Terror Alerts, Police and Crime

If you look at a simple correlation between crime per capita and police per capita you get this:

Police cause crime!  Some of the Foucault-inspired may buy into that conclusion but it's really no surprise that places with a lot of crime have a lot of police.  Estimating the true effect of police and crime from observational data is difficult because police and crime are determined jointly.

Ideally, to estimate the true causal effect of police on crime, we would run an experiment, randomly picking weeks in which we increased police presence and observing the effect on crime.  Experiments like this, however, are expensive and some might say unethical.  If we look carefully, however, we might find natural experiments - times when police presence increased for reasons that are random with respect to crime. 

Jon Klick and I look at just such a natural experiment in a paper published in the most recent JLE, Using Terror Alert Levels to Estimate the Effect of Police on Crime (subs. required, free version).  When the terror alert system kicks up a notch the police in Washington, DC put more police on the streets.  We find that crime in DC drops significiantly during these high-alert periods, especially in the National Mall area where most of the prime terror-targets are located.  Street crimes like auto theft and theft from automobiles show especially large decreases when more police hit the street.  We find no evidence that tourism or other demand side factors account for the decline in crime.