Paulo Matos

Does police discipline deter misconduct among an officer’s peers? How does the discretion of disciplinary decision-makers shape that deterrence? I study these questions in New York City, where a civilian oversight board recommends penalties and the police commissioner makes the final disciplinary decision. In 2021, the NYPD constrained the board’s discretion by requiring it to base its recommendations on standardized penalties, while leaving the commissioner’s final authority intact. I find that a disciplinary recommendation reduces misconduct among other officers in the same command by about one-fifth, even before any punishment is imposed, and that final discipline produces an additional reduction. The reform reallocated deterrence from the commissioner’s final decision to the board’s recommendation, allowing it to operate about a year earlier. The evidence points to penalty severity: recommended penalties became harsher, while final penalties became less severe. More broadly, the findings show that the effects of disciplinary reform depend on how authority and discretion are distributed across decision-makers.

Which oversight institutions best hold police officers accountable? I develop a game-theoretic model in which a civilian oversight agency recommends discipline for misconduct, but a police chief retains final authority. I compare a fixed-rule internal discipline system with two discretionary oversight agencies: an agency that considers the cost of discipline and a more aggressive agency that does not. The most effective institution depends on the chief’s cost of review and the agency’s cost of being overruled. As either cost rises, accountability is highest under the fixed-rule system, then under the agency that considers the cost of discipline, and finally under the more aggressive agency. Thus, neither replacing fixed rules with discretion nor making oversight more aggressive uniformly improves accountability. The model yields two empirical implications. A reform can affect accountability in opposing directions across departments, so averaging its effects across departments can hide heterogeneity. Greater oversight activity can indicate less, rather than more, accountability.

How does the strategic interaction of inmates within cells influence their future recidivism, and what are the implications for optimal incarceration policies? I address these questions by developing a game-theoretic model in which inmates assigned to the same cell first decide whether to cooperate and, subsequently, whether to reoffend after release. The benefits and costs of cooperation are endogenous: cooperation allows inmates to expand their criminal networks through knowledge transfer but also exposes their future criminal enterprises to their cellmates through information leakage. As a result, recidivism decisions become strategic substitutes between cellmates in equilibrium. The model yields three main findings. First, policies aimed at reducing recidivism, such as improving inmates’ economic conditions or increasing the cost of reoffending, are more effective when cooperation occurs. Second, when inmates differ in criminal skill, segregation minimizes recidivism when the low-skilled inmate’s detection probability is close to that of the high-skilled inmates, whereas non-segregation becomes optimal when he is sufficiently more likely to be detected. Third, when inmates differ in their legitimate outside opportunities, non-segregation minimizes recidivism when the low-opportunity inmate has sufficiently limited outside opportunities, whereas segregation becomes optimal as his outside opportunity approaches that of the other inmates.

We exploit an exogenous government policy in El Salvador that reallocated several high-ranking gang leaders from maximum-security prisons to lower-security facilities between 2012 and 2015, combined with highly detailed administrative incarceration data spanning more than a decade. Our key finding is that inmates exposed to these gang leaders within the first days of a new cell assignment exhibit higher probabilities of future gang-related recidivism, but not of non-gang-related crimes. We identify the transmission of criminal capital from gang leaders to other inmates as the main mechanism driving the increase in recidivism. Moreover, we find that the exposure effect is amplified among inmates with prior gang-related offenses—such as homicide, gang affiliation, or extortion—but is not affected by the pre-existing gang composition of the cell.

Community engagement policies are widely used to reduce crime by strengthening coordination between residents, police, and local service providers. This paper studies whether such policies remain effective in socially segmented settings. We exploit the interaction between a common criminal shock—the consolidation of Tren de Aragua by early 2022—and fixed cross-sectional variation in prior exposure to a local community engagement policy across police jurisdictions in Peru. We estimate difference-in-differences and common event-time specifications separately for municipalities with and without significant Venezuelan migrant presence. We find sharply divergent effects. In municipalities with significant Venezuelan presence, the policy increases extortion after gang consolidation by roughly 100% relative to the outcome mean. In municipalities without significant Venezuelan presence, the estimated effect is negative, with consistently declining post-consolidation patterns in the dynamic specification. We find no comparable divergence for non-gang-related outcomes. We interpret these findings as consistent with a mechanism in which participation requirements strengthen protection for socially integrated residents while leaving migrant populations more segregated and vulnerable to gang recruitment. More broadly, the results suggest that community engagement policies may backfire when access to protection is unequal and minority populations remain excluded from local institutions.