Women political leaders as agents of environmental change

Berniell, Inés; Marchionni, Mariana; Pedrazzi, Julián; Viollaz, Mariana Working Paper Women political leaders as agents of environmental change IDB Working Paper Series, No. IDB-WP-1704 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Berniell, Inés; Marchionni, Mariana; Pedrazzi, Julián; Viollaz, Mariana (2025) : Women political leaders as agents of environmental change, IDB Working Paper Series, No. IDBWP-1704, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0013521 This Version is available at: https://hdl.handle.net/10419/324789 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/igo/ Women Political Leaders as A gents of Environmental Change Inés Berniell Mariana Marchionni Julián Pedrazzi Mariana Viollaz WORKING PAPER No IDB-WP-1704 InterA merican Development Bank Gender and Diversity Division May 2025 * CEDLAS - IIE - Universidad Nacional de La Plata ** CEDLAS - IIE - Universidad Nacional de La Plata and CONICET Women Political Leaders as Agents of Environmental Change Inés Berniell* Mariana Marchionni** Julián Pedrazzi** Mariana Viollaz* InterA merican Development Bank Gender and Diversity Division May 2025 Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Women political leaders as agents of environmental change / Inés Berniell, Mariana Marchionni, Julián Pedrazzi, Mariana Viollaz. p. cm. — (IDB Working Paper Series ; 1704) Includes bibliographical references. 1. Environmental policy-Brazil. 2. Climatic changes-Government policy-Brazil. 3. Women in conservation of natural resources-Brazil. 4. Environmental impact analysis-Brazil. 5. Gender mainstreaming-Brazil. I. Berniell, Inés. II. Marchionni, Mariana. III. Pedrazzi, Julián. IV. Viollaz, Mariana. V. InterAmerican Development Bank. Gender and Diversity Division. VI. Series. IDB-WP-1704 http://www.iadb.org Copyright © 2025 Inter-American Development Bank ("IDB"). This work is subject to a Creative Commons license CC BY 3.0 IGO (https://creativecommons.org/licenses/by/3.0/igo/legalcode). The terms and conditions indicated in the URL link must be met and the respective recognition must be granted to the IDB. Further to section 8 of the above license, any mediation relating to disputes arising under such license shall be conducted in accordance with the WIPO Mediation Rules. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the United Nations Commission on International Trade Law (UNCITRAL) rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this license. Note that the URL link includes terms and conditions that are an integral part of this license. The opinions expressed in this work are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. Abstract1 This paper explores how female political leaders impact environmental outcomes and climate change policy actions using data from mixed-gender mayoral races in Brazil. We rely on a Regression Discontinuity design that compares municipalities where women narrowly won the election with those where men narrowly won. This strategy allows us to identify the causal effect of a woman winning the mayoral election. We find that, compared to male mayors, female mayors significantly reduce greenhouse gas emissions and deforestation in the municipalities with Amazon biome. Specifically, when a woman wins the election, an-nual greenhouse gas emissions decrease by 1,510 thousand tons of CO2e per municipality in the Amazon. This effect alone represents 23% of the average annual emissions of all municipalities within the Amazon biome and 6.4% of Brazil’s nationwide average. This reduction is driven by a reduction in emissions intensity (CO2e/GDP) in the Land Use sector, without changes in municipal economic activity. Part of the reduction in emissions in the Land Use sector is attributable to a decline in deforestation. Specifically, female-led municipalities in the Amazon experience a reduction in deforestation, with a 3 percentage-point decrease in the loss of forest formations relative to the baseline forest cover. This represents a 32%reduction compared to deforestation levels in the comparison municipalities. We examine potential mechanisms that could explain the positive environmental impact of narrowly electing a female mayor over a male counterpart and find that in Amazon municipalities, female elected mayors allocate more space to the environment in their government proposals and are more likely to invest in environmental initiatives. Differences in the enforcement of environmental regulations and the level of education of elected female and male mayors do not explain the results. JEL classifications: J16, D72, Q54, Q56, Q58 Keywords: Gender, Climate change, Mayoral elections, Amazon, Brazil, Latin America 1This research received funding from the Gender and Diversity Knowledge Initiative, a GDLab, InterAmerican Development Bank, as part of the project “Climate Change and Its Nexus with Gender and Diversity.” We are particularly indebted to Felipe Jordan and Isabelle Chort for their extensive and invaluable feedback. We are also very grateful to Cristian Bonavida, Matias Ciaschi, Guillermo Cruces, Guillermo Falcone, Leonardo Gasparini, Carlos Lamarche, Ignacio Lunghi, and Mariano Rabassa, as well as the GDLab team, for their insightful suggestions. We also thank seminar participants at Universidad Nacional de La Plata, Universidad de Buenos Aires, and the LIX Annual Meeting of Asociación Argentina de Economía Política at Universidad Nacional de Salta for helpful discussion. The usual disclaimer applies. 1 Introduction Climate change is a pressing global concern that requires adequate and timely policies. Rising greenhouse gas emissions are rapidly increasing average global temperatures and altering rainfall patterns. These environmental changes have significant social and economic implications, including job losses, reduced productivity and hours of work, deficient development of foundational cognitive skills, changes in migration patterns, and increased poverty (e.g., Graff Zivin and Neidell,2014;Jafino et al.,2020;Saget et al., 2020;Salemi,2021;Pazos et al.,2024). Climate change intersects with gender in important ways. First, political leadership plays a vital role in addressing climate change, and previous research shows that a leader’s gender can influence policy decisions (Chattopadhyay and Duflo,2004). Women leaders tend to promote better social outcomes, including increased spending and better health and education, reduced gender-based violence, lower levels of corruption, improved institutional quality, and higher rates of economic growth (Dollar et al.,2001;Jayasuriya and Burke,2013;Bhalotra and Clots-Figueras,2014;Bruce et al.,2022;Delaporte and Pino,2022;Bochenkova et al.,2023). Second, studies indicate that perceptions of climate change differ by sex, with women generally exhibiting more awareness and concern than men (Ergas and York,2012;Dechezleprêtre et al.,2022). These differences are often attributed to women’s traditional roles as caregivers, subsistence food producers, and water and fuelwood collectors, as well as gender differences in values such as cooperation and attentiveness (Ergas and York,2012). Given these patterns, we expect female leaders to adopt distinct environmental approaches compared to male leaders. Furthermore, since climate change exacerbates existing gender inequalities—disproportionately affecting women and girls through resource scarcity, increased risks of violence during crises like droughts, and higher likelihood of displacement during climate disasters (Habtezion, 2016;Marcos Morezuelas,2021;Dehingia et al.,2024)—it is plausible to anticipate an additional effect if leaders are more likely to implement policy actions that address the needs of their own gender (Chattopadhyay and Duflo,2004). Therefore, we hypothesize that female political leaders might perform better than men in addressing climate change. We test this hypothesis using information on mixedgender close mayoral electoral races in Brazilian municipalities. There are several reasons for focusing our analysis on Brazil. The country holds 60% of the Amazon rainforest, the largest tropical forest in the world, which plays a key role in global climate regulation. In addition, Brazil has more than five thousand municipalities with information on various outcomes related to climate change and on policy variables relevant to our analysis. Finally, Brazil is a highly decentralized country, and its municipalities have decisionmaking power on environmental issues. Previous evidence supports our hypothesis. The cross-country studies of Zhike and Deng (2019), Ergas and York (2012) and Asongu et al. (2022) report a negative association between women’s political empowerment and total carbon dioxide (CO2) emissions, 2 women’s political representation and CO2 emissions per capita, and between women’s political representation and vulnerability to climate change, respectively, using data from developed and developing countries around the world. Similarly, other cross-country studies use information from countries and regions of various development levels to show that there is a positive relationship between the share of women in parliament and the ratification of environmental treaties (Norgaard and York,2005), stringency of climate change policies (Mavisakalyan and Tarverdi,2019), production of renewable energy (Bansal and D’Agosti,2023), and consumption of renewable energy (Slamon,2023). Only a few studies, however, provide causal analysis on this topic. For instance, Beraldi and Fosco (2024) analyze the case of Italy using a difference-in-difference instrumental variables approach. They find that an increase in the percentage of female councilors in Italy decreases air pollution, and the effect is likely explained by an increase in the density of bicycle lanes, urban green spaces, bicycle and car sharing services, use of centralized heating systems, and use of traffic blockage. Closer to our study, Jagnani and Mahadevan (2021) implement an RD strategy to show that the election of women legislators in India leads to a reduction in crop-fire incidents, while Neumman López (2024) uses an RD strategy based on mixed-gender electoral races in Mexico to show that electing a woman as municipal president over a man improves water management, a measure related to the consequences of climate change. To the best of our knowledge, there is no causal evidence about how female political leaders impact direct measures of climate change—i.e., greenhouse gas emissions and deforestation—compared to their male counterparts in Latin America, particularly within the Amazon region, or on how their policy decisions regarding climate change mediate this relationship. Identifying the causal effect of female political leaders on environmental and policy outcomes is challenging because of the presence of other factors, such as societal attitudes toward women, that may be correlated with both women winning an election and the outcome variables. To tackle this issue, we adopt a Regression Discontinuity (RD) design that compares outcomes in municipalities where a woman won an election by a narrow margin against a male candidate with municipalities where the winner was, by a narrow margin, a man (Lee et al.,2004). The intuition behind this comparison is that in close election races, the probability of winning is the same for women and men. Then, municipalities with a male mayor who won by a narrow margin are a good counterfactual for municipalities with female mayors who also won by a narrow margin. Previous studies have applied this approach in Brazil to analyze the impact of female leadership on other outcomes such as corruption (Brollo and Troiano,2016), health outcomes (Bruce et al., 2022), and gender-based violence (Bochenkova et al.,2023;Delaporte and Pino,2022). We analyze outcome variables related to greenhouse gas emissions and deforestation at the municipal level. These two outcomes are related, since deforestation is one of the main contributors to the levels of emissions. In fact, in Latin America, around 40% of the greenhouse gas emissions in 2019 originated from land use changes, such as converting forests to pastures or agricultural lands, and the burning of forest residues (Brassiolo 3 et al.,2023). For greenhouse gas emissions, we consider the average annual emissions per municipality within each four-year mayoral term, both in tons of carbon dioxide equivalent (CO2e) and relative to municipal economic activity (CO2e/GDP). Additionally, we distinguish between total municipal emissions and sector-specific emissions. As for deforestation, we focus on changes in forest formations in each municipality over the four-year term of the mayors’ mandates, measured as a percentage of the municipal forest cover. We conduct our analysis for all Brazilian municipalities with close mixed-gender elections and the subsamples of municipalities with vegetation typical of the Amazon biome (Amazon municipalities) and those without it (non-Amazon municipalities). Additionally, we innovate by analyzing the mechanisms behind the impacts of having a woman elected as a mayor on climate change-related outcomes. These potential mechanisms include the importance assigned to the environment in elected mayors’ government proposals, public spending and institutions devoted to environmental protection in each municipality, and enforcement of environmental regulations. For the first mechanism we consider the percentage of words in a candidate’s government proposal that are related to the environment. For the second set of mechanisms we define three outcome variables: an indicator of whether the municipality has established an environmental council, a second capturing whether the municipality incurred any environmental expenditures during each mayoral term, and a measure of the percentage of the municipal budget allocated to environmental expenditures over each term. Enforcement efforts are measured as the number of fines issued due to deforestation infractions detected in each municipality. Our findings show that there is a positive impact on environmental outcomes at the municipal level when a woman wins a mayoral election against a man by a narrow margin and that these effects are driven by municipalities with Amazon biome. In these municipalities, when a woman wins the election, annual greenhouse gas emissions decrease by 1,510 thousand tons of CO2e per municipality. This effect alone represents 23% of the average annual emissions of all municipalities within the Amazon biome and 6.4% of Brazils nationwide average. Importantly, this reduction is due to a reduction in emissions intensity (CO2e/GDP) in the Land Use sector, without changes in municipal economic activity. Part of the reduction on emissions in the Land Use sector is attributable to a decline in deforestation. Female-led municipalities in the Amazon experience a reduction in deforestation, with a 3 percentage-point decrease in the loss of forest formations relative to municipal forest cover. This represents a 32% reduction compared to deforestation levels in the comparison municipalities. These results are robust to changes in key parameters of our RD specification and to adjustments in the estimation sample, including changes in the bandwidth choice, kernel, and polynomial order, and excluding observations near the cutoff. We also show that the main results are not driven by the subset of municipalities that were part of the List of Priority Municipalities (LPM), a federal initiative that sought to reduce deforestation rates establishing targets of reduction for municipalities with the highest deforestation rates and sanctions in case of non-compliance. 4 A potential challenge in identifying the effects of the gender of the winning politician is that female candidates in close elections may possess compensating attributes, such as higher ability, to overcome gender biases of the voters (Marshall,2024). These attributes could influence outcomes, causing our estimates to reflect a compound effect of bundled treatment rather than an isolated gender effect. In our analysis, most predetermined characteristics are balanced near the cutoff, except that female mayors are more educated. While higher education could influence our results if it correlates with environmentally protective policies, we present evidence showing that education does not significantly affect our key outcomes. Additionally, we show that voters probably did not prioritize environmental issues when voting, making compensatory differentials targeting such concerns unlikely and not biasing our results. However, unobservable factors, such as non-cognitive skills, political experience, or networks, can act as compensating differentials that help women win elections and implement environmental policies. Therefore, we acknowledge that our estimates may capture both gender effects and these additional influences. The analysis of mechanisms indicates that female leaders respond differently to climate change in terms of public policy, especially in Amazon municipalities. In this area, policy proposals from female elected mayors contain 0.16 percentage points more environmental-related terms compared to those from male elected mayors, which represents a 50% increase relative to the baseline. Additionally, when a woman wins the election, the likelihood of these municipalities investing in environmental initiatives increases by 13 percentage points. The evidence also suggests differences in the enforcement of environmental regulations do not explain our main results. This paper contributes to three strands of literature. First, we add to the literature that analyzes the impacts of female political leaders on various outcomes. As previously mentioned, the election of women as political leaders impacts positively on children’s health and education, economic growth rates, and quality of institutions, and negatively on corruption and violence against women. Second, we relate to the literature exploring the political economy of environmental outcomes. Previous evidence shows that deforestation at the municipal level tends to be higher when mayors are farmers—a powerful interest group that opposes to conservation policies (Bragança and Dahis,2022)—or when campaign donations come from landholders (Katovich and Moffette,2024); in contrast, deforestation and greenhouse gas emissions are lower when the mayor is a young politician, likely due to their greater exposure to climate change-oriented education and to recent cultural shifts towards environmental protection, compared to senior mayors (Dahis et al.,2023). Finally, we also add to the environmental justice literature by showing that women political leaders achieve better environmental outcomes when in office, thereby giving voice and representation to the group of women who are disproportionately impacted by climate change. The remainder of the paper is organized as follows. Section 2delves into the institutional context, focusing on Brazil’s federal organization and delineating the powers vested 5 in state sobserved during the four-year mayoral term t.F emaleMayor equals 1 if a woman won the mayoral race in the election held in the year prior to the start of term t in municipality iand state s.F emaleV oteMarginist is the running variable in this RD design, and it is defined as the vote share of the female candidate minus the vote share of the male candidate. The polynomial function f(FemaleV oteMarginist)represents the fitted polynomials in the female vote margin on both sides of the threshold, which is estimated following Calonico et al. (2014).8In our main analysis, we use a first-degree polynomial and show in the robustness section that our findings hold for higher-order polynomials. ϵist represents the error term. The baseline specification of Equation 1 includes mayoral term fixed effects (γt) and state fixed effects (λs). In other specifications we include additional controls to improve precision in our estimates (Calonico et al., 2019). These controls include predetermined municipality characteristics (population size and value added by sector), contemporaneous mayoral characteristics (age, level of education, marital status, and party of affiliation), and the value of the dependent variable in the year prior to the start of term t. We cluster standard errors at the municipal level. βis the coefficient of interest. While we cannot guarantee that our strategy isolates the causal effect of the mayor’s gender on the outcomes of interest (see Section 5.5), it does capture the causal effect of a combined treatment, i.e., the impact of a woman winning the election (Marshall,2024), provided that the density of the running variable is continuous at the threshold and that predetermined characteristics are balanced. To show that near the cutoff, municipalities with a female mayor are similar to those with a male mayor in terms of observable characteristics, we estimate equation 1for a range of characteristics at both the mayor level—political party, education, age, marital status, among others—and municipal level—total population, female population, per capita GDP, literacy rate, among others. Figure 1a reports the t-statistics of βfrom equation 1, where such characteristics are the outcome variables. We find that our sample is balanced in most predetermined characteristics, with the exception of the candidates education. We complement these results with a graphical illustration of the RD effects for each pre-determined variable, providing additional evidence that they do not exhibit discrete jumps at the cutoff (see Figures A.2 and A.3 in the Appendix).9Additionally, in Figure 1b we conduct a similar analysis for the subsample of municipalities with vegetation typical of the Amazon biome (Amazon municipalities), confirming that the characteristics are also generally balanced within this subsample. This subsample is particularly 8It is worth clarifying that the optimal bandwidth in our RD design does not define what constitutes aclose election. Rather, it determines the range of observations used to fit the local polynomial for estimation (Calonico et al.,2014) Thus, our coefficient of interest is identified from elections with narrow margins (i.e., close elections), aided by a local polynomial estimated using observations slightly further from the zero-margin cutoff. 9Previous studies utilizing an RD design in mixed-gender close elections in Brazil have demonstrated the balance of predetermined municipal and mayoral characteristics at the threshold, supporting the validity of our identification strategy. Like our findings, these studies show that these characteristics are generally balanced, with the exception of education, where female mayors tend to be more educated than their male counterparts. For instance, see the work of Brollo and Troiano (2016) on corruption and Bochenkova et al. (2023) on violence against women. 12 relevant for our analysis, as Section 5shows that the environmental effects of electing a woman are concentrated in the Amazon region. Secondly, although unlikely in our context, we verify the absence of manipulation of the running variable in a local neighborhood near to the cutoff (FemaleV oteMarginist = 0). We run the manipulation test based on the density discontinuity developed in Cattaneo et al. (2018), and find a p-value of 0.6, i.e., we fail to reject the null hypothesis of no difference in the density of municipalities with a female mayor and with a male mayor at the cutoff. Figure A.4 in the Appendix graphically illustrates the continuity in the density test approach: the density estimates on both sides of the cutoff are very similar, and the confidence intervals overlap for both the entire sample of municipalities with mixed-gender elections and the subsample of municipalities in the Amazon region. One potential identification concern with our RD strategy is the presence of spatial spillovers between municipalities. For example, if a female mayor leads to a decrease in deforestation and emissions in her own municipality but worsens environmental outcomes in neighboring municipalities governed by men, our point estimates could be artificially inflated. Moreover, such spatial spillover effects suggest that our analysis captures only partial effects—the local environmental impact in municipalities governed by women– rather than the general equilibrium effect on overall deforestation and emissions driven by female mayors. In an extreme case, reductions in deforestation and emissions in municipalities led by women could be fully offset if negative spillovers in neighboring municipalities governed by men are large enough. While these spillovers warrant a cautious interpretation of the estimated effects magnitude, it is important to note that if we estimate that female mayors improve environmental outcomes in their municipalities, the direction of the effect should remain unchanged, even in the presence of spatial spillovers. 13 Figure 1: Covariates Balance around the Threshold (a) Total Literacy rate (2000) Sh. Men (2000) Sh. Women (2000) Sh. Urban population (2000) Sh. Rural population (2000) Sh. Active population (2000) Sh. Active men (2000) Sh. Active women (2000) Sh. Employed population (2000) Sh. Employed men (2000) Sh. Employed women (2000) Sh. Low−Skill (2000) Sh. Med−Skill (2000) Sh. High−Skill (2000) Share Agro (2000) Share Industrial (2000) Share Services (2000) Share Public administration (2000) Total Population (2000) Total Men population (2000) Total Women population (2000) Total urban population (2000) Total Rural population (2000) Total Population > 10 (2000) Literate population (2000) GDP (2000) College Married Age PT PMBD DEM PSDB −4 −3 −2 −1 0 1 2 3 4 5 6 T−statistic (b) Amazon region Literacy rate (2000) Sh. Men (2000) Sh. Women (2000) Sh. Urban population (2000) Sh. Rural population (2000) Sh. Active population (2000) Sh. Active men (2000) Sh. Active women (2000) Sh. Employed population (2000) Sh. Employed men (2000) Sh. Employed women (2000) Sh. Low−Skill (2000) Sh. Med−Skill (2000) Sh. High−Skill (2000) Share Agro (2000) Share Industrial (2000) Share Services (2000) Share Public administration (2000) Total Population (2000) Total Men population (2000) Total Women population (2000) Total urban population (2000) Total Rural population (2000) Total Population > 10 (2000) Literate population (2000) GDP (2000) College Married Age PT PMBD DEM PSDB −4 −3 −2 −1 0 1 2 3 4 5 6 T−statistic Notes: The figures show the t-statistics from our RD (Equation 1) using predetermined municipal and mayor characteristics as outcomes for all municipalities with mixed-gender elections (Panel a) and for the subsample of Amazon municipalities (Panel b). Estimates obtained using local linear estimators with a triangular kernel. Optimal bandwidth based on Calonico et al. (2014). Robust standard errors clustered at the municipal level. The red, blue and green lines indicate the 1%, 5% and 10% significance level thresholds respectively. 5 Main Results In this section we present the main results about the impact on environmental outcomes of women winning municipal elections in Brazil. Our analysis focuses on greenhouse gas emissions and deforestation, two key indicators that can capture efforts to mitigate climate change. Figure 2begins by providing preliminary evidence on the relationship between the 14 margin of victory and greenhouse gas emissions. The horizontal axis measures the margin of victory in favor of a woman, defined as the difference between the percentage of votes received by the female candidate and the percentage of votes received by the male candidate in mixed-gender municipal elections. A positive margin of victory indicates a win for the woman. The figure shows a discontinuity in emissions at the municipal level depending on whether women or men won: in municipalities where a woman won (i.e., to the right of zero), greenhouse gas emissions are lower compared to municipalities where a man won, both across the entire country (Figure 2a) and when focusing solely on municipalities within the Amazon biome (Figure 2b). The graphical evidence suggests an improvement in environmental outcomes at the municipal level when a woman wins the election.10 Figure 2: Female Mayor and Emissions (in tons of CO2e) (a) All municipalities −200000 0 200000 400000 600000 Emissions (in tons of CO2e) −12 −6 0 6 12 Difference in votes share (b) Amazon municipalities −2000000 0 2000000 4000000 6000000 Emissions (in tons of CO2e) −12 −6 0 6 12 Difference in votes share Note: These figures graphically illustrate the effect of female mayors on total greenhouse gas emissions in municipalities with mixed-gender elections, both across the entire country (left panel) and within the Amazon region (right panel). Emissions are measured in tons of COe, averaged per year over each four-year mayoral term. The solid lines illustrate first-degree polynomials fitted to the running variable on either side of the threshold. Gray dots correspond to averages for bins of the running variable, with vertical lines showing the 90% confidence intervals for these averages. In what follows, we report the formal estimates based on the RD design described in Section 4. Overall, these results validate the preliminary findings from Figure 2. 5.1 Effect on Greenhouse Gas Emissions Table 2shows the RD estimates regarding greenhouse gas emissions in tons of CO2e, calculated as the annual average over the four years of each mayoral term. Column 1 10Figures A.5a and A.5b in the Appendix present another descriptive analysis that suggests environmental outcomes at the municipal level improve relatively more when a woman wins the election. These figures show the evolution of total emissions and the share of deforestation, on average, for the four years before and during the four years in office, separately for male and female mayors. The sample includes mixed-gender close elections, where a man and a woman are the two main candidates. The figures suggest that while emissions and deforestation decrease for both genders, the decline is sustained when a woman wins the election, whereas, when a man wins, the levels remain similar by the end of the term. 15 corresponds to the baseline model in equation 1, which only includes controls for mayoral term and state fixed effects. This is our preferred specification. Columns 2 to 4 gradually introduce additional controls—municipality controls, mayor controls, and the outcome from the previous year. In particular, column 3 includes controls at the mayor’s level— age, level of education, marital status, and party of affiliation. Notably, the results are highly robust across specifications, indicating that the effect we identify from a woman winning the election is not due to differences between female and male mayors in terms of education, political party, or other included observable characteristics. In what follows, we primarily focus on the baseline model in column 1 to describe our findings. Panel A in the table reports a statistically significant reduction in annual CO2e emissions in municipalities with female mayors compared to those with male mayors. Specifically, in municipalities where women narrowly won the election, annual emissions are, on average, 219 thousand tons of CO2e lower than in municipalities where men won by a similar margin. This represents a 75% reduction in the average annual municipal emissions per mayoral term when a woman is elected.11 Next, we re-estimate our RD model for two subsamples: municipalities with vegetation typical of the Amazon biome (Amazon municipalities) and those without (nonAmazon municipalities). Panels B and C of Table 2report these results, confirming that the Amazon municipalities are driving the effect identified for the entire sample. When a woman wins the election in Amazon municipalities, annual greenhouse gas emissions decrease by 1,510 thousand tons of CO2e per municipality. This effect is larger than that estimated for the total set of municipalities. In contrast, we find no evidence of effects on emissions in the non-Amazon municipalities. The key takeaway from these results is the statistically significant and negative effect of female leadership on emissions, particularly in the Amazon region. However, the magnitude of the estimated impact should be interpreted with caution. As is often the case in this type of studies, estimated effects tend to be relatively large, reflecting local treatment effects around the threshold.12 Indeed, in a robustness check accounting for potential outliers, our effect remains negative and statistically significant, although its magnitude is reduced by half (see Section 5.4). That said, it is useful to illustrate the scale of these effects with some back-of-theenvelope calculations. If we focus on the 113 Amazon municipalities with mixed-gender elections, they collectively produce an average of 88 thousand tons of CO2e per year 11We compute the baseline mean outcome for electoral term tas the average annual CO2e emissions in the preceding term. This is specifically calculated for municipalities where men candidates won the election corresponding to term t, with the margin of victory falling within the optimal bandwidth. 12Large effect sizes are commonly reported in the literature when using similar methodologies. For instance, Dahis et al. (2023) study the impact of young politicians on emissions in Brazil and find estimated effects ranging from 72% to over 130%. Similarly, Jagnani and Mahadevan (2021) show that the election of women legislators in India leads to a reduction in crop-fire incidents, with effects between 33% and 100%. These large effects are not unique to environmental outcomes—Bochenkova et al. (2023) find impacts on domestic homicide rates between 50% and 70%. Other examples include Eslava (2024) and Chauvin and Tricaud (2024), where estimated effects relative to mean consistently exceed 60%. 16 Table 2: Female Mayor and Emissions (in tons of CO2e) (1) (2) (3) (4) Panel A: Total Female mayor -218,756** -225,477** -230,113** -265,355** (101,193) (101,179) (101,512) (104,064) Mean outcome 291,668 291,668 291,668 291,668 Bandwidth 12.8 12.8 12.8 12.8 Observations [1156, 1035] [1156, 1035] [1156, 1035] [1156, 1035] Panel B: Amazon region Female mayor -1,509,865*** -1,348,315** -1,098,764** -1,428,845** (585,299) (558,625) (546,566) (575,735) Mean outcome 1,449,452 1,449,452 1,449,452 1,449,452 Bandwidth 12.8 12.8 12.8 12.8 Observations [140, 125] [140, 125] [140, 125] [140, 125] Panel C: Non-amazon region Female mayor -27,744 -36,305 -42,404 -44,069 (37,958) (37,936) (41,676) (40,598) Mean outcome 184,980 184,980 184,980 184,980 Bandwidth 16.2 16.2 16.2 16.2 Observations [1275, 1051] [1275, 1051] [1275, 1051] [1275, 1051] Year & State FE Yes Yes Yes Yes Municipality controls No Yes Yes Yes Mayor controls No No Yes Yes Outcome t-1 No No No Yes Notes: The dependent variable is the average annual emissions in tons of CO2e over the four-year term in all municipalities with mixed-gender elections (Panel A), the subsample of Amazon municipalities (Panel B), and the subsample of non-Amazon municipalities (Panel C). Municipality controls include population size, GDP, and share of value added by sector (agriculture, industry, services, and public administration). Mayor controls are age, level of education, marital status and party of affiliation (PT, PSDB, DEM, PMDB). Estimates obtained using local linear estimators with a triangular kernel. Optimal bandwidth based on Calonico et al. (2014). Robust standard errors clustered at the municipal level in parentheses * p < 0.1, ** p < 0.05 and *** p < 0.01. during the period under analysis.13 Without the effect of female mayors in these municipalities, annual emissions would have reached 160 thousand tons of CO2e—i.e., an 81% increase for this group. Notably, this change alone represents 23% of the average annual emissions of all municipalities within the Amazon biome and 6.4% of Brazil’s nationwide average. If we considered extending our findings to the rest of the Amazon municipalities (i.e., to those excluded from our sample due to the absence of mixed-gender elections), increasing the share of female mayors from 2% to just 20%—half the proportion in our sample—could reduce emissions by 104 thousand tons of CO2e. This represents a 35% decrease relative to the current annual emissions of these municipalities. While such extrapolation requires caution, this estimate provides a useful illustration of the potential scale of these effects. Furthermore, the effect of women winning municipal elections on emissions is primarily driven by changes in land use within Amazon municipalities. Table 3shows that 13For this calculation, we consider the average number of municipalities per electoral round by dividing the number of municipality-terms reported in Table A.1 by four, as the dataset spans four electoral rounds and their respective mayoral terms. 17 the estimated effect is an annual reduction in emissions from land use activities of 1,462 thousand tons of CO2e per municipality when women won the election relative to when men won. In contrast, we find very small and statistically insignificant effects of a woman winning the election on emissions in other sectors, such as agriculture, energy, or waste (columns 2, 3, and 5, respectively).14 Table 3: Female Mayor and Emissions by Sector (in tons of CO2e) Total Agriculture Energy Land use Waste (1) (2) (3) (4) (5) Panel A: Total Female mayor -218,756** 5,628 2,913 -232,379** 882 (101,193) (17,163) (5,017) (93,063) (781) Mean outcome 291,668 91,178 17,003 172,653 6,348 Bandwidth 12.8 13.6 7.4 13.1 11.7 Observations [1156, 1035] [1224, 1063] [722, 680] [1179, 1041] [1077, 977] Panel B: Amazon region Female mayor -1,509,865*** -40,342 -7,676 -1,461,500*** 2,215 (585,299) (93,096) (9,404) (550,012) (1,399) Mean outcome 1,449,452 262,855 25,436 1,130,693 8,179 Bandwidth 12.8 10.8 11.7 13.2 6.4 Observations [140, 125] [130, 110] [133, 114] [144, 125] [84, 77] Panel C: Non-amazon region Female mayor -27,744 8,114 4,717 -45,960 920 (37,958) (10,685) (5,489) (33,695) (780) Mean outcome 184,980 71,466 15,388 85,424 6,162 Bandwidth 16.2 14.7 7.3 15.6 14.1 Observations [1275, 1051] [1178, 1003] [633, 598] [1235, 1035] [1141, 978] Year & State FE Yes Yes Yes Yes Yes Notes: The dependent variable is the average annual emissions in tons of CO2e over the four-year term in all municipalities with mixed-gender elections (Panel A), the subsample of Amazon municipalities (Panel B), and the subsample of non-Amazon municipalities (Panel C). The results correspond to our baseline specification, similar to that in Column 1 of Table 2, which controls for year and state fixed effects. Estimates obtained using local linear estimators with a triangular kernel. Optimal bandwidth based on Calonico et al. (2014). Robust standard errors clustered at the municipal level in parentheses * p < 0.1, ** p < 0.05 and *** p < 0.01. 5.2 Emissions Intensity vs. Economic Activity Using an IPAT decomposition framework as a lens (Ehrlich and Holdren,1971), the reduction in emissions in tons of CO2e can be understood as stemming either from a decline in emissions intensity (CO2e/GDP) or from a contraction in economic activity. Table 4presents the RD estimates of emissions in tons of CO2e per 1,000 BRL (Brazilian 14These results for emissions are based on average emissions measured over the entire four-year mayoral term. We also conducted a year-by-year analysis for both the full sample of municipalities and the subsample of Amazon municipalities. Whether we consider total emissions or those specifically from the Land Use sector, we find no clear temporal pattern within the mayoral term. The point estimates do not suggest any significant changes over time, and no statistically significant differences were found across the years. Additionally, to examine whether the decrease in emissions under female mayors persists or reverses in subsequent terms, we estimated our main specification using outcomes from the next mayoral term. While we also observe a decline in future emissions, the effects are mostly not statistically significant. Results are available upon request. 18 Reais) of GDP, calculated as the annual average over the four years of each mayoral term, which allows us to assess the first channel. The results show that emissions intensity decreases when a woman mayor is elected by a narrow margin compared to when a man is elected. This effect is driven by changes in land use within Amazon municipalities. When a woman wins the election in these municipalities, total greenhouse gas emissions in the Land Use sector decrease by 3.5 tons of CO2e per 1,000 BRL of GDP. In contrast, we find no evidence of effects on emissions intensity in other sectors or in non-Amazon municipalities. Table 4: Female Mayor and Emissions Intensity (CO2e/GDP) by Sector Total Agriculture Energy Land use Waste (1) (2) (3) (4) (5) Panel A: Total Female mayor -0.546** 0.036 -0.001 -0.584** -0.000 (0.274) (0.054) (0.006) (0.254) (0.001) Mean outcome 1.124 0.435 0.039 0.618 0.023 Bandwidth 10.4 14.2 11.5 10.4 11.7 Observations [989, 892] [1262, 1089] [1055, 956] [985, 890] [1077, 980] Panel B: Amazon region Female mayor -3.504** 0.020 0.001 -3.509** -0.006 (1.652) (0.296) (0.015) (1.606) (0.004) Mean outcome 3.871 0.799 0.041 2.996 0.025 Bandwidth 10.9 12.9 12.6 10.9 12.6 Observations [130, 110] [141, 125] [139, 122] [130, 110] [140, 124] Panel C: Non-amazon region Female mayor -0.103 0.055 -0.001 -0.159 0.001 (0.142) (0.045) (0.007) (0.131) (0.001) Mean outcome 0.847 0.406 0.039 0.381 0.023 Bandwidth 13.9 13.2 12.8 13.5 12.3 Observations [1131, 971] [1081, 943] [1033, 924] [1101, 952] [1001, 904] Year & State FE Yes Yes Yes Yes Yes Notes: The dependent variable is emissions intensity, measured in tons of CO2e per 1,000 BRL of GDP, calculated as the annual average over the four years of each mayoral term, in all municipalities with mixed-gender elections (Panel A), the subsample of Amazon municipalities (Panel B), and the subsample of non-Amazon municipalities (Panel C). The results correspond to our baseline specification, similar to that in Column 1 of Table 2, which controls for year and state fixed effects. Estimates obtained using local linear estimators with a triangular kernel. Optimal bandwidth based on Calonico et al. (2014). Robust standard errors clustered at the municipal level in parentheses * p < 0.1, ** p < 0.05 and *** p < 0.01. Regarding the second channel, we use our RD model to assess the impact of a woman winning an election on economic activity. Our findings indicate that the observed reductions in emissions are driven solely by changes in intensity and not by shifts in economic activity. In fact, Table 5shows that when we apply the same RD design using municipal GDP, whether in levels or in logs, as the outcome variable, the effect of a woman winning the election is never statistically significant, which means that the reduction in emissions primarily operates through improved CO2e efficiency, i.e., fewer emissions per GDP. 19 Table 5: Female Mayor and Economic Activity GDP Log GDP (1) (2) Panel A: Total Female mayor -194776 -0.018 (177825) (0.091) Mean outcome 385,406 12.065 Bandwidth 8.0 12.8 Observations [781, 737] [1152, 1033] Panel B: Amazon region Female mayor 40460 -0.176 (129346) (0.219) Mean outcome 381,909 12.329 Bandwidth 8.2 13.7 Observations [103, 95] [149, 126] Panel C: Non-amazon region Female mayor -158560 -0.006 (191352) (0.098) Mean outcome 402,326 12.053 Bandwidth 10.4 13.5 Observations [857, 766] [1068, 913] Year & State FE Yes Yes Notes: The dependent variables are municipality GDP (Column 1) and its logarithm (Column 2), in all municipalities with mixed-gender elections (Panel A), the subsample of Amazon municipalities (Panel B), and the subsample of non-Amazon municipalities (Panel C). The regressions control for year and state fixed effects. Estimates obtained using local linear estimators with a triangular kernel. Optimal bandwidth based on Calonico et al. (2014). Robust standard errors clustered at the municipal level in parentheses * p < 0.1, ** p < 0.05 and *** p < 0.01. 5.3 Effect on Deforestation According to MapBiomas (2024), over 90% of deforestation in the Amazon is driven by the creation of pastureland, a major contributor to greenhouse gas emissions in the Land Use sector. Deforestation plays a critical role in emissions, primarily by releasing carbon stored in trees and soil. When deforestation is reduced, the forest’s carbon sequestration capacity is preserved, leading to lower emissions. To better understand how changes in land use contribute to the negative impact on emissions that we find, we now examine the impact of a woman winning the election on deforestation. Table 6presents the RD estimates of deforested hectares as a share of forest cover at the municipal level. The findings indicate that women winning elections lead to a reduction in deforestation exclusively in Amazon municipalities. According to the baseline specification in Column 1, a woman’s victory in these municipalities reduces the 20 deforested area by 3 percentage points relative to the total forest cover, which represents a 35% decrease relative to the baseline level. These results are consistent and highly robust across different specifications. In contrast, no significant effects on deforestation are observed in non-Amazon municipalities.15 Table 6: Female Mayor and Deforestation (as a share of forest cover) (1) (2) (3) (4) Panel A: Total Female mayor -0.004 -0.004 -0.005 -0.005 (0.004) (0.005) (0.005) (0.005) Mean outcome 0.049 0.049 0.049 0.049 Bandwidth 11.8 11.8 11.8 11.8 Observations [1054, 950] [1054, 950] [1054, 950] [1054, 950] Panel B: Amazon region Female mayor -0.033* -0.032* -0.033* -0.029* (0.018) (0.017) (0.017) (0.016) Mean outcome 0.104 0.104 0.104 0.104 Bandwidth 11.5 11.5 11.5 11.5 Observations [130, 110] [130, 110] [130, 110] [130, 110] Panel C: Non-amazon region Female mayor -0.005 -0.004 -0.004 -0.005 (0.004) (0.004) (0.004) (0.004) Mean outcome 0.045 0.045 0.045 0.045 Bandwidth 12.6 12.6 12.6 12.6 Observations [974, 862] [974, 862] [974, 862] [974, 862] Year & State FE Yes Yes Yes Yes Municipality controls No Yes Yes Yes Mayor controls No No Yes Yes Outcome t-1 No No No Yes Notes: The dependent variable is the total deforested area of forest in each municipality over each four-year term as a share of the total forest cover in each municipality during the baseline year (i.e., the year prior to the start of each term) in all municipalities with mixedgender elections (Panel A), the subsample of Amazon municipalities (Panel B), and the subsample of non-Amazon municipalities (Panel C). Municipality controls include population size, GDP and share of value added by sector (agriculture, industry, services, and public administration). Mayor controls are age, level of education, marital status and party of affiliation (PT, PSDB, DEM, PMDB). Estimates obtained using local linear estimators with a triangular kernel. Optimal bandwidth based on Calonico et al. (2014). Robust standard errors clustered at the municipal level in parentheses * p < 0.1, ** p < 0.05 and *** p < 0.01. 15Primary and secondary forests together make up 70% of the natural vegetation cover in Amazon municipalities. Among these forest formations, 83% correspond to primary forest—i.e., natural forest that has remained intact since the beginning of the series (1987) through the year of analysis—while the remaining 17% is secondary forest—i.e., forest that has regenerated after the original vegetation was removed or significantly altered by human activity. The results in Table 6are primarily driven by changes in primary forest. In contrast, we find no effects of women winning the election on changes in secondary forest. When we replicate our analysis considering changes in the entire natural vegetation cover—including savanna, mangrove, wetland, grassland, and other non-forest natural formations—rather than just forest, the results remain robust. Results are available upon request. 21 Figure 3: Political Proposals, Public Spending and Institutions (a) Government proposals Administration Child Culture Education Environmental Health Security Social Assistance Women −1 −.5 0 .5 1 RDD coefficient Amazon Total (b) Environmental policies % Environmental expenditure Any Environmental expenditure Create Environmental council −.2 0 .2 .4 .6 RDD coefficient Amazon Total Notes: These figures show the βcoefficient and its 90% confidence interval from estimating Equation 1 for all municipalities with mixed-gender elections and for the subsample of Amazon municipalities. The dependent variables are the following: number of words related to each topic in relation to the total number of words in the proposal (Panel a) and a binary variable indicating if the mayor created an environmental council during the four-year term, another binary indicator showing whether the municipality made any environmental expenditure, and a variable measuring the percentage of the municipal budget allocated to environmental expenditures (Panel b). Estimates obtained using local linear estimators with a triangular kernel and optimal bandwidth based on Calonico et al. (2014). All estimates account for state and round fixed effects. ics, and operation of municipal public institutions carried out by the Instituto Brasileiro de Geografia e Estatística (IBGE). Based on these data, we construct several variables that serve as proxies for the effort municipalities devote to environmental policy. Our outcomes of interest are: a binary variable indicating if the municipality established an environmental council during the mayors’ mandate, another binary indicator showing whether the municipality made any environmental expenditure over the four-year term of the mayors mandates, and a variable measuring the percentage of the municipal budget allocated to environmental expenditures in each four-year term. Table A.1 shows the averages of the variables related to Public spending and Institutions. In the sample with mixed-gender elections, 55% of the municipalities have an environmental council and, while 57% of the municipalities made any environmental expenditure, this expenditures represents only 0.48% of the total budget on average. Although the percentage of municipalities with an environmental council is slightly higher in the remaining municipalities (60%), there are no statistically significant differences between municipalities with and without mixed-gender elections for the other variables. The results in Figure 3b suggest that, once in office, female leaders respond differently to climate change in terms of the allocation of resources to environmental management. Specifically, in line with the findings presented in Section 5, the difference is evident in municipalities with Amazon biome. When a woman wins the election in Amazon municipalities, the likelihood of these municipalities investing in environmental initiatives increases by 13 percentage points, and the share of the budget allocated to these initiatives rises by 0.2 percentage points, though this effect is not statistically significant. Amazon 28 municipalities led by women are also more likely to create an environmental council compared with those led by men, but this effect is not significant in statistical terms. Enforcement of environmental regulations. To explore this potential mechanism, we examine whether female and male mayors differ in the number of environmental fines issued during their mayoral term due to environmental infractions detected in their municipalities.18 We use data on the number of environmental fines issued throughout the mayoral term due to deforestation infractions detected in each municipalities. These data come from the Brazilian Institute of Environment and Renewable Natural Resources. According to this source, municipalities with mixed-gender elections issued an average of three fines related to deforestation infractions (see Table A.1), and this number is not significantly different from the average number of fines issued in other municipalities. Although differences between female and male mayors in inspection activities may lead to discrepancies in the number of fines, the direction of the effect is not clear. For instance, higher enforcement efforts could reduce the number of fines in equilibrium if land users update their perceived probability of being caught, thereby reducing illegal deforestation. Figure 4shows no statistically significant gender differences in the number of fines due to deforestation infractions in the entire sample of municipalities or those with Amazon biome. Given the limitation of the number of environmental fines as a proxy of enforcement effort mentioned above, we interpret this result as suggestive evidence of differential enforcement efforts between female and male mayors not driving our main results. 18Although the institution responsible for issuing environmental infraction reports is the Brazilian Institute of Environment and Renewable Natural Resources, a federal institution, the responsibility for environmental inspections is shared with the states, municipalities, and the federal district. 29 Figure 4: Female Mayor and Number of Environmental Fines Total fines −40 −20 0 20 40 RDD coefficient Amazon Total Notes: These figures show the βcoefficient and its 90% confidence interval from estimating equation 1 for all municipalities with mixed-gender elections and for the subsample of Amazon municipalities. The dependent variables are the following: total deforested area in hectares per municipality over each fouryear mandate (Panel a), emissions per unit of municipal GDP (Panel b), and total number of fines due to deforestation infractions per mayoral term (Panel c). Estimates obtained using local linear estimators with a triangular kernel and optimal bandwidth based on Calonico et al. (2014). All estimates account for state and round fixed effects. 7 Conclusions Climate change has significant social and economic negative implications and requires adequate and timely policies. Understanding the role of women in policy decisions regarding climate change is important given the evidence indicating that many social outcomes improve when the leader is a woman and that women are generally more aware and concerned than men about climate change. In this paper, we have analyzed how female political leaders impact climate change policy actions and environmental outcomes using data from mixed-gender close mayoral races in Brazilian municipalities and applying a Regression Discontinuity design. Our findings reveal a significant positive effect on environmental outcomes at the municipal level when a woman narrowly defeats a male opponent in a mayoral election, particularly in Amazon municipalities. In these municipalities, annual greenhouse gas emissions decrease by 1,510 thousand tons of CO2e when a woman is elected mayor. This implies that, without the effect of female mayors, annual emissions would have increased by 81% in mixed-gender election municipalities in the Amazon. Notably, this change alone represents 23% of the average annual emissions of all municipalities within the Amazon biome and 6.4% of Brazil’s nationwide average. Moreover, the reduction in emissions is driven by a decrease in emissions intensity (CO2e/GDP) within the Land Use sector, without changes in municipal economic activity. Part of the reduction on emissions in the Land Use sector is attributable to a decline in deforestation. Specifically, female-led municipalities in the Amazon experience a reduction in deforestation compared to male-led municipalities, with a 3 percentage 30 point decrease in the deforested area relative to total vegetation cover, representing a 32% reduction compared to the baseline deforestation levels. Furthermore, our exploration of the underlying mechanisms indicate that female mayors adopt distinct approaches to climate change public policy, particularly in the Amazon region. The policy proposals put forth by female elected mayors feature 0.16 percentage points more references to environmental-related terms than those proposed by their male counterparts, representing a 50% increase relative to the baseline. Moreover, the election of a woman significantly increases the likelihood of investing in environmental initiatives by 13 percentage points. Importantly, our findings suggest that differences in the enforcement of environmental regulations do not explain our findings. Finally, we show that these results are robust across various checks. Specifically, they hold when applying different bandwidths, excluding municipalities located very close to the cutoff, and employing different kernel and polynomial functions. Additionally, our findings hold when excluding municipalities there were part of the List of Priority Municipalities, indicating that the results are not driven by this policy, and we also show that differences in the observed skill levels of female and male mayors do not explain the results. While our study focuses on the Amazon, a region of critical global environmental significance, further research is needed to explore whether similar patterns emerge in other contexts. Expanding this analysis to different geographic regions, particularly those facing distinct environmental and governance challenges, could provide a broader understanding of the role of female leadership in climate policy. For instance, examining the impact of female political leaders in regions affected by industrial pollution, water scarcity, or extreme weather events could shed light on whether their influence extends beyond deforestation and land use policies. Additionally, it would be of great interest to investigate how the presence of strong national environmental policies or external international pressures, such as global environmental agreements or climate finance mechanisms, might interact with gender differences in leadership. By expanding this analysis to a broader range of regions, we could better assess whether women’s leadership has consistently positive environmental impacts or if such effects are context-dependent, thus offering more generalizable insights into the role of women in climate governance. Summarizing, our analysis shows that electing a woman as mayor leads to significant improvements in climate-related outcomes compared to electing a male mayor in Amazon municipalities. These effects likely stem from gender differences in public policy decisions. Although extrapolating these results to other contexts is not straightforward, this evidence underscores the value of increasing womens political participation, as it not only strengthens environmental governance but also addresses key climate challenges. In short, our findings highlight the role of women as agents of environmental change. 31 References Asongu, S. A., Messono, O. O., and Guttemberg, K. T. (2022). Women political empowerment and vulnerability to climate change: evidence from 169 countries. Climatic Change, 174(3-4):30. 2 Bansal, M. and D’Agosti, N. (2023). Women in power: the role of gender in renewable energy policymaking. Environmental Research: Climate, 3(2). 3 Beraldi, A. L. and Fosco, G. (2024). Clearing the air: Women in politics and air pollution. MPRA Paper 121377.3 Bhalotra, S. and Clots-Figueras, I. (2014). Health and the political agency of women. American Economic Journal: Economic Policy, 6(2):164–197. 2 Bochenkova, A., Buonanno, P., and Galletta, S. (2023). Fighting violence against women: The role of female political representation. Journal of Development Economics, 164:103140. 2,3,12,16 Bragança, A. and Dahis, R. (2022). Cutting special interests by the roots: Evidence from the brazilian amazon. Journal of Public Economics, 215:104753. 5,7 Brassiolo, P., Estrada, R., Vicuña, S., Odriozola, J., Toledo, M., Juncosa, F., Fajardo, G., and Schargrodsky, E. (2023). Desafíos globales, soluciones regionales: América latina y el caribe frente a la crisis climática y de biodiversidad. 3 Brollo, F. and Troiano, U. (2016). What happens when a woman wins an election? evidence from close races in brazil. Journal of Development Economics, 122:28–45. 3, 12 Bruce, R., Cavgias, A., Meloni, L., and Remígio, M. (2022). Under pressure: Womens leadership during the covid-19 crisis. Journal of development economics, 154:102761. 2,3 Calonico, S., Cattaneo, M. D., Farrell, M. H., and Titiunik, R. (2019). Regression discontinuity designs using covariates. Review of Economics and Statistics, 101(3):442– 451. 12 Calonico, S., Cattaneo, M. D., and Titiunik, R. (2014). Robust data-driven inference in the regression-discontinuity design. The Stata Journal, 14(4):909–946. 12,14,17,18, 19,20,21,28,30,38,39,41,42,43,44,45,46 Cattaneo, M. D., Idrobo, N., and Titiunik, R. (2019). A practical introduction to regression discontinuity designs: Foundations. Cambridge University Press. 22,23 Cattaneo, M. D., Jansson, M., and Ma, X. (2018). Manipulation testing based on density discontinuity. The Stata Journal, 18(1):234–261. 13,40 Chattopadhyay, R. and Duflo, E. (2004). Women as policy makers: Evidence from a randomized policy experiment in india. Econometrica, 72(5):1409–1443. 2 Chauvin, J. P. and Tricaud, C. (2024). Gender and electoral incentives: evidence from crisis response. Technical report, National Bureau of Economic Research. 16 32 Dahis, R., de las Heras, I., and Saavedra, S. (2023). Young politicians and long-term policy. Technical report, Documento de Trabajo 20694, Universidad del Rosario. 5,16 Dechezleprêtre, A., Fabre, A., Kruse, T., Planterose, B., Chico, A. S., and Stantcheva, S. (2022). Fighting climate change: International attitudes toward climate policies. Technical report, National Bureau of Economic Research. 2 Dehingia, N., McDougal, L., Silverman, J., Reed, E., Urada, L., McAuley, J., Singh, A., and Raj, A. (2024). Climate and gender: Association between droughts and intimate partner violence in india. American Journal of Epidemiology, 193(4):634–645. 2 Delaporte, M. and Pino, F. (2022). Female political representation and violence against women: Evidence from brazil. IZA Discussion Paper.2,3,6 Dollar, D., Fisman, R., and Gatti, R. (2001). Are women really the fairer sex? corruption and women in government. Journal of Economic Behavior & Organization, 46(4):423– 429. 2 dos Santos Massoca, P. and Brondízio, E. (2022). National policies encounter municipal realities: a critical analysis of the outcomes of the list of priority municipalities in curbing deforestation in the brazilian amazon. World Development, 158. 7,24 Ehrlich, P. R. and Holdren, J. P. (1971). Impact of population growth: Complacency concerning this component of man’s predicament is unjustified and counterproductive. Science, 171(3977):1212–1217. 18 Ergas, C. and York, R. (2012). Womens status and carbon dioxide emissions: A quantitative cross-national analysis. Social Science Research, 41(4):965–976. 2 Eslava, F. (2024). Conflict and female leadership: Evidence from colombia. Technical report, Working Paper. 16 Graff Zivin, J. and Neidell, M. (2014). Temperature and the allocation of time: Implications for climate change. Journal of Labor Economics, 32(1):1–26. 2 Habtezion, S. (2016). Gender and climate change: Overview of linkages between gender and climate change. United Nations Development Programme.2 Jafino, B., Walsh, B., Rozenberg, J., and Hallegatte, S. (2020). Revised estimates of the impact of climate change on extreme poverty by 2030. Policy Research Working Paper 9417.2 Jagnani, M. and Mahadevan, M. (2021). Women leaders improve environmental outcomes: Evidence from crop fires in india. Presented at the 4th IZA Workshop on Gender and Family Economics. 3,16 Jayasuriya, D. S. and Burke, P. J. (2013). Female parliamentarians and economic growth: Evidence from a large panel. Applied Economics Letters, 20(3):304–307. 2 Katovich, E. and Moffette, F. (2024). Does local politics drive tropical land-use change? property-level evidence from the amazon. 5 Lee, D. S., Moretti, E., and Butler, M. J. (2004). Do voters affect or elect policies? evidence from the us house. The Quarterly Journal of Economics, 119(3):807–859. 3 33 MapBiomas (2024). Project MapBiomas - Collection 8 of Brazilian Land Cover Use Map Series. https://brasil.mapbiomas.org/en/estatisticas/. [Accessed: Mar-2024]. 20 Marcos Morezuelas, P. (2021). Gender, forests and climate change. IDB Technical Note 2124.2 Marshall, J. (2024). Can close election regression discontinuity designs identify effects of winning politician characteristics? American Journal of Political Science, 68(2):494– 510. 5,12,24,25 Mavisakalyan, A. and Tarverdi, Y. (2019). Gender and climate change: Do female parliamentarians make difference? European Journal of Political Economy, 56:151– 164. 3 Neumman López, N. C. (2024). Clearing the air: Women in politics and air pollution. ITAM Bachelor of Economics Thesis.3 Neves, E. M. S. C. (2016). Institutions and environmental governance in Brazil: the local governments’ perspective. Revista de Economia Contemporânea, 20:492–516. 6,7 Neves, E. M. S. C. and Whately, M. (2016). Municipalities and policies against deforestation in the brazilian amazon. Novos estudos CEBRAP, 35:67–83. 6,7 Norgaard, K. and York, R. (2005). Gender equality and state environmentalism. Gender & Society, 19(4):506–522. 3 Pazos, N., Favara, M., Sánchez, A., Scott, D., and Behrman, J. (2024). Long-term effects of early life rainfall shocks on foundational cognitive skills: Evidence from peru. Economics & Human Biology, 54. 2 Rodrigues Afonso, J. and Amorim Araujo, E. (2006). Local government organization and finance: Brazil, chapter 11, pages 381–416. The World Bank. 7 Saget, C., Vogt-Schilb, A., and Luu, T. (2020). Jobs in a net-zero emissions future in latin america and the caribbean. Inter-American Development Bank and International Labour Organization.2 Salemi, C. (2021). Refugee camps and deforestation in sub-saharan africa. Journal of Development Economics, 152. 2 SEEG (2021). Soluções para redução gas emissões de gases de efeito estuga nos municípios brasileros. 6 Sills, E., Pfaff, A., Andrade, L., Kirkpatrick, J., and Dickson, R. (2020). Investing in local capacity to respond to a federal environmental mandate: Forest & economic impacts of the green municipality program in the brazilian amazon. World Development, 129:104891. 6,7 Slamon, H. (2023). The effect of women’s parliamentary participation on renewable energy policy outcomes. European Journal of Political Research, 63:174–196. 3 Souza, C. (2002). Brazil: the prospects of a centerconstraining federation in a fragmented polity. Publius, 32(2):3–22. 6 34 Zhike, L. and Deng, C. (2019). Does women’s political empowerment matter for improving the environment? a heterogeneous dynamic panel analysis. Sustainable Development, 27(4):603–612. 2 35 A Tables and Figures Figure A.1: Distribution of Mixed-gender Elections (a) Round 2004 (b) Round 2008 (c) Round 2012 (d) Round 2016 Note: Mixed-gender elections are defined as elections where the two candidates with the largest share of votes are a man and a woman. “Others” are the rest of the municipalities. 36 Table A.1: Descriptive Statistics Municipalities with Rest of mixed-gender elections municipalities Mean Obs. Mean Obs. Difference P-value Panel A: Geographic and economic characteristics Total area in ha. 170213 3,888 145671 16,890 24541 0.01 GDP (thousands of constant Brazilian reais) 719438 3,889 852787 16,893 -133349 0.28 GDP per capita (constant Brazilian reais) 21360 3,889 23982 16,893 -2623 0.00 Agriculture value added/GDP 0.176 3,889 0.187 16,893 -0.010 0.00 Industry value added/GDP 0.116 3,889 0.127 16,893 -0.011 0.00 Services value added/GDP) 0.357 3,889 0.375 16,893 -0.018 0.00 Public administration value added/GDP 0.296 3,889 0.256 16,893 0.040 0.00 Panel B: Population characteristics Total population 25,862 3,889 27,429 16,893 -1,566 0.48 Share of urban population in 2000 0.574 3,864 0.585 16,695 -0.010 0.01 Share of rural population in 2000 0.426 3,864 0.415 16,695 0.010 0.01 Share of active women in 2000 0.372 3,864 0.394 16,695 -0.022 0.00 Share of employed women in 2000 0.313 3,864 0.335 16,695 -0.021 0.00 Share of low skilled in 2000 0.850 3,864 0.839 16,695 0.012 0.00 Share of medium skilled in 2000 0.120 3,864 0.131 16,695 -0.011 0.00 Share of high skilled in 2000 0.016 3,864 0.019 16,695 -0.003 0.00 Panel C: Mayor characteristics Female 0.42 3,889 0.02 16,893 0.40 0.00 College 0.53 3,889 0.45 16,893 0.08 0.00 Married 0.73 3,889 0.79 16,893 -0.06 0.00 Age 47.23 3,889 47.77 16,893 -0.54 0.00 Party PT 0.08 3,887 0.09 16,893 -0.01 0.17 Party PMBD 0.19 3,887 0.20 16,893 -0.00 0.75 Party DEM 0.08 3,887 0.08 16,893 -0.00 0.31 Party PSDB 0.14 3,887 0.14 16,893 -0.01 0.35 Other political party 0.51 3,887 0.49 16,893 0.02 0.03 Panel D: Emissions per municipality Emissions Agriculture 101,252 3,889 95,258 16,893 5,994 0.08 Emissions Energy 36,971 3,855 38,741 16,778 -1,770 0.63 Emissions Waste 8,285 3,889 8,915 16,893 -630 0.02 Emissions Land use 82,885 3,888 70,858 16,890 12,027 0.43 Emissions Total 229,047 3,889 213,496 16,893 15,552 0.36 Emissions Total Amazon 779,483 453 774,242 1,538 5,241 0.97 Emissions Total non-Amazon 170,750 3,295 177,300 14,985 -6,550 0.40 Panel E: Deforestation per municipality Total deforestation/baseline forest 0.05 3,770 0.05 16,552 0.00 0.09 Total Amazon deforestation/baseline forest 0.10 453 0.08 1,538 0.02 0.00 Total non-Amazon deforestation/baseline forest 0.04 3,295 0.04 14,985 -0.00 0.65 Panel F: Environmental governance and expenditures Municipality with an Environmental Council 0.55 3,889 0.60 16,893 -0.04 0.00 Municipality with environmental expenditure 0.57 3,875 0.59 16,836 -0.01 0.12 Percentage of environmental expenditure 0.48 3,875 0.47 16,836 0.01 0.32 Total environmental fines 3.08 3,889 2.57 16,893 0.51 0.09 Percentage of environmental related words 0.28 1,892 0.28 6,884 -0.00 0.76 Notes: The table considers all municipalities where elections were resolved in the first round, without the need for a rerun and without irregularities. These municipalities are divided into two groups: municipalities with mixed-gender elections (defined as those where the two candidates with the largest share of votes are a man and a woman) and the remaining municipalities. The total number of observations refers to the municipality-term units. Difference and p-values refer to the difference between the means of the two samples and the statistical significance of this difference. Emissions are measured as the average annual emissions in tons of CO2e for each four-year term of the mayors mandates. Deforestation is the total deforestation of forest cover in each municipality over the four-year term of the mayors’ mandates, measured as a percentage of the total forest cover in each municipality during the baseline year, which is the year prior to the start of each term. The share of active women in 2000 is measured as the total number of economically active women aged 10 years or older divided by the total number of women aged 10 years or older in 2000, while the share of employed women in 2000 is calculated as the total number of employed women aged 10 years or older divided by the total number of economically active women aged 10 years or older in 2000. The share of low-skilled, medium-skilled, or high-skilled individuals is measured as the total number of people aged 10 years or older with 0 to 8 years of education, 9 to 13 years of education, or 14 or more years of education in 2000, respectively, divided by the total number of people aged 10 years or older in 2000. 37 Figure A.8: Sensitivity to Observations Near the Cutoff A. All municipalities with mixed-gender elections. (a) Emissions in tons of CO2e . −500000 −400000 −300000 −200000 −100000 0 RDD coefficient 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Exclude Abs(share)<= X (b) Emissions in tons of CO2e Land use −500000 −400000 −300000 −200000 −100000 RDD coefficient 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Exclude Abs(share)<= X (c) Deforestation as share of forest cover −.015 −.01 −.005 0 .005 RDD coefficient 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Exclude Abs(share)<= X B. Amazon municipalities. (d) Emissions in tons of CO2e . −3000000 −2000000 −1000000 0 RDD coefficient 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Exclude Abs(share)<= X (e) Emissions in tons of CO2e Land use −3000000 −2000000 −1000000 0 RDD coefficient 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Exclude Abs(share)<= X (f) Deforestation as share of forest cover −.08 −.06 −.04 −.02 0 RDD coefficient 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Exclude Abs(share)<= X Notes: These figures show the βcoefficient and its 90% confidence interval from estimating equation 1 10 times, starting from no exclusions (the baseline case) and progressively excluding observations within 0.1%, 0.2%, 0.3%, up to 1% of the cutoff. Estimates are obtained using local linear estimators with a triangular kernel. The optimal bandwidth is based on Calonico et al. (2014). Table A.3: Female Mayor and Emissions: Different Kernel and Polynomial Order Total Emissions in tons of CO2e Land Use Emissions in tons of CO2e (1) (2) (3) (4) (5) (6) (7) (8) Panel A: Total Female mayor -218,756** -234,122** -230,211** -278,043** -232,379** -258,011** -242,291** -263,267** (101,193) (112,829) (110,210) (121,037) (93,063) (102,760) (102,288) (108,029) Mean outcome 291,668 262,323 339,591 285,693 172,653 181,222 217,123 160,848 Bandwidth 12.8 20.8 8.6 14.6 13.1 23.3 8.5 16.4 Observations [1156, 1035] [1644, 1304] [830, 774] [1296, 1113] [1179, 1041] [1742, 1373] [826, 769] [1411, 1185] Panel B: Amazon region Female mayor -1,527,424*** -1,199,905* -1,547,957** -1,148,758* -1,477,081*** -1,103,449* -1,476,340** -1,090,148* (587,018) (641,794) (648,061) (681,953) (552,254) (591,302) (613,005) (631,903) Mean outcome 1,440,656 1,394,211 1,713,731 1,534,648 1,138,774 1,117,241 1,394,030 1,115,417 Bandwidth 13.0 14.3 8.9 9.9 13.4 13.6 8.9 12.0 Observations [142, 125] [154, 133] [107, 99] [122, 104] [143, 125] [147, 125] [107, 99] [135, 120] Year & State FE Yes Yes Yes Yes Yes Yes Yes Yes Order polynomial 1 2 1 2 1 2 1 2 Kernel Triangular Triangular Uniform Uniform Triangular Triangular Uniform Uniform Notes: The dependent variable is the average annual emissions in tons of CO2e over the four-year term in all municipalities with mixed-gender elections (Panel A), the subsample of Amazon municipalities (Panel B). Estimates obtained using local linear estimators with a triangular kernel. Optimal bandwidth based on Calonico et al. (2014). Robust standard errors clustered at the municipal level in parentheses * p < 0.1, ** p < 0.05 and *** p < 0.01. 44 Table A.4: Female Mayor and Deforestation: Different Kernel and Polynomial Order Deforestation as a share of forest cover (1) (2) (3) (4) Panel A: Total Female mayor -0.004 -0.003 -0.006 -0.004 (0.004) (0.005) (0.005) (0.006) Mean outcome 0.049 0.049 0.049 0.049 Bandwidth 11.8 17.0 8.0 14.3 Observations [1054, 950] [1407, 1172] [761, 710] [1243, 1064] Panel B: Amazon region Female mayor -0.033* -0.030 -0.031* -0.039* (0.018) (0.020) (0.017) (0.022) Mean outcome 0.104 0.103 0.104 0.104 Bandwidth 11.5 17.8 11.9 13.9 Observations [130, 110] [175, 151] [131, 113] [148, 125] Year & State FE Yes Yes Yes Yes Order polynomial 1 2 1 2 Kernel Triangular Triangular Uniform Uniform Notes: The dependent variables is deforestation as share of the baseline forest cover in each municipality over the four-year term. Panel A estimates equation 1in the total sample and Panel B restricts the sample to Amazon biome municipalities. Estimates obtained using local linear estimators with a triangular kernel. Optimal bandwidth based on Calonico et al. (2014). Robust standard errors clustered at the municipal level in parentheses * p < 0.1, ** p < 0.05 and *** p < 0.01. Figure A.9: Excluding Municipalities in the LPM. (a) Emissions in tons of CO2e . LU Emissions Total emissions −6 −4 −2 0 RDD coefficient Amazon Total (b) Deforestation as a share of forest cover Deforestation (share) −.06 −.04 −.02 0 RDD coefficient Amazon Total Notes: These figures show the βcoefficient and its 90% confidence interval from estimating Equation 1for all municipalities with mixed-gender elections and for the subsample of Amazon municipalities. The dependent variables are the following: the average annual emissions in tons of CO2e over the fouryear term in all municipalities (Panel a) and deforestation as share of the baseline forest cover in each municipality over the four-year term (Panel b). Estimates obtained using local linear estimators with a triangular kernel and optimal bandwidth based on Calonico et al. (2014). All estimates account for state and round fixed effects. 45 Figure A.10: Differential Skills between Female and Male Mayors (a) College effect: Emissions in tons of CO2e LU Emissions (levels) Total emissions (levels) −1000000 −500000 0 500000 1000000 RDD coefficient Amazon Total (b) College effect: Deforestation as share of forest cover Deforestation (share) −.02 −.01 0 .01 .02 .03 RDD coefficient Amazon Non−amazon Notes: These figures show the βcoefficient and its 90% confidence interval from estimating equation 1, using mayor’s education level (indicator of having college education) as treatment variable for all municipalities with mixed-gender elections and for the subsample of Amazon municipalities. The dependent variables are the following: the average annual emissions in tons of CO2e over the four-year term in all municipalities (Panel a) and deforestation as share of the baseline forest cover in each municipality over the four-year term (Panel b). Estimates obtained using local linear estimators with a triangular kernel and optimal bandwidth based on Calonico et al. (2014). All estimates account for state and round fixed effects. 46