Heterogeneity in the Effect of Federal Spending on Local Crime: Evidence from Causal Forests

Abstract

Federal place-based policy could improve efficiency if it targets areas with large amenity or agglomeration externalities. We begin by showing that positive shocks to federal spending in a county and their associated economic stimulus substantially decrease crime, an important amenity. We then employ two machine learning algorithms---causal trees and causal forests---to conduct a data-driven search for heterogeneity in this effect. The effect is larger in below-median income counties, and the difference is economically and statistically significant. This heterogeneity likely improves the efficiency of the many place-based policies that target such areas.

Keywords: place-based policies; amenities; machine learning; crime

JEL Classification: R1, H2, R23