Fighting Violence Against Women: The Role of Female Political Representation

We investigate the eﬀect of women’s political leadership in local government on homicide and violence against women. Using a regression discontinuity design we compare Brazilian municipalities where a female candidate barely won to those where a female candidate barely lost mayoral elections. Having a female mayor reduces homicide rate of women by 20% and violence (physical, psychological, sexual) by 40%. These results are not due to pre-existing municipalities characteristics or other observable mayor characteristics. Overall, our ﬁndings provide compelling evidence that women holding oﬃce are eﬀective in addressing violence against women


Introduction
A growing body of academic research studies the effect of female political representation on policy decisions and outcomes, suggesting that female policymakers are more socially oriented than male ones (Hessami and da Fonseca, 2020;Brollo and Troiano, 2016;Bruce et al., 2022).In this paper, we provide evidence of the effect of female political representation on violence.In particular, we analyze whether the gender of the policymaker affects violence against women (i.e., femicide, physical violence and sexual violence) by focusing on mixed-gender electoral races in Brazilian municipalities.As emphasized by the United Nations in the sustainable development goal "eliminating all forms of violence against all women and girls in the public and private spheres" is a crucial objective to achieve gender equality and empower all women and girls.Official statistics provide a clear picture of this global tragedy.Indeed, one in three women worldwide experience physical or sexual violence affecting both women well-being and their participation in society and politics.1Moreover, more than 50% of intentional women homicide are perpetrated by intimate partners or other family members.2Despite being a global issue, violence against women is much more prevalent in low and lower-middle income countries and regions, forcing many countries to adopt specific legislation to criminalize femicide and gender-based violence. 3For instance in Brazil, according to official statistics, a woman is killed every two hours and assaulted every 15 seconds (Cerqueira and Bueno, 2020).
Over the past few decades, the share of women in politics has significantly increased in almost every country, shaping social and economic policy (Hessami and da Fonseca, 2020).Several empirical and experimental studies have documented that female empowerment and political representation affect policy decisions and outcomes, favoring social policies and interventions and reducing corruption and bribing (Chattopadhyay and Duflo, 2004;Brollo and Troiano, 2016;Eckel and Grossman, 2008).More limited is the evidence about the role of female representation in affecting violent crime.
Identifying the effects of female leaders on violence is challenging since policy decisions may be correlated to municipality characteristics.Therefore, we apply a Regression Discontinuity (RD) design focusing on close elections assuming that municipalities where municipalities where a female candidate won against a man by a narrow margin represent a good counterfactual for those municipalities where the opposite occurred (i.e., a male candidate won against a woman by a narrow margin).
We find that the presence of a female mayor in Brazilian municipalities is associated with a significant and sizable reduction in femicide by 17% and in all the other measures for violence against women (physical violence, psychological violence, sexual violence and sexual harassment) by 30% to 40%.Our results are robust to the inclusion of standard controls and to several validation and falsification tests.Theoretically, female mayors can affect violence against women via a number of mechanisms.First, female mayors could enact policies that deter violence and increase awareness about this issue.Second, the presence of local female leaders can directly affect crime through a "role-model" effect.
Third, law enforcement could become more sympathetic toward female victims (e.g., attitudes or incentives).Finally, female mayors might differ in their policy preferences in building a peaceful and equitable society: having a less adverse environment could give female victims greater self-confidence and a lower tolerance for being badly treated.
Second, this paper is related to the expanding literature on violence against women.For instance, Iyengar (2009) find that mandatory arrest in the case of domestic violence increased femicide, while Chin and Cunningham (2019) find no conclusive evidence.Luca et al. (2015) suggest that policies that restrict access to alcohol may help reduce gender violence in India, while Aizer (2010) provides evidence that a decrease in the gender wage gap reduce violence against the women.Iyer et al. (2012) finds that an increase in female representation in local government induces a large and significant rise in documented crimes against women in India reflecting improvements in reporting rather than a rise in actual crimes.Besides these results on what causes violence against women, Sabia et al. (2013) and Siddique (2021) find that sexual violence against women has important effects on the subsequent labor market outcomes of the victims.Thus, we complement the related literature by providing the first clear evidence of the existing link between female political representation and femicide and violence against women.
The paper proceeds as follows.Section 2 provides institutional details.Section 3 and Section 4 describe our data and empirical strategy, respectively.Section 5 presents our results and discusses the possible mechanisms at play.Section 6 concludes.

Violence in Brazil
Similar to other countries in Latin America, Brazil has a high level of violence against women.This is a long-term and persistent phenomenon that only in the recent years has gained attention thanks to the effort of women's activists and politicians that were able to push forward several specific legislative reforms to criminalize femicide and gender-based violence.Indeed, in 2006 the Brazilian legislators pass the Law no.11.340 (known as the "Maria da Penha" Law on Domestic and Family Violence, which establishes criminal sanctions against perpetrators of domestic violence against women, domestic violence courts but also requires Brazilian authorities to protect and assist the victims of violence through special police bodies and stations and shelters for women.In the following years additional initiatives took place and further legislation was approved.For instance, in 2013 was promoted the initiative "Mulher, Viver sem Violencia" that aimed at improving public policies in favor of women victims of violence, while in 2015 was enacted the Feminicide Law, which changed the Brazilian Penal Code, by including feminicide as a qualifier for the crime of homicide.Despite this effort in fighting violence against women, there were no substantial changes to the overall level of violence. 4For example, the female homicide rate (FHR) was 4.2 per 100 thousand in 2018, slightly decreasing from the previous 15 years in which the average homicide rate was about 4.3.Looking at the geographical distribution, we can see that the most violent states in 2018 are Roraima (FHR= 18.8), Ceará (FHR= 10.2) and Acre (FHR=8.0),while the less violent are São Paulo (FHR= 1.9), Santa Catarina (FHR= 2.6) and Piauí (FHR= 3.1).The heterogeneity in the level of violence across municipalities is quite sizable, as demonstrated by the fact that, in 2018, around 75% of municipalities did not have cases of female homicides in their territory, while those municipalities with at least one case FHR ranges between almost 0 to more than 100, with an average of 13.4.
Our empirical analysis aims to highlight whether this cross-sectional heterogeneity could be in part explained by the gender of the local political leader.

Local politics
Brazil is a federal republic governed under a presidential system, and organized in a federal government, 26 states and 5,570 municipalities.Each municipality has an autonomous local government, comprising a mayor (Prefeito) and a legislative body (Câmara Municipal).Local governments are responsible for the provision of several local public goods (e.g., primary education, culture, health care, housing, transportation and municipal infrastructure).The mayor plays a central role in defining the expenditure programs, while the city council is responsible for enacting municipality laws and overseeing the mayor on the usage of public resources.
Mayors are elected in a one-round election in municipalities with less than 200,000 registered voters, while a run-off could take place in municipalities with more 200,000 voters, when no mayoral candidate achieves at least 50% of the votes in the first round.
Mayors can be in office for up to two four-year terms.City councilors are elected based on an open list proportional representation system, in which parties' share of seats is proportional to the number of votes cast on their candidates.According to population size, the number of councilors varies from a minimum of 9 to a maximum of 55.All elected municipal officials take office from the 1st of January of the year subsequent the elections.
With respect to equal gender representation in politics, since 1997 the electoral law requires a minimum of 30% of candidates of each sex on electoral lists (e.g., party or coalition).Despite the electoral quota, the percentage of women in politics in Brazilian is relatively low both in the national and the local governments.Currently, 75 of the 513 deputies are women (14.6%), as well as 11 out of 81 senators (13.6%).Appendix Table A.1 provides some statistics on the presence of women in local politics in the sample (three consecutive terms) we employ for the analysis.Looking at the share of female mayoral candidates and female mayors we see that the participation of women in local elections is relatively small (in 2004, 8% and 7% respectively), but increases over time (an increase of 4 p.p for both measures from 2004 to 2012).Instead, the share of female councilors is steadily around 13%.

Homicide and violence data
Our analysis considers two main categories of crime against the person: i) homicide and ii) violence against the person.Data on homicide come from the Brazil Ministry of Health's TABNET Platform and cover the period 2000-2016.The Mortality Information System (Sistema de Informação de Mortalidade -SIM) provides detailed data at the municipality-year level about the causes of individuals' death.We consider intentional homicides defined as the number of deaths provoked by external causes through aggression-the group X85-Y09 of the International Classification of Diseases (ICD 10).
However, for additional analysis in the robustness section we also consider from the same source deaths due to traffic accidents and suicides collected as well as a general measure of mortality excluding homicide.All measures are expressed as rate for 100.000inhabitants.One of the main advantages of using murder as a proxy for violence is related to under-reporting.Indeed, it is well-known that official crime statistics may suffer from under-reporting that is much less relevant for murders (MacDonald, 2002).Data on violence comes from the Violence and Accidents Surveillance System (Sistema de Vigilância de Violências e Acidentes -VIVA) which provides municipality-year level data about different types of violence for the period 2009-2016.Our analysis focuses on the cases of physical, psychological, sexual violence and sexual harassment against women expressed at a rate of 100.000 inhabitants.Relevantly, the law mandates the health providers to report suspected or confirmed cases of domestic violence and/or other violence as well as sexual violence.To a certain extent, this provision of the law reduces the relevance of under-reporting issues.

Local Elections data
We focus on data about municipal elections for three electoral terms (2005-2008, 2009-2012, 2013-2016).Our data source is the Brazilian Electoral Court (Tribunal Superior Eleitoral).For each municipality-election we know for each candidate: vote share, sex, education (graduated or not), age, party of affiliation.It is worth noting that, as we apply an RD design in the empirical analysis, only municipalities with mix-gender race are considered, therefore the final sample will be composed of all municipalities-term in which the two top candidates were of different sex.

Other data
We complement the previous data with a set of municipal characteristics from Brazilian Institute of Geography and Statistics (IBGE) collected for the 2000 Brazilian decennial census.The data includes municipality-level covariates, such as population, the share of female population, average income per capita, the percentage of active individuals in the total population and of individuals employed in different economic sectors, income inequality with a GINI index, the percent of population living below national poverty line and illiterate individuals older than 15 years.

Empirical Strategy
Identifying the causal effect of having a female mayor on violence against women is challenging.Simply comparing violent outcomes of municipalities governed by a female to those governed by a male mayor would not deliver a causal estimate as the assignment of mayor sex is not random.Therefore, we apply a RD design in mixed gender electoral races using the following empirical specification: where the dependent variable, Y ist , denotes the sum of cases of violent events that took place in municipality i, belonging to state s, in term t.F ist is a dummy variable indicating if a woman win the mayoral race in election t in municipality i, while the running variable M V ist is the margin of victory in elections defined as the difference in the votes received by the two most voted candidates.f () is a polynomial function calculated on the margin of victory.X ist include a set of municipal pre-determined covariates, contemporaneous mayoral characteristics, term and state fixed effects that we include in our preferred specification to improve precision in our estimates (Calonico et al., 2019).Finally, ist is the error term.β is our coefficient of interest, which under specific assumptions (i.e., continuity of the density of the margin of victory and that the treatment does not affect other covariates), its estimate provides a causal effect.We show in the Appendix the results from standard validity checks of RD design.Specifically, we show that the density of the running variable is continuous at the threshold (Appendix Figure A.1) and that pre-determined characteristics are balanced (Appendix Table A.4).5 For the actual implementation, we use a linear function with rectangular kernel and employ a mean-squared error (MSE) optimal bandwidth (Calonico et al., 2014), while errors are clustered at the municipality level to account for serial correlation in the error component.

Violence against women
Our results are graphically presented in Figures 1 and 2 and show the relationship between the margin of victory and the per capita number of violent outcomes (in log) once we partial out covariates and fixed effects. 6In Figure 1 we report the results when focusing on female homicides and violent acts against women (i.e., aggregating all type of violence), using alternatively the contemporaneous or prior term outcome.Interestingly, we can identify discontinuity at the threshold for the contemporaneous outcome, while no clear discontinuity is displayed for the outcome on the previous term.Overall, the graphical evidence hints at the presence of an effect of having a female mayor on violent acts against women, which is not confounded by pre-existing differences.In Figure 2 we detailed more the type of violent acts by reporting results separately for physical  In Table 1 we report the formal estimates as defined in the empirical strategy section. 7In the first panel we show the RD estimates when we do not include controls, while in the second panel we include as covariates municipal level pre-determined charac- and 5%.As the homicide rate in a term is on average 29 per 100 thousand female inhabitants, the estimated effect would imply a reduction of around 5 cases per 100 thousand women.For all types of violence (from column 2 to 4) and sexual harassment (column 5) we find again significant and negative effects, which are larger than the one estimated for homicide rates.Physical violence is reduced by 43% to 46%, psychological violence of between 38% and 40%, sexual violence of around 33% and sexual harassment of between 54% to 39%.All coefficients reach the conventional level of statistical significance.Notes: The dependent variable is defined as the sum of violent events in an electoral term in per-capita terms.The column-heads identify the type of violent events.Covariates include municipality level and mayoral level characteristics as well as electoral term and state fixed effects.Municipality features include population size, occupational composition, income level, income inequality and previous experience with a female mayor.Mayoral features are age, level of education and party of affiliation (PT, PSDB, DEM, PMDB).Estimates in Column 1 are based on data from three electoral terms (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016), while in Columns 2 to 5 the estimates use data for two electoral terms (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016).The coefficients are constructed using local quadratic estimators with rectangular kernel.Robust standard errors clustered at the municipality level in parenthesis * p < 0.1, ** p < 0.05 and *** p < 0.01.

Additional Results
In this section, we provide evidence about the relation of having a female mayor and other outcomes.First, we test whether the effect of having a female mayor on violent acts is present also when looking at men as victims.Therefore, we report in panel A of Table 2 the estimates from our main specification, using this time as dependent variables violence against men.Column 1 and 2 show that there is no effect on homicides rate and (total) violence, respectively. 9This result seems to reinforce the idea that having a female mayor does not have a general effect on violent crimes, but, rather the effect is limited to violence against women.It also rules out the possibility that our main result could be driven by some change associated with the arrival of a female mayor in power that would alter the overall level of violence.For example, this would be the case if one expects a female mayor to apply stronger policies to contrast inequality or to be effective in positively affecting economic growth, given the link between economic condition and violence (Aizer, 2010).
Next, we report a set of additional results addressing other possible concerns related to the main evidence being just the consequence of some structural reforms affecting unintentional homicide occurring during a female mayor's tenure.Again in Table 2, we show the effect of having a female mayor on general mortality (column 3), motor vehicle fatality accidents rate (column 4) and suicide rate (column 5) for both males (panel A) and females (panel B) separately.Overall, the coefficients are very close to zero and none of them is statistically significant, suggesting the specific deterrence effect towards crime against women.

Discussion on the mechanisms
Our results point to a clear "reduced form" effect of having a female mayor on violence against women.However, the actual mechanisms that made this possible are not simple to clearly identify with the available data.Therefore, in this section, we discuss a set of potential channels that could explain our evidence and that are in line with previous findings.First, women might propose or enact policies with different objectives from 8 To provide the correct percentage effect of the estimated treatment we apply the transformation 100 × [exp(estimated effect)−1]. 9We do not report the results separately for each type of violence because of the limited number of events, which drastically reduces the number of observations and therefore the reliability of the estimates.
those preferred by men simply because gender matters in the kind of life experiences one has.This seems plausible in our context as women can be expected to be more sympathetic than men toward the specific type of crime studied in this paper.Yet this is unlikely to happen directly via law and order as in the Brazilian setting police activity and law enforcement are tasks assigned to upper levels of government.Nevertheless, we cannot exclude some indirect influences that make the police more responsive to crimes against women, for instance by making the issue more salient.In addition, it is worth noting that violence against women could be indirectly affected by other policies that female mayors are more likely to pursue.For instance, female mayors might be more willing to help women to have access to economic and social resources and to support the introduction of pro-women institutions to protect women under threat.
Next, besides policy changes one can expect our results to be explained by a role model hypothesis where having a female mayor would help change the norms that accept violence against women.For instance, one can expect the exposure of women to successful female politicians to empower women which will be more likely to stand against acts of violence (Jensen and Oster, 2009).Similarly, there might be changes in how men perceive the role of women in society, therefore, changing the interactions between the two sexes.

Conclusion
This paper highlights the effect of female political leadership on femicide and violence against women.By using Brazilian municipality level data for the period 2005-2016 and exploiting an RD design in close elections, we provide evidence that the election of a female mayor causes a large, negative and significant effect on femicide and violence against women.Our findings are robust to placebo regressions and to standard validation and falsification tests in the RD design.This effect is not confounded by the initial presence of violence and is not part of a more general reduction in violence.Our evidence emphasizes one positive aspect of increasing female representation in public offices.More research is needed to empirically test what are the channels through which women in politics are affecting violence (e.g., role model or changes in policies).Notes: The table reports the share of women among the top two candidates in mayoral elections, the share of female mayor and the share of female in the city council, for the three electoral terms used in the analysis.[385,390] [279,282] [295,275] [197,177] [111,110] Notes: The dependent variable are defined as the sum of violent events in an electoral term in per-capita terms.
Overall, we have 3,080 observations, of which 804 are from the term 2005-2008, 1023 from the term 2009-2012 and 1253 from the term 2012-2016.

Figure
Figure 1: Female Mayor and Violence against Women (a) Homicides (b) Homicides in t-1

Figure 2 :
Figure 2: Female Mayor and different type of Violence against Women (a) Physical Violence (b) Psychological Violence

Figure A. 1 :
Figure A.1: Continuity of the density of the margin of victory

Figure
Figure A.2: Different cut-off (a) Homicides Table A.2, while their description and sources are in Appendix Table A.3.

Table 2 :
Female Mayor and Violence: Placebo

Table A
Notes: The variable All type of violence includes physical, sexual and psychological; The variable Other homicides includes all cases of homicides except caused by agression (ICD10, All categories except X85-Y09) Table A.8: Female Mayor and Violence Against Women: Interaction with level of education