Replication Data for: Political Geography of Violence: Municipal Politics and Homicide in Brazil

Violence has harmful effects on individuals and society. This is especially true in Latin America, a region that stands out globally for its high homicide rate. Building on research on subnational politics, democratization, and an inter-disciplinary literature that seeks to understand sources of violence, we examine the effect of municipal politics on homicide rates in Brazil while controlling for conventional socio-structural accounts. Specifically, we test the effect of four key political variables – party identification of mayors, partisan alignment of mayors and governors, electoral competition, and voter participation – and examine the locally varying effect of these variables with geographically weighted regressions (GWR). Our emphasis on political explanations of criminal violence is a rare departure from dominant accounts of violent crime, suggesting comparisons with the literature on political violence, and the spatial approach allows an analysis of the territorially uneven effect of political variables. The results show the statistical significance, direction, and magnitude of key political factors vary substantially across Brazil’s 5562 municipalities, showcasing the uneven effect of predictors of violence across space, and generating new hypothesis regarding the conditional effect of key predictors. In the time period examined (2007–2012), the largest left party in Brazil, Workers' Party (PT), had a beneficial effect, reducing violence in large parts of Brazil, the center party that held most local governments (PMDB) had a harmful effect in certain areas of Brazil, and the largest center-right party (PSDB) had mixed effects – helpful in some parts of Brazil and harmful in others. These results help us understand key features of the relationship between Brazilian politics and public security across different parts of the country, illuminating the political geography of violence in the region's largest country.