidates as in Meyersson (2014). Regressions control for quadratic polynomials in the running variable. See Appendix Table A4 for robustness to alternative bandwidth selection methods. Table 6 Women's Electoral Success and Future Candidacy: Heterogeneity by Party Gender Female population share Low High 1 2 Female population share Low High 3 4 Panel A: Female share of major party candidates Major parties headed by women Woman won previous election 0.002 0.008 [0.012] [0.023] Observations 3714 2160 Major parties headed by men 0.050** 0.132*** [0.022] [0.028] 3714 2160 Panel B: New female share of major party candidates Major parties headed by women Major parties headed by men Woman won previous election -0.015** -0.012 -0.018* 0.010 Observations [0.006] 3714 [0.009] 2160 [0.009] 3714 [0.019] 2160 Standard errors in brackets, clustered at the district level. ***, **, * indicate significance at 1%, 5% and 10% level respectively. Sample includes all races with any female candidates as in Meyersson (2014). Regressions control for quadratic polynomials in the running variable. Table 7 Women's Electoral Success and Political Candidacy Before and After Local Government Quotas Local government quotas No Yes 1 2 Local government quotas No Yes 3 4 Female population share is low 0.019 0.094*** [0.030] [0.035] 2220 1494 Female population share is high 0.224*** 0.085** [0.054] [0.041] 855 1305 Panel B: New female share of major party candidates Female population share is low Woman won previous election -0.029** -0.038* [0.015] [0.021] Observations 2220 1494 Female population share is high -0.014 -0.001 [0.032] [0.025] 855 1305 Panel A: Female share of major party candidates Woman won previous election Observations Standard errors in brackets, clustered at the district level. ***, **, * indicate significance at 1%, 5% and 10% level respectively. Sample includes all races with any female candidates as in Meyersson (2014). Regressions control for quadratic polynomials in the running variable. Table 8 Muslims' Electoral Success and Future Political Candidacy: Regression Discontinuity Estimates 1 2 Panel A: Muslim share of major party candidates Muslim won previous election Observations 3 4 Bandwidth Party controls Hindu-Muslim literacy differential High Low 7 8 # incidents of religious violence High Low 9 10 0.028 [0.020] 0.037 * [0.020] 0.094 *** [0.024] 0.050 [0.031] 0.015 [0.053] 0.045 ** [0.021] 0.033 [0.023] 0.067 * [0.037] 0.02 [0.021] 0.133 *** [0.045] 7636 7636 2447 1392 1787 5849 4820 2816 5718 1918 -0.005 [0.026] -0.085 ** [0.041] -0.01 [0.020] -0.016 [0.022] -0.04 [0.032] -0.009 [0.018] -0.014 [0.042] Panel B: New Muslim candidate share in major parties Muslim won previous -0.028 -0.023 -0.009 election [0.019] [0.019] [0.017] Observations Polynomial Muslim population share Low High 5 6 7636 7636 2nd order 2nd order yes 3430 1444 optimal (IK) yes optimal (CCT) yes 1787 5849 2nd order 2nd order yes yes 4820 2816 2nd order 2nd order yes yes 5718 1918 2nd order 2nd order yes yes Standard errors in brackets, clustered at the district level. ***, **, * indicate significance at 1%, 5% and 10% level respectively. Sample includes all races with any Muslim candidates. Optimal bandwidths are determined by the algorithms suggested in Imbens and Kalyanaraman 2011 (IK) and Calonico, Cattaneo and Titiunik 2014 (CCT). In each case, the first column in a pair indicates a stronger potential Muslim disadvantage. Muslim population share and Hindu-Muslim literacy differential data are from 2001 state-level census for 16 major states. Classification for population share is bottom 8 states (low population share) versus top 8. Classification for literacy differential is less than 1 percentage point (low differential) versus others. Incidents of religious violence are measured over 1980-2010. Appendix Figures and Tables Figure A1: Women's Political Candidacy in Major Indian States, 1980-2007 2005-2007 2000-2004 1995-19994 1990-1994 1985-1989 1980-1984 .02 .04 .06 .08 .1 A. Women's Political Candidacy over Time 5-year election period Fraction of candidates that are female Fraction of candidates that are female (major parties) Fraction of winners that are female B. Female Share of Major Party Candidates in Major Indian States 0.120 0.100 0.080 no need to update 0.060 0.040 0.020 0.000 1980-84 2000-07 0 1 2 t stat 3 4 5 Figure A2: Testing for Discontinuities at "Fake" Discontinuity Points -.6 -.4 -.2 margin 0 .2 Note: Graph plotes the t-statistics of different placebo tests conducted at alternative placebo discontinuities between percentiles 5 and 95 of the distribution. The t-statistic of the discontinuity at zero is the one of our main estimates. Specifications restrict the sample to observations either at the left or at the right of the discontinuity and include a split second order polynomial aproximation. Figure A3: Women's Electoral Success and the Entry of New Female Candidates 0 .05 .1 .15 A. States with lower female population share −50 0 Victory margin between female and male candidate New female share of major party candidates Fitted values (female) 50 Fitted values (male) Victory Margin Aggregated into 2.5% Bins 0 .05 .1 .15 B. States with higher female population share −50 0 Victory margin between female and male candidate New female share of major party candidates Fitted values (female) Victory Margin Aggregated into 2,5% Bins 50 Fitted values (male) Table A1 Descriptive Statistics Whole Sample Female share of major party candidates New female share of major party candidates Female share of competitive candidates Female share of independent candidates Female voter turnout Male voter turnout Woman won previous election (dummy) Regression Discontinuity Sample Female share of major party candidates New female share of major party candidates Female share of competitive candidates Female share of independent candidates Female voter turnout Male voter turnout Woman won previous election (dummy) Obs Mean Std. Dev Min Max 22420 22420 22478 20649 22421 22415 22296 0.056 0.038 0.050 0.037 0.587 0.664 0.048 0.156 0.129 0.134 0.116 0.154 0.124 0.214 0 0 0 0 0 0.004 0 1 1 1 1 1 1 1 Obs Mean Std. Dev Min Max 5874 5874 5881 5504 5871 5871 5881 0.114 0.047 0.103 0.055 0.575 0.652 0.182 0.209 0.142 0.182 0.140 0.146 0.123 0.386 0 0 0 0 0.009 0.011 0 1 1 1 1 0.975 0.989 1 Note: the regression discontinuity sample includes all constituencies with at least one female candidate. Table A2 Women's Electoral Success and Future Political Participation: OLS Estimates Woman won previous election R-squared Observations Any female major party candidate 1 0.244** [0.018] Female share of major party candidates 2 0.110 *** [0.009] 22296 0.38 22238 New female Female share of share of major competitive party candidates candidates 3 4 0.097 *** -0.083 *** [0.008] [0.007] 0.40 22296 0.22 22238 Woman wins election Vote share received by women candidates 5 0.010 [0.016] 6 0.084 *** [0.009] 0.33 22296 0.42 22296 Standard errors in brackets, clustered at the constituency level. ***, **, * indicate significance at 1%, 5% and 10% level respectively. All regressions control for constituency and election cycle fixed effects, and district specific trends. Table A3 Regression Discontinuity Estimates: Robustness to Changes in Functional Form, Bandwidth and Sample Dependent variable: Female share of major party candidates Woman won previous election R-squared Observations polynomial bandwidth State*year fixed effects Constituency level clustering 1 0.074*** [0.023] 2 0.080 *** [0.028] 3 0.066 ** [0.031] 4 0.100 *** [0.022] 5 0.081*** [0.019] 6 0.085 *** [0.019] Restricted sample 7 8 0.092 *** 0.060** [0.023] [0.031] 5874 3rd order 5874 4rd order 853 linear 449 none 5874 2nd order 5881 2nd order 1897 2nd order 0.1 0.05 931 linear optimal (IK) 0.121 yes yes Standard errors in brackets, clustered at the district level, except as indicated. ***, **, * indicate significance at 1%, 5% and 10% level respectively. Sample includes all races with any female candidates as in Meyersson (2014). Restricted sample consists of electoral races in which a man and a woman were the top two candidates. Optimal bandwidths are determined by the algorithms suggested in Imbens and Kalyanaraman 2011 (IK). Table A4 Heterogeneous Effects by Indicators of Gender Prejudice: Robustness to Measures Male-female literacy differential High Low 1 2 Panel A: Female share of major party candidates Woman won previous election Observations 0.070*** [0.023] 3769 Panel B: New female share of major party candidates Woman won previous election -0.030** [0.012] Observations 3769 Female population share (1981 state level) Low High 3 4 Female population share (2001 constituency level) Low High 5 6 0.113*** [0.035] 2105 0.044 [0.029] 2232 0.112*** [0.025] 3642 0.066** [0.027] 2765 0.100*** [0.028] 3109 -0.005 [0.020] 2105 -0.051*** [0.015] 2232 -0.006 [0.014] 3642 -0.043*** [0.015] 2765 -0.007 [0.015] 3109 Standard errors in brackets, clustered at the district level. ***, **, * indicate significance at 1%, 5% and 10% level respectively. Sample includes all races with any female candidates as in Meyersson (2014). Optimal bandwidths are determined by the algorithms suggested in Imbens and Kalyanaraman 2011 (IK) and Calonico, Cattaneo and Titiunik 2014 (CCT). Male-female literacy differential is based on 2001 state-level data. Low and High female population share refer to states with female population share above or below the nationwide average. These measures are alternatives to the 2001 state-level female population share used in Table 4 as a proxy for gender bias.