Thanks for the link, that's an interesting and useful article. Updated my probabilities in a few areas. Also an amazingly civil, rational comments section given the nature of the material.

It didn't help me very much on police shootings though - lot of way-out-of-date data that points both ways depending on area. No racial bias in shooting recorded in New York, seriously significant effect in Tennessee. Given what I know about American racism-by-state that's almost disappointingly obvious. (Again though, out of date - relevant laws in Memphis have changed since).

Okay, strap in, this is a long one. First I'm going to cheerfully steal a chunk directly from one of those comments on SSC for your consideration:

Suppose that you notice that on average, green shows up twice as often as red, but you can’t see a pattern to it. If you want to maximise your winnings, should you on average bet on green twice as often as red to match the frequencies you’re seeing? No, you should strictly bet on green every time.

(Anyone know where I might have read about this before? Pretty sure it’s somewhere on LW, but I can’t find it.)

Similarly, if a random black person is statistically more likely to be a criminal than a white person, then a police officer’s or prosecutor’s career incentive is to focus on them.

Of course this wouldn’t fly as a practical policy. Green and red lights may be independent from your choices, but humans are not. If you completely removed police overwatch from non-black populations, this would encourage crime in those populations. Even if you didn’t care, there’s no way you’d defend yourself from accusations of blatant racism and unfairness.

Still, the incentive is there. And it’s based on math – racial prejudice not required.

Because I believe people tend to follow incentives, my current best guess is that police do over-profile (target the higher risk groups more than the actual risk differences would suggest), and they are going to, and the only question is to what extent this can be mitigated.

I'd be interested to see what you think of that.

Now, speaking for myself -

I find no flaw with your buggy-robot analogy. That would indeed in result in more shootings of innocent black people without any need for racial prejudice influencing the decision to shoot, simply because of disproportionate exposure given higher crime rates.

My contention, however, is that racial prejudice is a factor in real-world police shootings/violence. So the question I ask as a half-decent wannabe rationalist is how my belief should constrain my expectation - how do I expect my world to look different from a non-racist buggy-robot world? How do I test it? Honestly I don't have a particularly satisfactory answer. Some attempts follow, feel free to skim if you're not arsed reading them.

1. I would expect the disproportionate impact of police shootings/violence upon innocent black people (innocent meaning unarmed/non-dangerous here, not necessarily innocent of any crime) to be measurably higher even when adjusted for the higher crime-rate/residential grouping effect as it was in the Memphis study. (Though not to such a degree - important to state I do believe that this problem is getting better). However, the way a shooting event or other incident of violence is recorded depends very heavily on the word of the officer involved (the event is assumed to be "an "assault on law enforcement" and the officer is referred to as "the victim" in the report on a fatal shooting by default). If racial prejudice influenced his/her decision to shoot it could also - with or without any deliberate lying - influence his/her assessment of whether the individual shot was behaving as a threat. (You'd be amazed what some police officers will call "assault on an officer" or "resisting arrest" with a straight face - that's a problem even without touching any racial-disparity issues). So that's also in my model, and the fact that the two effects counter each other means they're of little use to me as a measurable anticipation-constraint.

1. I'd expect a higher impact on black people of what I'd call "WTF shootings" - shootings where the victim could not have been deemed a threat by a reasonable observer. Unresisting arrestees, fleeing suspects holding nothing in their hands, kids holding toy guns shot before being given any chance to comply with verbal directions - or being given no verbal directions. Not tragic-but-understandable mistake type shootings - "itchy trigger finger" shootings that baffle reasonable explanation and appear to proceed directly from some kind of gut feeling on the part of the officer.

Interesting to think of this in relation to what seems like an odd number of reports of police shooting securely-tethered pet dogs that barked at them. I've seen an actual cop try to explain this phenomenon by saying that police often have terrifying, dangerous encounters with vicious guard dogs owned by drug dealers and the like, and develop a fear that leads them to react with instinctive aggression to a barking dog without taking the time to evaluate whether or not it's a threat. Interesting, that. I have no data for it though, so just an idea.

Anyway, I'd expect these "WTF shootings" to hit black people harder, but one can only apply the "reasonable observer" test if the incident is recorded on camera or there are a decent number of witnesses - and in this case I'm willing to admit that the political heat around this issue might lead to WTF shootings of black people being *over-represented" or identified where they don't exist. So, measurement problem here again.

1. Police shooting disproportionately affecting black people even when you only count the shooting events that occur in locations that mitigate the grouping effect. Put simply, your robot that shoots innocent people is disproportionately likely to hit a black person because it's in a black neighborhood interacting with black people for a disproportionate number of hours of the day. But real policemen are assigned to patrol specific areas - the racial mix of the people they interact with is governed by the demographics of their "beat", not overall crime stats. There are whole all-white towns in America. Pretend for a second that the crime rate among black Americans is four times that of white Americans. Now imagine a police officer in a neighborhood - or city, even - with a 3% black population. Adjust for higher crime rate and their interactions with black citizens go up to 12% which neatly matches the actual demographics (if I remember the figures.) That could be an adequate sort of "controlled environment" where interactions mirror actual population demographics. If black people are disproportionately shot within that neighbourhood, I'd say that's a measurable indication that racism is playing a part. If they aren't, it indicates the opposite.

But after all that, I just don't know where to find current objective data or how to look at it, and at the end of the day I'm looking for something that is capable of covering its own tracks. I'm not quite at "no sabotage is evidence of Fifth Column" yet but I'm brushing dangerously close to a universally applicable argument. Racism not evident in data? Data could be skewed by racism! Sounds dodgy but common sense says it is possible and has happened before and I can't discount it. But given my difficulty making my beliefs pay rent, I've revised my certainty down a bit just by writing all this out.

But not a lot down and here's why - the other side - the part two of this ridiculously long comment.

A certain percentage of Americans are racists. Lots are a teeny bit racist (arguably we all are), but a few are massively, viciously racist. This isn't distributed equally over the states or within them - there are clusters. If American police are a fair sample of the American population, then many of them are a wee bit racist and a few are massively racist, and there are clusters in certain areas. (By the way, I wouldn't be overly surprised if American cops were less racist on average than Americans as a whole. That still leaves a goodly few "bad apples"). How could that not impact their treatment of minority-group individuals? What negates the effect of that bias in a given situation? I'm willing to accept the impact could be neutralized to a large extent by complex structures and redundancies within, say, the court system - but in the personal, individual, encounters - split-second decisions whether or not to shoot, whether or not to resist the impulse to kick someone in the head while they're on the ground, what check is there? You can say that the buggy-robot is a simpler explanation, but to me it's just a shorter one. The more complex idea, as I see it, is that somehow there's far less racism among police officers than among the genpop, or that somehow the racism there is is prevented from impacting its targets in situations where no apparent check is provided. The absence of racism as a motive force in any instances of police misbehavior or misjudgment would need explaining to me.

I'm done. Sorry about the novel, it's been a slow day at work.

TFL;DFR - the evidence is complex, patchy and difficult to interpret but doesn't appear to be stronger for my position than the converse; however cops are people, some people are racist, therefore some cops are racist, and cops have a lot of discretion as to how and when it's appropriate to use physical force which means some whacking great racists make decisions about whether or not to shoot or brutalize black people, and I don't see how that can't equal disproportionate impact, at least in certain states or areas.

My reasoning on some parts is probably lousy with holes, so if you've slogged through this far, have at it is with a hatchet, and if you haven't, I don't blame you.

Is it fair? No. Is it racist? Also no.

Agree, and I think this is a really important and overlooked implication, that two tribes will talk past each other on. Unfair discrimination persists even with rational, non-racist, greedy capitalist.

A less charged example would be life insurance policies. Almost everyone would agree that mortality tables are acceptable; almost everyone could also imagine themselves getting older, and could imagine themselves as above average with in their group. The insurer will rationally charge the older group more premium. Atypical healthier older people within this group have experience unfair discrimination and the insurer is rationally non-prejudiced.

So when one tribe says that markets will punish racist, it doesn't fix unfair discrimination. And when other tribe says that there is unfair discrimination, that doesn't mean there is rampant racism. I personally feel a lot of compassion towards atypical individuals within a disadvantaged group, but how could we improve?

I don't know if this is the right place to have this conversation but I can't help myself. Mods - feel free to kill this.

Disclaimer, I'm not American. I don't have a dog in this fight one way or another, but I can pattern-match.

People object to "All Lives Matter" because it derails the discussion and implies that it's somehow unfair to focus, as you said, on "Whichever Lives Are Most Affected By Police Brutality At The Moment" - which in America means black people specifically. It's the same reason people object when a discussion of sexual violence is cluttered up with comments insisting that everyone recognize "women can commit rape too!" or when a discussion of social discrimination faced by disabled people meets a response like "able-bodied people can be bullied too! I was bullied for being ginger!". I've seen that kind of "what about me" response in a dozen different forms and it's almost never useful. It's a cry of not fair along the same lines as "Why can't we have a Straight Pride Parade?" "Why isn't there a White History Month?" and so on and so on.

Nobody was tweeting "All Lives Matter" before "Black Lives Matter". It's not the slogan of any particular group or movement. It's a response, and a clear implied criticism. While I wouldn't go so far as to say it's inherently racist, I'm not surprised in the least to see that motive attributed to it. If I was American I'd certainly be objecting to it too.