My dissertation collects four papers investigating changes in political behavior in violent and unequal societies. These four independent, but theoretically interconnected papers, work around three main questions: how do citizens form preferences about security policies when stressed about risks of crime victimization? How does exposure to crime interact with income differences to explain citizens' preferences for police allocation and voting behavior? How do these concerns ultimately enter into the electoral arena via support for candidates campaigning on tough-on-crime policies? I integrate these questions across my four papers with a general theory considering both the micro-level dynamics behind preferences for security policies, and the supply of politicians framing the menu of security policies available to voters. My first chapter brings together fine-grained observational data and an endorsement experiment to understand the effect of crime victimization and partisanship on voting for law and order candidates for legislative elections in Brazil. My second chapter develops an insurance model to explain preferences for crime deterrence policies and uses a behavioral experiment to assess the model's empirical implications. The third chapter uses computational text analysis on a corpus with more than one hundred thousand Congressional speeches to discuss issue ownership and how politicians use their professional history in law enforcement agencies as informational heuristics about their security preferences. The fourth chapter uses novel network models and a conjoint design to uncover the effects of exposure to criminal violence on citizens' preferences in Mexico using a conjoint design. Chapter one shows that a local exogenous crime shocks right before the election increases the vote share of law-and-order candidates in cities more afflicted by violence. This effect is only present in municipalities with more robust support for more conservative presidential candidates and driven mainly by wealthier voters. Experimental results converge with macro-level results. Chapter two's main finding shows with experimental data that income and fear of crime follow a positive joint distribution, making wealthier respondents with high fear of crime more supportive of greater levels of police allocation on high-crime and low-income geographical areas. Chapter three shows that occupation on law enforcement explains which politicians "own" the issue of security in the Brazilian Congress. To conclude, using a conjoint design, chapter four finds that higher exposure to crime using network information increases support for punitive policies and candidates previously employed in the local police forces. Chapter four's findings combine new models to measure crime exposure, using information from respondents' friendship networks and a conjoint candidate-choice design. My dissertation contributes empirically and theoretically to deepen our understanding of preferences for protection in violent and unequal democracies. My most general result provides observational and experimental evidence for a positive joint distribution between income and risk. This dynamic explain how wealthy voters in Latin America form the main electoral and social support behind the emergence of populist, iron-fist politicians in the continent.