More geography of police shootings

In July, I had a post on state differences in the rate of fatal shootings by police, which differ by a lot--the rate in New Mexico or Wyoming is about ten times the rate in New York and Connecticut.  Peter Moskos, a professor at John Jay College of Criminal Justice, also noted the regional differences (I got the reference from Andrew Gelman's blog) and suggested that they resulted from differences in the training and procedures in different police departments.  That suggests that we should go to the level of individual cities.  I took the 100 largest cities and calculated the expected number of deaths if they were proportional to population (which comes to about 1 per 135,000 in the period covered by the data).  The deviations were statistically significant by any standard you want (chi-square of about 267 with 99 degrees of freedom), so the differences in the rates are not just a matter of chance.

The cities with the highest ratio of actual to predicted deaths:

                Deaths  Predicted
Miami              13       3.2
San Bernardino      5       1.6
St Louis            7       2.3
Orlando             6       2.0
Baton Rouge         5       1.7
Bakersfield         8       2.7
Las Vegas          13       4.6
Reno                5       1.8
Norfolk             5       1.8
St Paul             6       2.2
Albuquerque        10       4.1

For the lowest ratio, it's a tie among eight cities--Hialeah, Irvine, Jersey City, Lexington, Lubbock, Plano, Riverside, and Winston-Salem--which had no fatal shootings.  Those are all in the lower reaches of the top 100 in population, and the expected numbers are about two in each.  With those expected values, a zero can easily come up by chance, so we can't be sure that the actual risk in those cities is actually different from the average.  But the next two lowest ratios are in big cities:  New York, with eight actual deaths and 63.6 expected, and Philadelphia, with three actual and 11.5 expected.  Those differences definitely cannot be attributed to chance.  

There seems to be a lot of geographical clustering--New York and Philadelphia are less than 100 miles apart, and the #41, 44, 45, 46, and 47 cities are all in Texas.  Maybe there is some general cultural similarity in regions that makes a difference, or maybe police departments just tend to model themselves after departments in nearby cities.  But whichever it is, there is something that needs to be explained.