Police, Crime and the Usefulness of Economics

In 1994 the noted criminologist David Bayley wrote:

The police do not prevent crime. This is one of the best kept secrets of modern life. Experts know it, the police know it, but the public does not know it.

Economists were skeptical based on intuition but in truth the empirical work from economists at that time was mixed with some papers showing little or no effect of police on crime, just as Bayley argued. Since Levitt’s pioneering paper, however, there have been many papers applying a wide variety of more credible research designs like natural experiments, regression discontinuity, matching and other techniques. Any one of these papers is subject to criticism but as group the results have been remarkably consistent: police reduce crime with a 10% increase in police reducing property crime by about 3-4% and violent crime a little bit more perhaps by 4-5% (average elasticities of .35 and .48 from my review paper).

Two interesting new papers add to this literature. The University of Pennsylvania has a large private police force, some 100 officers who patrol the Penn campus and a substantial fraction of the surrounding neighborhood. The city police also police the Penn neighborhood but the UP police stay within a known (but not demarcated) region. Thus, there are more police on the Penn side of the border than on the other side. MacDonald, Kick and Grunwald apply regression discontinuity to look at what happens to crime around the border region and they find that it drops as one crosses the border. Their measures of the elasticity of crime with respect to police are similar to those found elsewhere in the literature.

Chalfin and McCrary take another approach. I always assumed that the reason standard (OLS) techniques do not pick up an effect of police on crime was reverse causality, places with a lot of crime also have a lot of police. Chalfin and McCrary argue that an even more serious problem may have been measurement error. The usual measure of police is produced by the FBI and the Uniform Crime Reports. CM find another measure produced by the Annual Survey of Governments. The two measures are close enough in levels but the relationship is surprisingly weak when looking at growth rates. Although we can’t say which measure is correct (or if either are correct) just knowing that they are different tells us that measurement error is important and measurement error will bias results downward (i.e. away from showing a significant effect of police on crime.) Moreover, if you know that measurement error exists it’s also possible to correct for it (surprisingly one can do this even without knowing the truth!) and when CM do this they find large and significant effects of police on crime, very much in line with earlier results. CM also show that there is lots of variability in police numbers that is not accounted for by crime so reverse causality is not as big a problem as one might imagine.

Using a range of reasonable elasticity estimates from the new literature and a back of the envelope calculation, Klick and I argue that it would not be unreasonable to double the number of police officers in the United States. At current levels, it’s also my belief that police are much more effective than prisons at reducing crime and with far fewer of the blowback effects. Chalfin and McCrary do a more detailed cost-benefit calculation for individual cities and they also find that many cities are severely underpoliced (and some are overpoliced–the police force of Richland County, South Carolina probably does not need a tank).

Estimates of the elasticity of crime with respect to police are largely consistent across many papers which suggests that the new techniques are more credible.  The elasticity estimates are also important because their size implies that major changes in policy could improve social welfare. I see the empirical economics of crime as one of the more useful areas in economics in which substantial progress has been made in recent years.


Comments for this post are closed