ABSTRACT
Understanding what phenomena cause delinquent and criminal behaviors or which policy interventions work in curbing crime have been the central goals of decades of crime research, not only within the blurred boundaries of “criminology.” This chapter first aims at reviewing the most important statistical approaches to estimate causal effects both in experimental and observational studies, starting from Sampson's call against golden myths concerning Randomized Controlled Trials. Second, it reasons about the convergence of econometrics and machine learning approaches for causal discovery, building upon the debate originated from Leo Breiman's “Two Cultures” argument. Third, it specifically surveys some of the most promising methods integrating machine learning and causality, with a specific focus on heterogeneous treatment effects, techniques for temporal data, agent-based modeling, and causal inference in social networks.
