Chronic Murmuring

Do Police Deter Crime?

In Becker's seminal article, "Crime and Punishment: An Economic Approach," he modeled a model of crime as a labor phenomenon in which utility-maximizing agents choose to engage in either illicit or legal labor market opportunities. He theorized that depending the level of risk aversion in the criminal, penalties, probabilities of arrests and the severity of those penalties will deter crime. This question has been empirically tested numerous times. Isaac Ehrlich, in 1973, found that capital punishment reduced the number of violent crimes and the paper - while villifying him and more than likely costing him tenure at numerous universities - found its way into the hands of the Supreme Court and helped overturn the temporary moratorium on capital punishment. One of my professors conducted a similar study involving gun control legislation. Professor David Mustard found that in states where law-abiding citizens were allowed to carry firearms, the crime level was reduced since the presence of firearms in the civilian population theoretically should represent a cost to the criminal. His chances of successfully engaging in criminal activity is reduced, and the the costs of engaging have risen, thus resulting in a negative net gain in those states. Empirically, that was what he found.

But one thing that has troubled economists for a long time has been the fact that empirically, police per capita does not seem to have the expected negative effect on crime rates. Probabilities of arrests, severity of arrests, probabilities of conviction - these and other deterrent variables will be shown to negatively affect crime rates in a region, but police per capita strangely will not. This has always been interpreted to be a problem of "endogeneity," though, and economists have not usually been found to believe the data. The problem is, police hirings are often associated with rising crime rates, and thus isolating the causal effect of police hirings on crime rates is difficult since they are usually correlated. The solution which is technically straightforward is to use "instrumental variables." Find some variable correlated with police hirings which is not correlated with crime rates, and you will be able to effectively to remove that endogeneity.

While a graduate student at MIT, Steven Levitt had been studying congressional and mayoral races, and he had noticed that when mayors came up for re-election, they hired police. It was a simple solution, yet a creative one, and when Levitt used this instrument in his regressions, he found police did indeed reduce crime rates.

All this is to give the reader some background to this classic problem, so that I can simply post a working paper from George Mason I found a minute ago. I post it because it, too, is an ingenious use of an instrument, and in their study, is found to reduce crime. They look at changes in the terror alert system issued by Homeland Security. The name of the working paper is "Using Terror Alert Levels to Estimate the Effect of Police on Crime." The co-authors are Jonathan Klick and Alexander Tabarrok. They also found there to be a negative relationship between police and crime rates.