Crime Statistics Analysis: how to properly compare apples to apples? first this is non-partisan. Now, I have heard a lot of people in the news using certain simple statistics such as the number of uses of lethal forces by cops particularly where the victim is a minority to draw the conclusion that there is excessive force and racial bias. I have also heard on Planet Money a comparison to the use of lethal force by law enforcement in the US vs in Britain (https://www.npr.org/transcripts/871298161). Is this really a sufficient analysis from which to draw conclusions? If not what is? Don't we need to ensure we are evaluating the data:

*

*per capita when comparing separate regions

*removing incidents where lethal force was necessary (e.g. a shootout) from the dataset

*identifying the probability of minority vs. non-minority interaction for an officer given a location's crime statistics and racial statistics

[Again, this is non-partisan, but I would like to see a much more rigorous analysis. I understand that incidents like Floyd's and Eric Garner are abhorrent and should not go unpunished- plus the subsequent defense of these people by the union or fellow officers is at least an indication of a bad system. Furthermore, there seems to be lack of available data sometimes.]
Any links to the type of data analysis I am searching for would be great. Thank you.
 A: The objections you mention are certainly noteworthy considerations. A lot of the issues with statistics on issues like this stem from an endogeneity problem: the error term of the model is correlated with the independent variable, such as race. Specifically, I think it is easy in this case to fall prey to the missing variable bias. That happens when there are unobserved variables (population size, interactions with the police, criminal behavior, mental illness, income, geography) that are related to the response, but not included in the model. These variables may correlate with race in which case their effects are erroneously assigned to race.
If you were to examine police shootings in the United States, and compare cases of white victims and native Hawaiian/pacific islander victims without accounting for anything else, you would attribute the effect of a much higher population to race.  You might conclude that police shoot white people at 300 times the rate at which they shoot native Hawaiin people, but you could not conclude that the effect was due to the difference of race. That example is an oversimplification, and any honest statistic should account for the most obvious covariates such as population size.
However, the process of determining exactly what to control for is complicated and surprisingly subjective. If you look at this paper: https://www.pnas.org/content/116/32/15877 you'll get a picture of a scientific approach to estimating the degree of racial bias in police shootings in the United States. And, if you look at the letters/corrections at the top, you'll also get an idea of some of the controversy involved, and at least one example of how small changes in the analysis can yield conflicting results.
In any case, you can be sure that a single statistic cannot truthfully tell a story as complicated as racial bias in police brutality. And yet, you can be just as sure that they will be used to tell stories, many of which were calculated before the statistic.
A: You can read Roland Fryer's work in the respected Journal of Political Economy: https://scholar.harvard.edu/fryer/publications/empirical-analysis-racial-differences-police-use-force. He finds that

On nonlethal uses of force, blacks and Hispanics are more than 50 percent more likely to experience some form of force in interactions with police. Adding controls that account for important context and civilian behavior reduces, but cannot fully explain, these disparities. On the most extreme use of force—officer-involved shootings—we find no racial differences either in the raw data or when contextual factors are taken into account.

However, Fryer also notes several limitations to his analysis, including that

all but one dataset was provided by a select group of police departments. It is possible that these departments only supplied the data because they are either enlightened or were not concerned about what the analysis would reveal...There may be important selection in who was willing to share their data.

Clearly, more research on this topic is necessary.
