1/n I believe I've largely replicated Fryer's (2019) non-lethal force findings using the 2015 Police-Public Contact survey (his data was older, running from 1996-2011).
2/n First, the base-rates across each racial category for self-reporting officer-use of force show the expected pattern: blacks and Hispanics report that an officer used force at ~2.5 times the rate of whites. Asians reported it a somewhat lower rate than whites (p<0.1) 
3/n Keep in mind that these base-rates are extremely low + the data is limited to respondents who had contact with police over the previous 12 months (which means these aren't the base rates for ANYBODY reporting the use of force, but the minority of the sample that had contact)
4/n They're also self-reported, which introduces the possibility of respondent bias (hence, these might be liberal estimates).
5/n Anyways, if you check out the middle panel, you'll see that the controls added to the model largely wipe out the differences.
6/n In other words, holding all these variables constant, blacks/hispanics are only a bit more likely to report the use of non-lethal force as whites (and the difference with whites only reaches conventional levels of significance for hispanics; p < 0.05)
7/n Finally, in the 3rd and last panel, I tested a respondent race x officer race interaction. Similar to Fryer (2019), I find that non-white officers are just as likely to use non-lethal force against non-whites as white officers.
8/n I suppose a major difference between my and Fryer's analysis is that he combines multiple datasets, which results in greater statistical power. It's thus very possible that the differences would have remained at conventional levels of significance with more observations.
9/n I will try adding the earlier datasets when I have the time. But with that being said, I would expect to find the same general pattern of results even with more data.
10/n For those wondering, the dependent variable--officer use of non-lethal force--is coded as 1 if the respondent reported any of the following, and 0 if otherwise
11/n Self-reported non-compliant/uncooperative behavior is an index consisting of the following (which, for convenience sake, I took from Fryer's appendix)
12/n So is the above evidence of systemic racism? If it is, it's not very impressive. But you be the judge.
13/n I should belatedly add that the above in no way disposes of the question of racial bias in police use of force. Much more research needs to be done before one can conclude with any degree of certainty one way or another. But herein lies an important point...
14/n In declaring that there is 'sytemic racism' in policing, media and political elites have already 'settled' the question for us. They derive strong and sweeping conclusions that are not warranted by the existing (and often poor) data.
15/n Update: So I combined the 2011 and 2015 data and re-ran the models with a new control for whether there was more than one officer present at the scene (as well as year). The results are largely similar, with a few small but still noteworthy differences.
16/n The controlled model in the original graph left a ~0.008 difference in the black and white odds. The latest model sees this gap narrow further to 0.005 (which is insignificant, p=0.26), and the one with hispanics from 0.009->0.006 (p=0.04)
17/n As before, Asians are less likely than whites to report use of force, with this difference now reaching conventional levels of significance (p=0.02).
18/n Surprisingly, and turning to the interaction model, whites are now slightly but significantly (p <0.01) more likely to report the use of force from a white officer than a non-white officer.
19/n If anybody is curious, the rate of reporting use of force for those who had contact with police did not significantly (p=0.176) change from 2011 (0.0244) to 2015 (0.0287). Nor were there any significant changes by race.
