Steven Levitt's "Using Electoral Cycles in Police Hiring to Estimate the Effect of Police on Crime": Replication and Extensions

Abstract

Are police effective in maintaining public order and safety? Levitt (1997) approaches this question through the use of instrumental variables. Using election cycles as instruments for changes in police, Levitt finds that increased police presence causally reduces violent crime but does not have an effect on property crime. We replicate Levitt’s results with the inclusion of the corrections suggested by McCrary (2002). We also consider the non-parametric bootstrap method to estimate standard errors in addition to 3SLS. Our replication results indicate that elections may be a weak instrument for changes in police, making our estimates biased.

We then investigate the effectiveness of police on crime using an unconventional theory. Most empirical literature seeks to show that police are effective by reducing crime levels in the long-run. However, we argue that short-run effectiveness should instead be measured by increases in the number of crimes reported: increases in the number of police on patrol consequently should increase the number of crimes reported. Using regression discontinuity, event study, and differences-in-differences (synthetic control method) designs, we take police academy graduations as an exogenous shock to police levels and find that this causes a short-term increase in reported crimes. We also verify these results using monthly crime data around mayoral elections in Chicago, finding that reported crime levels increase in the months after mayoral elections.