Education and the Intergenerational Transmission of Violence (Figures)

0.008 0.006 0.000 0.002 0.004 Density 0.010 0.012 0.014 Figure 1: McCrary Density Test −80 −40 0 40 80 Born After January 1987 (In Months) Note: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. The graph shows the results of the McCrary test of whether there is a discontinuity in the density of the forcing variable, the month of birth. Figure 2: Balanced Covariates -40 0 40 80 .8 -80 -40 0 40 80 -80 -40 0 40 Born after January 1987 (in months) Born after January 1987 (in months) (d) Childhood Region: North (e) Childhood Region: East (f) Non-Turkish Speaker 1 .3 .8 0 .2 0 0 .2 .1 .4 .4 .2 .6 .8 .6 80 .4 Born after January 1987 (in months) 1 -80 0 .2 .4 .6 .8 .6 .4 .2 0 0 .2 .4 .6 .8 1 (c) Childhood Region: Central 1 (b) Childhood Region: South 1 (a) Childhood Region: West -80 -40 0 40 Born after January 1987 (in months) 80 -80 -40 0 40 Born after January 1987 (in months) 80 -80 -40 0 40 80 Born after January 1987 (in months) Note: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. The figures plot predetermined covariates in monthly bins against the month-year of birth of being born in January 1987. The vertical line in each graph represents the cut-off point, January 1987. Gray lines show 95 percent confidence intervals around the mean level. Variable definitions are listed in Online Appendix A. Figure 3: RD Treatment Effects on Junior High School Completion .9 .8 .7 .6 .5 .4 .3 .3 .4 .5 .6 .7 .8 .9 1 Men with Completed Junior High School 1 Women with Completed Junior High School -60 -30 0 30 Born after January 1987 (in months) 60 -60 -30 0 30 Born after January 1987 (in months) 60 Note: Data are from the 2014 Household Labor Force Survey. Figures plot junior high school completion rates in monthly bins for women on the left and men on the right. Gray lines show 95 percent confidence intervals around the mean level. Figure 4: Treatment and Placebo .9 .8 .7 .6 .5 .4 .3 .2 .1 0 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 2008 TNSDVW Survey 1 2014 TNSDVW Survey -60 -30 0 30 Born after January 1987 (in months) 60 -60 -30 0 30 Born after January 1981 (in months) 60 Note: Data are from the 2014 and 2008 National Surveys on Domestic Violence against Women in Turkey, respectively. The figures plot a dummy variable equal to one of the respondent completed junior high school in monthly bins. Gray lines show 95 percent confidence intervals around the mean level. Table 1: Summary Statistics for 20- to 34-Year-Old Women Who Have Children Region of Childhood Difference All (1) Mean (S.D.) Rural Urban (2) (3) Mean Mean (S.D.) (S.D.) (2) – (3) (4) Est. (S.E.) 7.52 (3.86) 0.51 (0.50) 0.31 (0.46) 0.89 (0.31) 6.73 (3.72) 0.42 (0.49) 0.22 (0.41) 0.87 (0.33) 8.55 (3.78) 0.63 (0.48) 0.42 (0.49) 0.92 (0.27) -1.82*** (0.21) -0.21*** (0.03) -0.20*** (0.03) -0.05*** (0.02) 1,807/1,101/686 0.48 (0.50) 0.41 (0.49) 0.51 (0.50) 0.44 (0.50) 0.45 (0.50) 0.38 (0.48) 0.06** (0.03) 0.06** (0.03) 1,800/1,096/683 0.38 (0.48) 0.29 (0.45) 0.41 (0.49) 0.30 (0.46) 0.32 (0.47) 0.27 (0.45) 0.09*** (0.03) 0.03 (0.03) 1,712/1,040/651 0.02 (0.52) 0.00 (0.65) 0.02 (0.52) 0.04 (0.52) 0.02 (0.65) 0.04 (0.52) -0.01 (0.53) -0.03 (0.65) -0.01 (0.53) 0.05* (0.03) 0.06 (0.04) 0.05 (0.03) 1,808/1,101/686 21.34 (3.68) 1.50 (1.18) 21.08 (3.55) 1.70 (1.21) 21.60 (3.77) 1.27 (1.08) -0.52*** (0.20) 0.43*** (0.05) 1,905/1,143/737 (5) Observations (All/Rural/Urban) Panel A: Education Years of schooling Completed junior high school Completed high school Completed primary school 1,808/1,101/686 1,808/1,101/686 1,808/1,101/686 Panel B: Violence against children Child abuse Frequent child abuse 1,800/1,096/683 Panel C: Attitudes against violence Men can beat their partners in certain situations. It may be necessary to beat children for discipline. 1,801/1,097/683 Panel D: Mental health outcomes Overall depression index Somatic depression index Nonsomatic depression index 1,808/1,101/686 1,808/1,101/686 Panel E: Fertility outcomes Age at first pregnancy † Number of children † 2,425/1,387/1,006 Table 1: Summary Statistics for 20- to 34-Year-Old Women Who Have Children, Cont’d Region of Childhood Difference All (1) Mean (S.D.) Rural Urban (2) (3) Mean Mean (S.D.) (S.D.) (2) – (3) (4) Est. (S.E.) 0.19 (0.39) 0.14 (0.34) 0.11 (0.31) -0.08 (0.45) 0.08 (0.35) 0.18 (0.38) 0.11 (0.31) 0.09 (0.29) -0.10 (0.44) 0.03 (0.34) 0.21 (0.41) 0.18 (0.38) 0.14 (0.35) -0.04 (0.47) 0.17 (0.34) -0.03 (0.02) -0.07*** (0.02) -0.05*** (0.02) -0.05** (0.03) -0.14*** (0.02) 8.81 (3.61) 24.81 (4.25) 0.01 (0.64) 21.28 (3.37) 0.57 (0.49) 0.06 (0.23) 8.43 (3.55) 24.60 (4.31) 0.06 (0.53) 21.09 (3.36) 0.52 (0.50) 0.05 (0.21) 9.31 (3.62) 25.04 (4.11) -0.07 (0.78) 21.48 (3.37) 0.64 (0.48) 0.06 (0.25) -0.88*** (0.21) -0.44* (0.24) 0.12** (0.04) -0.39** (0.19) -0.12*** (0.03) -0.02 (0.01) 0.00 (0.79) 0.05 (0.55) -0.04 (0.80) 0.00 (0.79) 0.05 (0.54) -0.05 (0.78) 0.00 (0.80) 0.04 (0.58) -0.05 (0.81) 0.00 (0.05) 0.01 (0.03) 0.00 (0.04) 0.59 (0.49) 0.01 (0.09) 0.14 (0.35) 1.00 (0.00) 0.01 (0.09) 0.14 (0.34) 0.00 (0.00) 0.00 (0.03) 0.14 (0.35) 1.00*** (0.00) 0.01*** (0.00) -0.01 (0.02) (5) Observations (All/Rural/Urban) Panel F: Labor market outcomes Employed Employed in services Social security Personal income index Asset ownership index 1,808/1,101/686 1,808/1,101/686 1,808/1,101/686 1,808/1,101/686 1,808/1,101/686 Panel G: Marriage market outcomes Partner’s years of schooling Partner’s age Partner’s religiosity index Marriage age Marriage decision Divorced 1,792/1,088/684 1,805/1,099/685 1,808/1,101/686 1,805/1,099/685 1,808/1,101/686 1,808/1,101/686 Panel H: Spousal violence outcomes Physical violence index Psychological violence index Financial control index 1,808/1,101/686 1,808/1,101/686 1,801/1,096/684 Panel I: Covariates Rural childhood region Non-Turkish speaker Childhood violence 1,787/1,101/686 1,808/1,101/686 1,742/1,057/664 Notes: The table presents the means, standard deviations, and number of observations from the 2014 National Survey on Domestic Violence against Women in Turkey. The sample includes women who have children and who were born within 85 months before or after January 1987. † denotes that age at first pregnancy is reported for women who have ever been pregnant, and number of children is reported for all women, born within 85 months around the discontinuity. Columns 1 - 3 report means and standard deviations in parentheses. Column 4 reports differences in the group means between columns 2 and 3 with standard errors in parentheses. ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively. The sum of rural and urban samples is less than the full sample due to missing observations in the region of childhood variable. The variables are described in Appendix A. Table 2: RD Treatment Effects on Schooling Outcomes Years of schooling Completed education: Junior high school High school Primary school (1) (2) (3) (4) (5) (6) Linear RD ĥ bandwidth Linear RD 0.75ĥ bandwidth Linear RD 1.5ĥ bandwidth Bandwidth N Mean 0.704** (0.303)†† 0.769** (0.349)† 1.031*** (0.249)††† 89 2,492 8.48 0.192*** (0.031)††† 0.125*** (0.044)†† -0.020 (0.024) 0.186*** (0.037)††† 0.081* (0.048) -0.031 (0.028) 0.186*** (0.027)††† 0.078** (0.038)† -0.020 (0.020) 118 3,308 0.60 65 1,837 0.40 93 2,630 0.91 Notes: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. Columns 1 – 3 report local RD regressions with linear polynomials in the month-year of birth using the optimal bandwidth ĥ, 0.75 ĥ and 1.5 ĥ, respectively. The optimal bandwidth, reported in column 4, is estimated by using the Imbens and Kalyanaraman (2009) algorithm. Column 5 reports the number of observations used in estimations with the optimal bandwidth ĥ, and column 6 reports the outcome mean within the optimal bandwidth ĥ. All results are reported for the full sample of women. The variables are described in Appendix A. All specifications control for a dummy variable for whether the respondent grew up in a rural location, a dummy variable for whether the respondent’s mother tongue is not Turkish, month-of-birth fixed effects, region fixed effects, and interactions of region fixed effects with an indicator of rural regions. Standard errors are clustered at the month-year cohort level. ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively (based on p-values unadjusted for multiple-hypothesis testing). †††, ††, and † denote significance at the 1, 5, and 10 percent levels, respectively (based on p-values adjusted for multiple-hypothesis testing using Simes adjustment). Table 3: RD Treatment Effects on Schooling by Region of Childhood Rural childhood region Bandwidth: (1) ĥ (2) 0.75 ĥ 1.134** (0.451) 7.42 85 1,396 1.080** (0.513) 7.40 64 1,052 (3) 1.5 ĥ Urban childhood region (4) ĥ (5) 0.75 ĥ (6) 1.5 ĥ 1.283*** 0.413 (0.364) (0.454) 7.47 9.68 128 98 2,038 1,147 0.560 (0.474) 9.73 74 866 0.763* (0.391) 9.51 147 1,710 1.103** (0.460) 6.68 120 1,455 -0.136 (0.552) 8.32 54 458 -0.395 (0.481) 8.55 143 825 Panel A: Sample of All Women Years of schooling Mean Bandwidth Observations Panel B: Sample of Women Who Have Children Years of schooling Mean Bandwidth Observations 1.115** (0.539) 6.81 80 1,032 1.039* (0.605) 6.80 60 779 -0.141 (0.514) 8.52 72 596 Notes: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. Columns 1 – 3, and 4 – 6 report local RD regressions with linear polynomials in the month-year of birth using the optimal bandwidth ĥ, 0.75 ĥ, and 1.5 ĥ, respectively. The outcome mean, optimal bandwidth estimated by the Imbens and Kalyanaraman (2009) algorithm, and observation numbers are reported in the rows under the dependent variables. Columns 1 – 3 report the results for the sample of women who grew up in a rural region, and columns 4 – 6 report them for the sample of women who grew up in an urban region. Panel A reports the results for the sample of all women, and Panel B reports them for the sample of women who have children. The variables are described in Appendix A. All specifications control for a dummy variable for whether the respondent grew up in a rural location, a dummy variable for whether the respondent’s mother tongue is not Turkish, month-of-birth fixed effects, region fixed effects, and interactions of region fixed effects with an indicator of rural regions. Standard errors are clustered at the month-year cohort level. ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively. Table 4: Effects of Education on Violence Against Children Overall sample Rural sample (1) OLS (2) RF (3) IV (4) OLS (5) RF (6) IV -0.021*** (0.003)††† 0.49 94 1,932 0.007 (0.046) 0.49 94 1,932 0.033 (0.226) 0.49 94 1,932 -0.024*** (0.005)††† 0.51 89 1,140 0.024 (0.067) 0.51 89 1,140 0.025 (0.068) 0.51 89 1,140 -0.020*** (0.003)††† 0.42 106 2,131 0.056 (0.043) 0.42 106 2,131 0.196 (0.269) 0.42 106 2,131 -0.021*** (0.005)††† 0.44 92 1,164 0.053 (0.072) 0.44 92 1,164 0.058 (0.085) 0.44 92 1,164 Panel A: RD Treatment Effects Child abuse Schooling Mean Bandwidth Observations Frequent child abuse Schooling Mean Bandwidth Observations Panel B: RD Treatment Effects by Exposure to Childhood Violence Child abuse Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations Frequent child abuse Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations -0.019*** (0.004)††† -0.000 (0.011) 0.198** (0.086)†† 0.49 94 1,864 0.032 (0.047) -0.060 (0.084) 0.224*** (0.049)††† 0.49 94 1,864 0.746 -0.023*** (5.790) (0.005)††† -0.623 0.002 (4.547) (0.018) 4.979 0.238* (34.975) (0.123) 0.49 0.51 94 89 1,864 1,096 0.083 (0.068) -0.224** (0.103)†† 0.341*** (0.056)††† 0.51 89 1,096 0.097 (0.096) -0.148** (0.074)†† 1.308** (0.531)†† 0.51 89 1,096 -0.018*** (0.003)††† -0.011 (0.011) 0.257*** (0.095)†† 0.42 106 2,055 0.080* (0.044) -0.060 (0.081) 0.204*** (0.046)††† 0.42 106 2,055 1.065 -0.020*** (4.903) (0.005)††† -0.957 -0.001 (4.309) (0.018) 7.475 0.250* (32.864) (0.134) 0.42 0.44 106 92 2,055 1,119 0.108 (0.073) -0.268** (0.103)†† 0.344*** (0.062)††† 0.44 92 1,119 0.134 (0.122) -0.185** (0.084)†† 1.557** (0.613)†† 0.44 92 1,119 Notes: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. The sample includes women who have children. The optimal bandwidth is estimated by using the Imbens and Kalyanaraman (2009) algorithm. Panel A reports the RD treatment effects of the reform, and Panel B reports them by exposure to childhood violence, i.e., whether the respondent experienced violence from her own family members during her childhood. Columns 1 reports OLS results using years of schooling as the independent variable for an optimal bandwidth ĥ estimated by the Imbens and Kalyanaraman algorithm. Columns 2 – 3 report reduced-form RD treatment effects and two-stage least-squares RD treatment effects (by using treatment as an instrument for years of schooling) of being born after January 1987 with a linear control function in the month-year of birth on each side of the discontinuity. Columns 1 – 3 report these results for the overall sample, and columns 4 – 6 report results from same specifications for the subsample of respondents whose childhood region of residence was rural. The variables are described in Appendix A. All specifications control for a dummy variable for whether the respondent grew up in a rural location, a dummy variable for whether the respondent’s mother tongue is not Turkish, month-of-birth fixed effects, region fixed effects, and interactions of region fixed effects with an indicator of rural regions. Standard errors are clustered at the month-year cohort level. *** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level. (Based on p-values unadjusted for multiple-hypothesis testing.) ††† Significant at the 1 percent level. †† Significant at the 5 percent level. † Significant at the 10 percent level. (Based on p-values adjusted for multiple-hypothesis testing using Simes adjustment.) Table 5: Effects of Education on Attitudes toward Violence Overall sample Men can beat their partners Schooling in certain situations. Schooling × Childhood violence Childhood violence Mean Bandwidth Observations It may be necessary to beat children for discipline. Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations Rural sample (1) OLS (2) RF (3) IV (4) OLS (5) RF (6) IV -0.024*** (0.004)††† -0.003 (0.011) 0.114 (0.091) 0.38 83 1,589 0.046 (0.055) -0.066 (0.092) 0.113** (0.048)†† 0.38 83 1,589 0.154 (0.233) -0.162 (0.206) 1.323 (1.561) 0.38 83 1,589 -0.023*** (0.005)††† -0.010 (0.013) 0.169 (0.103) 0.41 88 1,039 0.053 0.052 (0.068) (0.070) -0.067 -0.056 (0.110) (0.065) 0.120** 0.490 (0.056)† (0.446) 0.41 0.41 88 88 1,039 1,039 -0.018*** (0.003)††† -0.005 (0.010) 0.091 (0.088) 0.28 107 2,056 0.035 (0.042) 0.017 (0.065) 0.058 (0.039) 0.28 107 2,056 0.357 (1.522) -0.297 (1.388) 2.329 (10.558) 0.28 107 2,056 -0.024*** (0.004)††† -0.008 (0.013) 0.119 (0.104) 0.29 99 1,189 0.025 (0.049) 0.009 (0.084) 0.075 (0.052) 0.29 99 1,189 0.027 (0.055) -0.009 (0.048) 0.147 (0.344) 0.29 99 1,189 Notes: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. The sample includes women who have children. The optimal bandwidth is estimated by using the Imbens and Kalyanaraman (2009) algorithm. The RD treatment effects of the reform are reported by exposure to childhood violence, i.e., whether the respondent experienced violence from her own family members during her childhood. Columns 1 reports OLS results using years of schooling as the independent variable for an optimal bandwidth ĥ estimated by the Imbens and Kalyanaraman algorithm. Columns 2 – 3 report reduced-form RD treatment effects and two-stage least-squares RD treatment effects (by using treatment as an instrument for years of schooling) of being born after January 1987 with a linear control function in the month-year of birth on each side of the discontinuity. Columns 1 – 3 report these results for the overall sample, and columns 4 – 6 report results from the same specifications for the subsample of respondents whose childhood region of residence was rural. The variables are described in Appendix A. All specifications control for a dummy variable for whether the respondent grew up in a rural location, a dummy variable for whether the respondent’s mother tongue is not Turkish, month-of-birth fixed effects, region fixed effects, and interactions of region fixed effects with an indicator of rural regions. Standard errors are clustered at the month-year cohort level. *** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level. (Based on p-values unadjusted for multiple-hypothesis testing.) ††† Significant at the 1 percent level. †† Significant at the 5 percent level. † Significant at the 10 percent level. (Based on p-values adjusted for multiple-hypothesis testing using Simes adjustment.) 44 Table 6: Effects of Education on Mental Health Outcomes Overall sample (1) OLS Overall depression index Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations Somatic depression index Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations Nonsomatic depression index Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations (2) RF Rural sample (3) IV (4) OLS (5) RF (6) IV -0.023*** (0.005)††† 0.005 (0.012) 0.284*** (0.100)††† 0.02 77 1,584 0.001 0.114 -0.022*** (0.050) (0.277) (0.005)††† -0.126 -0.203 0.004 (0.096) (0.270) (0.015) 0.364*** 1.852 0.257** (0.044)††† (2.055) (0.111)†† 0.02 0.02 0.03 77 77 115 1,584 1,584 1,360 0.017 (0.067) -0.286** (0.115)†† 0.368*** (0.056)††† 0.03 115 1,360 0.039 (0.088) -0.169** (0.081)† 1.449** (0.575)†† 0.03 115 1,360 -0.037*** (0.006)††† 0.029** (0.014) 0.062 (0.106) 0.00 106 2,045 0.025 (0.059) -0.249** (0.111)† 0.350*** (0.054)††† 0.00 106 2,045 0.054 (0.075) -0.340** (0.143)†† 0.333*** (0.081)††† 0.03 96 1,167 0.071 (0.112) -0.192* (0.103)† 1.543** (0.736)†† 0.03 96 1,167 -0.022*** (0.005)††† -0.001 (0.013) 0.342*** (0.112)††† 0.03 75 1,518 0.002 0.077 -0.019*** 0.040 (0.055) (0.228) (0.005)††† (0.077) -0.098 -0.147 -0.011 -0.290** (0.102) (0.245) (0.017) (0.127)†† 0.370*** 1.443 0.378*** 0.407*** (0.049)††† (1.857) (0.120)††† (0.063)††† 0.03 0.03 0.04 0.04 75 75 95 95 1,518 1,518 1,156 1,156 0.054 (0.106) -0.160** (0.081)† 1.414** (0.575)†† 0.04 95 1,156 0.410 -0.033*** (1.264) (0.007)††† -0.583 0.013 (1.185) (0.025) 4.657 0.122 (9.023) (0.170) 0.00 0.03 106 96 2,045 1,167 Notes: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. The sample includes women who have children. The optimal bandwidth is estimated by using the Imbens and Kalyanaraman (2009) algorithm. The RD treatment effects of the reform are reported by exposure to childhood violence, i.e., whether the respondent experienced violence from her own family members during her childhood. Columns 1 reports OLS results using years of schooling as the independent variable for an optimal bandwidth ĥ estimated by the Imbens and Kalyanaraman algorithm. Columns 2 – 3 report reduced-form RD treatment effects and two-stage least-squares RD treatment effects (by using treatment as an instrument for years of schooling) of being born after January 1987 with a linear control function in the month-year of birth on each side of the discontinuity. Columns 1 – 3 report these results for the overall sample, and columns 4 – 6 report results from the same specifications for the subsample of respondents whose childhood region of residence was rural. The variables are described in Appendix A. All specifications control for a dummy variable for whether the respondent grew up in a rural location, a dummy variable for whether the respondent’s mother tongue is not Turkish, month-of-birth fixed effects, region fixed effects, and interactions of region fixed effects with an indicator of rural regions. Standard errors are clustered at the month-year cohort level. *** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level. (Based on p-values unadjusted for multiple-hypothesis testing.) ††† Significant at the 1 percent level. †† Significant at the 5 percent level. † Significant at the 10 percent level. (Based on p-values adjusted for multiple-hypothesis testing using Simes adjustment.) Table 7: Effects of Education on Fertility Outcomes Overall sample (1) OLS Age at first pregnancy Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations Number of children Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations (2) RF Rural sample (3) IV (4) OLS (5) RF (6) IV 0.442*** (0.027)††† -0.096 (0.078) 0.514 (0.601) 21.48 121 2,385 0.147 0.016 (0.259) (1.519) 0.823* 0.920 (0.448) (1.862) -0.407 -7.147 (0.325) (14.394) 21.48 21.48 121 121 2,385 2,385 0.360*** 0.650* 0.672 (0.037)††† (0.359) (0.437) -0.095 0.871 0.161 (0.083) (0.547) (0.401) 0.767 0.036 -0.778 (0.599) (0.411) (2.922) 21.14 21.14 21.14 106 106 106 1,336 1,336 1,336 -0.141*** (0.007)††† -0.023 (0.016) 0.065 (0.173) 1.51 73 1,963 -0.124 (0.102) -0.084 (0.150) -0.038 (0.108) 1.51 73 1,963 -0.146*** (0.010)††† -0.014 (0.026) -0.059 (0.242) 1.69 88 1,382 -0.149 (0.130) -0.014 (0.075) 0.023 (0.664) 1.51 73 1,963 -0.135 -0.139 (0.113) (0.120) -0.109 0.006 (0.198) (0.077) -0.045 -0.164 (0.125) (0.616) 1.69 1.69 88 88 1,382 1,382 Notes: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. The sample includes all women. The optimal bandwidth is estimated by using the Imbens and Kalyanaraman (2009) algorithm. The RD treatment effects of the reform are reported by exposure to childhood violence, i.e., whether the respondent experienced violence from her own family members during her childhood. Columns 1 reports OLS results using years of schooling as the independent variable for an optimal bandwidth ĥ estimated by the Imbens and Kalyanaraman algorithm. Columns 2 – 3 report reduced-form RD treatment effects and two-stage least-squares RD treatment effects (by using treatment as an instrument for years of schooling) of being born after January 1987 with a linear control function in the month-year of birth on each side of the discontinuity. Columns 1 – 3 report these results for the overall sample, and columns 4 – 6 report results from the same specifications for the subsample of respondents whose childhood region of residence was rural. The variables are described in Appendix A. All specifications control for a dummy variable for whether the respondent grew up in a rural location, a dummy variable for whether the respondent’s mother tongue is not Turkish, month-of-birth fixed effects, region fixed effects, and interactions of region fixed effects with an indicator of rural regions. Standard errors are clustered at the month-year cohort level. *** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level. (Based on p-values unadjusted for multiple-hypothesis testing.) ††† Significant at the 1 percent level. †† Significant at the 5 percent level. † Significant at the 10 percent level. (Based on p-values adjusted for multiple-hypothesis testing using Simes adjustment.) Table 8: Effects of Education on Labor Market Outcomes Overall sample Employed Schooling Schooling × Childhood violence Childhood violence Employed in services Mean Bandwidth Observations Schooling Schooling × Childhood violence Childhood violence Social security Mean Bandwidth Observations Schooling Schooling × Childhood violence Childhood violence Personal income index Mean Bandwidth Observations Schooling Schooling × Childhood violence Childhood violence Asset ownership index Mean Bandwidth Observations Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations (1) OLS (2) RF 0.019*** (0.004)††† -0.001 (0.009) 0.052 (0.067) 0.19 96 1,891 0.024*** (0.004)††† -0.003 (0.008) 0.045 (0.055) 0.14 94 1,871 0.024*** (0.003)††† -0.005 (0.009) 0.036 (0.057) 0.11 81 1,658 0.023*** (0.003)††† 0.003 (0.013) -0.028 (0.087) -0.05 143 2,532 0.030*** (0.002)††† -0.012** (0.006) 0.036 (0.039) 0.03 84 1,697 0.024 (0.032) -0.037 (0.064) 0.053 (0.042) 0.19 96 1,891 0.004 (0.027) -0.070 (0.052) 0.040 (0.038) 0.14 94 1,871 0.017 (0.026) -0.042 (0.043) 0.009 (0.035) 0.11 81 1,658 -0.040 (0.040) -0.061 (0.055) 0.001 (0.041) -0.05 143 2,532 -0.011 (0.021) 0.054 (0.038) -0.074*** (0.023)††† 0.03 84 1,697 Rural sample (3) IV (4) OLS 0.478 0.014*** (3.048) (0.004)††† -0.400 -0.003 (2.417) (0.014) 3.102 0.087 (18.546) (0.100) 0.19 0.18 96 93 1,891 1,139 0.114 0.021*** (0.592) (0.004)††† -0.146 -0.006 (0.479) (0.014) 1.117 0.063 (3.682) (0.093) 0.14 0.11 94 88 1,871 1,097 0.113 0.021*** (0.235) (0.004)††† -0.119 -0.007 (0.195) (0.012) 0.898 0.054 (1.507) (0.072) 0.11 0.10 81 97 1,658 1,186 -1.089 0.013*** (14.771) (0.005)††† 0.927 0.026 (13.656) (0.033) -7.134 -0.218 (104.576) (0.190) -0.05 -0.10 143 89 2,532 1,101 -0.087 0.031*** (0.213) (0.002)††† 0.109 -0.006 (0.174) (0.008) -0.877 -0.015 (1.332) (0.054) 0.03 0.00 84 87 1,697 1,076 (5) RF (6) IV 0.027 (0.037) -0.061 (0.079) 0.088 (0.055) 0.18 93 1,139 -0.001 (0.031) -0.059 (0.058) 0.046 (0.046) 0.11 88 1,097 -0.003 (0.028) -0.042 (0.051) 0.021 (0.046) 0.10 97 1,186 -0.052 (0.058) -0.069 (0.091) -0.020 (0.077) -0.10 89 1,101 0.019 (0.028) 0.046 (0.046) -0.070** (0.030) 0.00 87 1,076 0.034 (0.046) -0.044 (0.049) 0.378 (0.360) 0.18 93 1,139 0.001 (0.032) -0.029 (0.037) 0.224 (0.279) 0.11 88 1,097 -0.002 (0.030) -0.018 (0.032) 0.132 (0.250) 0.10 97 1,186 -0.053 (0.074) -0.011 (0.069) 0.019 (0.528) -0.10 89 1,101 0.017 (0.024) 0.015 (0.026) -0.151 (0.184) 0.00 87 1,076 Notes: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. The sample includes women who have children. The optimal bandwidth is estimated by using the Imbens and Kalyanaraman (2009) algorithm. The RD treatment effects of the reform are reported by exposure to childhood violence, i.e., whether the respondent experienced violence from her own family members during her childhood. Columns 1 reports OLS results using years of schooling as the independent variable for an optimal bandwidth ĥ estimated by the Imbens and Kalyanaraman algorithm. Columns 2 – 3 report reduced-form RD treatment effects and two-stage least-squares RD treatment effects (by using treatment as an instrument for years of schooling) of being born after January 1987 with a linear control function in the month-year of birth on each side of the discontinuity. Columns 1 – 3 report these results for the overall sample, and columns 4 – 6 report results from the same specifications for the subsample of respondents whose childhood region of residence was rural. The variables are described in Appendix A. All specifications control for a dummy variable for whether the respondent grew up in a rural location, a dummy variable for whether the respondent’s mother tongue is not Turkish, month-of-birth fixed effects, region fixed effects, and interactions of region fixed effects with an indicator of rural regions. Standard errors are clustered at the month-year cohort level. ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively (based on p-values unadjusted for multiple-hypothesis testing). †††, ††, and † denote significance at the 1, 5, and 10 percent levels, respectively (based on p-values adjusted for multiple-hypothesis testing using Simes adjustment). Table 9: Effect of Education on Partner Characteristics and Marriage Market Outcomes Overall sample (1) OLS Partner’s years of schooling Schooling Schooling × Childhood violence Childhood violence Partner’s age Mean Bandwidth Observations Schooling Schooling × Childhood violence Childhood violence Partner’s religiosity index Mean Bandwidth Observations Schooling Schooling × Childhood violence Childhood violence Marriage age Mean Bandwidth Observations Schooling Schooling × Childhood violence Childhood violence Marriage decision Mean Bandwidth Observations Schooling Schooling × Childhood violence Childhood violence Divorced Mean Bandwidth Observations Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations 0.515*** (0.028)††† -0.046 (0.066) -0.034 (0.542) 8.80 87 1,748 0.302*** (0.034)††† -0.172 (0.136) 1.985* (1.099) 24.89 106 2,059 0.004 (0.005) 0.023 (0.032) -0.457* (0.265) 0.01 84 1,718 0.325*** (0.030)††† -0.254** (0.110)† 1.547** (0.714) 20.94 37 820 0.035*** (0.004)††† 0.040*** (0.010)††† -0.352*** (0.080)††† 0.58 66 1,378 -0.001 (0.002) -0.003 (0.009) 0.069 (0.069) 0.05 71 1,444 (2) RF Rural sample (3) IV (4) OLS 0.494 2.371 0.502*** (0.351) (5.235) (0.034)††† 1.081** -1.095 -0.150 (0.537) (4.608) (0.095) -0.885*** 8.326 0.632 (0.322)†† (35.514) (0.765) 8.80 8.80 8.47 87 87 79 1,748 1,748 979 0.138 0.250 0.272*** (0.398) (3.792) (0.043)††† 0.965 0.664 0.122 (0.869) (3.705) (0.132) 0.419 -4.102 -0.174 (0.380) (28.070) (0.851) 24.89 24.89 24.71 106 106 111 2,059 2,059 1,323 0.001 0.023 -0.001 (0.068) (0.292) (0.004) -0.028 -0.043 -0.017 (0.169) (0.303) (0.032) -0.282*** 0.033 -0.084 (0.106)†† (2.323) (0.242) 0.01 0.01 0.06 84 84 82 1,718 1,718 1,022 0.006 -0.253 0.286*** (0.445) (8.375) (0.036)††† -0.130 0.055 -0.236* (0.660) (4.236) (0.127) -0.265 -0.814 1.766** (0.376) (32.998) (0.779)† 20.94 20.94 20.74 37 37 40 820 820 539 0.131** 0.346 0.038*** (0.051)† (0.560) (0.005)††† 0.100 -0.108 0.041*** (0.081) (0.430) (0.011)††† -0.094* 0.795 -0.292*** (0.050) (3.302) (0.092)†† 0.58 0.58 0.53 66 66 87 1,378 1,378 1,076 -0.015 -0.037 -0.001 (0.023) (0.101) (0.002) -0.006 0.018 -0.001 (0.046) (0.106) (0.006) 0.046 -0.095 0.031 (0.035) (0.817) (0.050) 0.05 0.05 0.05 71 71 122 1,444 1,444 1,414 (5) RF (6) IV 1.043** 1.149* (0.440)† (0.669) 0.647 -0.301 (0.726) (0.566) -0.730* 1.901 (0.420) (4.052) 8.47 8.47 79 79 979 979 0.281 0.287 (0.549) (0.611) 0.481 0.134 (0.889) (0.585) 0.579 -0.152 (0.550) (4.186) 24.71 24.71 111 111 1,323 1,323 -0.019 -0.037 (0.069) (0.078) 0.217 0.129 (0.160) (0.110) -0.279** -1.112 (0.124) (0.824) 0.06 0.06 82 82 1,022 1,022 -0.308 -7.786 (0.556) (152.057) -0.213 3.345 (0.669) (68.125) 0.062 -25.265 (0.345) (512.013) 20.74 20.74 40 40 539 539 0.162*** 0.146* (0.057)†† (0.086) 0.265*** 0.064 (0.085)†† (0.072) -0.103* -0.409 (0.060) (0.518) 0.53 0.53 87 87 1,076 1,076 0.012 0.013 (0.025) (0.026) -0.025 -0.019 (0.040) (0.027) 0.030 0.156 (0.032) (0.203) 0.05 0.05 122 122 1,414 1,414 Notes: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. The sample includes women who have children. The optimal bandwidth is estimated by using the Imbens and Kalyanaraman (2009) algorithm. The RD treatment effects of the reform are reported by exposure to childhood violence, i.e., whether the respondent experienced violence from her own family members during her childhood. Columns 1 reports OLS results using years of schooling as the independent variable for an optimal bandwidth ĥ estimated by the Imbens and Kalyanaraman algorithm. Columns 2 – 3 report reduced-form RD treatment effects and two-stage least-squares RD treatment effects (by using treatment as an instrument for years of schooling) of being born after January 1987 with a linear control function in the month-year of birth on each side of the discontinuity. Columns 1 – 3 report these results for the overall sample, and columns 4 – 6 report results from the same specifications for the subsample of respondents whose childhood region of residence was rural. The variables are described in Appendix A. All specifications control for a dummy variable for whether the respondent grew up in a rural location, a dummy variable for whether the respondent’s mother tongue is not Turkish, month-of-birth fixed effects, region fixed effects, and interactions of region fixed effects with an indicator of rural regions. Standard errors are clustered at the month-year cohort level. ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively (based on p-values unadjusted for multiple-hypothesis testing). †††, ††, and † denote significance at the 1, 5, and 10 percent levels, respectively (based on p-values adjusted for multiple-hypothesis testing using Simes adjustment). Table 10: Effects of Education on Spousal Violence Overall sample Physical violence index Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations Psychological violence index Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations Financial control index Schooling Schooling × Childhood violence Childhood violence Mean Bandwidth Observations Rural sample (1) OLS (2) RF (3) IV (4) OLS -0.016*** (0.006)†† -0.061*** (0.022)†† 1.005*** (0.195)††† 0.00 83 1,682 -0.112 (0.079) -0.124 (0.187) 0.600*** (0.115)††† 0.00 83 1,682 -0.371 (0.813) 0.127 (0.704) -0.462 (5.387) 0.00 83 1,682 -0.021*** (0.007)††† -0.044* (0.026) 0.754*** (0.233)††† 0.01 108 1,311 -0.021*** (0.004)††† 0.006 (0.016) 0.298*** (0.114)††† 0.04 115 2,177 0.024 (0.057) 0.027 (0.112) 0.338*** (0.051)††† 0.04 115 2,177 0.271 -0.025*** (1.600) (0.005)††† -0.215 0.004 (1.494) (0.027) 1.987 0.252 (11.353) (0.189) 0.04 0.06 115 79 2,177 980 -0.008 (0.005) -0.041 (0.026) 0.546** (0.228)†† -0.04 83 1,675 0.064 (0.070) -0.082 (0.162) 0.280*** (0.091)††† -0.04 83 1,675 0.335 (0.606) -0.303 (0.507) 2.569 (3.884) -0.04 83 1,675 -0.009 (0.007) -0.076** (0.032)† 0.661** (0.298)†† -0.04 74 926 (5) RF (6) IV -0.040 -0.031 (0.089) (0.104) -0.303 -0.141 (0.197) (0.129) 0.547*** 1.419 (0.130)††† (0.919) 0.01 0.01 108 108 1,311 1,311 0.036 (0.084) -0.097 (0.133) 0.315*** (0.076)††† 0.06 79 980 0.041 (0.090) -0.071 (0.090) 0.785 (0.648) 0.06 79 980 0.141 (0.108) -0.191 (0.174) 0.220 (0.141) -0.04 74 926 0.147 (0.132) -0.177 (0.131) 1.426 (0.981) -0.04 74 926 Notes: Data are from the 2014 National Survey on Domestic Violence against Women in Turkey. The sample includes women who have children. The optimal bandwidth is estimated by using the Imbens and Kalyanaraman (2009) algorithm. The RD treatment effects of the reform are reported by exposure to childhood violence, i.e., whether the respondent experienced violence from her own family members during her childhood. Columns 1 reports OLS results using years of schooling as the independent variable for an optimal bandwidth ĥ estimated by the Imbens and Kalyanaraman algorithm. Columns 2 – 3 report reduced-form RD treatment effects and two-stage least-squares RD treatment effects (by using treatment as an instrument for years of schooling) of being born after January 1987 with a linear control function in the month-year of birth on each side of the discontinuity. Columns 1 – 3 report these results for the overall sample, and columns 4 – 6 report results from the same specifications for the subsample of respondents whose childhood region of residence was rural. The variables are described in Appendix A. All specifications control for a dummy variable for whether the respondent grew up in a rural location, a dummy variable for whether the respondent’s mother tongue is not Turkish, month-of-birth fixed effects, region fixed effects, and interactions of region fixed effects with an indicator of rural regions. Standard errors are clustered at the month-year cohort level. *** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level. (Based on p-values unadjusted for multiple-hypothesis testing.) ††† Significant at the 1 percent level. †† Significant at the 5 percent level. † Significant at the 10 percent level. (Based on p-values adjusted for multiple-hypothesis testing using Simes adjustment.)