Police Shootings Revisited

This time last year, I published a blogpost that examined police shootings in America. In it, I tried to determine whether police shootings of black people outpaced their expected value in terms of their population share from state to state. I found that the majority of states were within the expected distribution, i.e., in 33 states there was insufficient evidence to suggest that the police were more likely to fatally shoot black people than anyone else. However, the remaining 17 states (my home state of Minnesota among them) showed evidence of racial bias. That is to say, far more black people were shot than could be explained by randomness alone. Additionally, in the most concerning case — when the victim was unarmed, not resisting, and not fleeing the scene — black victims constituted a plurality of the cases. They not only outpaced their population share but were the most common victims of police shootings under those conditions.

Since it’s been a year† I thought it would be a good time to revisit the question and see how we’ve progressed in this area. Despite the best efforts of Colin Kaepernick and his like-minded cohort, it has been my impression that this issue has gotten significantly less press attention in the last year than in the year that preceded it. Is this a reflection of real, quantifiable improvement in this problem? Or have we simply turned our attention to other things?

The results, as one might expect, are something of a mixed bag. Nationwide, the shooting of black people is down roughly 8.5%, while every other category is up slightly. In total, this amounts to an extra three fatal shootings per month, or about a 4.2% rise from the preceding period.

Fatal Shootings Per Month

It’s not immediately clear what’s driving the increase in police shootings. I compared state per capita police shootings to per capita violent crime and there is a loose correlation between the two; however, one would expect states that have seen an increase in violent crime to also show an increase in police shootings, but that does not appear to be the case:

Police Shootings by Per Capita Violent Crime

Police Shootings by Change in Violent Crime

Perhaps these results would be more meaningful if examined on the city level. I am unaware of a city level violent crime rate data set.

When segmented by state, the picture looks quite similar to last year. Every state that showed evidence of racial bias last year continued to show it once the next year of data was incorporated. Moreover, three states and the District of Columbia moved from within the expected distribution to borderline status.

Iowa (n=13, p=.027, k=4), for example, is just barely off the expected distribution, though its small sample size doesn’t inspire compelling conclusions be drawn from its example:

Iowa Example

Washington D.C. (n=11, p=.490, k=10) is similar to Iowa in this respect. D.C. only had three fatal police shootings in 2017, but all of the victims were black men.

DC Example

North Carolina (n=75, p=.216, k=28) has a sample size that justifies its placement among the problematic states. Twelve of North Carolina’s twenty-five victims of fatal police shootings in 2017 were black.

North Carolina Example

Washington state (n=70, p=.037, k=11) had a surprising number of total shootings. Based on national averages, one would expect to see about 30% fewer such shootings, or about 20 (!) fewer fatal shootings since January 2015. Though black victims made up a somewhat smaller proportion of the total shootings in Washington, they well outpaced the population share.

Washington Example

California (n=457, p=.067, k=72) continues to be the worst offender in the country, killing the most of every race and significantly outpacing its population share. In fact, if California’s total shootings were simply the difference between its actual shootings and its expected shootings by population size, it would still rank second in the nation.


It should be noted, of course, that police shootings are not randomly distributed by state, nor are they consistently proportionate to population size. California, Texas, Florida, and New York are the four most populous states in the country and they rank 1st, 25th, 34th, and 51st respectively the rate of police shootings. An important aspect of this conversation is to determine how, exactly, to make California, Arizona, and Oklahoma more like New York, Pennsylvania, and Michigan. (To the best of my knowledge, no one has studied variance in police shootings by state.)  Racial parity here is a laudable goal, but it will be incomplete — and of small comfort to grieving families — if it is not also accompanied by a rapid decline in overall police shootings.

Total Shootings minus Expected Shootings

It’s not all bad news. While states such as Washington, Ohio, Florida, and Missouri are killing more people overall, states like Texas, New Mexico, Nebraska, and Colorado are killing fewer. Texas, for instance, averaged reduced their rate of fatal shootings by more than one per month over the course of 2017. This, unfortunately, was not enough to offset the gains in states like Washington, Ohio, and Missouri. Perhaps closer examination of either extreme — with a contrast with the states that saw little or no change — would provide some insights.

Change in shootings.PNG

Furthermore, states like Nevada, Alabama, and Kentucky were among eleven in total that significantly improved their performance with respect to racial bias. Nevada, for example, didn’t kill a single black person in 2017, despite the black population of Nevada hovering around 5%.

Lastly, the category I referred to last year as the “most insidious form of police shootings” — when the victim was unarmed, not resisting, and not fleeing the scene — has inched closer to the expected distribution. Black people are no longer the plurality of victims of this category. This superficially positive change masks the devastating fact that the number of this type of shooting has skyrocketed: 35% more in the last twelve months than in the twenty one months that preceded it.

Unarmed not resisting not fleeing

On the whole, I would argue that there is reason for optimism. While it’s unclear what is driving the changes that are reflected in the data, the overall picture is one that increasingly matches the population demographics of our country. (It should be pointed out that this facet is understated in this analysis as the country has gotten less white since the last census, meaning the population demographics used here would make us more likely to find evidence of bias than if more recent figures were available.) Whether this change has been driven by new policies, by increased media attention and protesting, or is just an artifact of having a more robust data set, that aspect of the trend seems to be positive. No state is perfect, and there is a lot of work yet to be done, but the data suggest that, with respect to racial bias, we have taken a step in the right direction. Police are killing more white people, however. Perhaps this is tied to the recent spike in the crime rate, but the data are unclear on this point.


† Every time “2017” is used in this post, I am referring to the span from October 10th, 2016 to October 11th, 2017, and “2015-2016” refers to January 1st, 2015 to October 9th 2016. Apologies for the confusion: this choice was made to avoid making the date references more cumbersome than need be.