What is bias?

A correspondent asks, citing this quote from this article

"To demonstrate the existence of a racial bias, it’s not enough to cite the fact that black people comprise 14% of the population but ~35% of unarmed Americans shot dead by police. You must control for confounding variables (in this case, the effect of a suspect’s race on a cop’s decision to pull the trigger)."

Does he mean the confounding variable is the race of the cop, compared to the race of the person who was shot?

And in that case, are white cops killing white people, and black cops killing black people?

Or, is it that the overall number of encounters with cops is higher for black people? In which case, isn't that still an (indirect) result of systemic racism ?

To which I answered:

So. It’s the correlation vs causation issue. And it’s quibbling over what one may call “bias”. In my mind you can demonstrate bias just by the crosstab and controlling for nothing. What that doesn’t show is whether or not being black was the causal factor (either directly or indirectly). There may be things that are associated with being black that are the true causal factors. That all being said, people have actually controlled for all these things and found that yes, being black was a causal factor

There are two things at issue here:

1. What can we call “bias”? Does race have to be the direct effect of a police shooting in order for there to be bias, or if race causes something that is, in turn the cause of the shooting — is that also bias?

2. Where, actually, is the causal effect?

For (1), I don’t think one needs to actually control for confounds in order to demonstrate the existence of bias. Controlling for confounds does help determine where the bias is caused, but not whether or not it exists. Is the problem because cops are racist, or because being Black puts one in situations where one (regardless of race) is more likely to be shot by a cop? We could call the latter systemic racism, which is what many people do.

For (2), well — this has been studied, and I’m going to just leave you with two papers that investigated this in depth:

Analysis of Racial Bias in Police Shootings

A large-scale study that found the risk of being black, unarmed, and shot by police, is about 3.5x the same risk for white Americans.

Analysis of Racial Disparities in Police Stops

This study uses an ingenious natural experiment by analyzing traffic stops right before and right after sunset. Relative stops of black Americans go down after sunset (when it becomes harder for police to see who they’re stopping).