Using Large-Scale Network Data to Measure Social Returns and Improve Targeting of Crime-Reduction Interventions

Regents Of The University Of Michigan - Ann Arbor, Ann Arbor MI

Investigators

Abstract

Many randomized experiments have tested whether policies surrounding punishment, work, and education can prevent criminal behavior. Evaluations of prevention programs typically assume that one individual’s behavior does not affect anyone else’s. If individuals exposed to crime intervention transmit their improvements in skills, beliefs, or time use to their peers, ignoring these spillovers could understate or alternatively overstate an intervention’s total impact. This project will document how changes in criminal behavior spread through social networks by combining four existing experiments with measures of social networks derived from multiple sources of administrative data. It will quantify the direct and indirect effects of these violence-reduction interventions on participants and their peers. The results will generate a better understanding of the overall effect of each program, as well as the role of peers in shaping criminal decision-making more broadly. The project will provide actionable information on how to effectively target crime prevention programs in the future by describing whom program operators should serve to maximize net crime reduction benefits. The results will also help to establish the US as global leader in crime reducing policies. This project will estimate the effects of social networks on crime intervention programs. Estimating social spillovers faces two key challenges: measuring social networks and causally identifying peer effects. This research addresses the first challenge by combining population-wide administrative police and school records in Chicago to capture different kinds of social connections. The PIs solve the second challenge by combining the network information with exogenous variation in crime and violence generated by four existing RCTs in Chicago, all focusing on low income youth. Using these data, the PIs will construct the social network between individuals at the time each intervention was randomly assigned. The random variation from the RCTs will allow the PIs to test whether and how treatment changes behavior—both via direct participation and via indirect exposure to treated peers—as well as which types of social ties matter. The PIs will test for heterogeneity in peer effects based on demographic, criminal history, and network characteristics of individuals to help understand and model how social interactions generate criminal decisions. Finally, the PIs will us the model to evaluate alternative targeting strategies for future RCTs. The results of this research project will provide important inputs into policies to decrease crime and thus improve public safety in the US. The results will also help to establish the US as global leader in crime reducing policies. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.