| HN Mirror

Me. Racist. Against white people. I think you might need more practice at your local comedy club's open mic night.

What I said was, there was a hypothesis, it made a testable prediction, and measurements support that hypothesis. This is sometimes called evidence.

I then pointed out that "proof" is not a simple concept. Quoting https://en.wikipedia.org/wiki/Proof_(truth) :

> The concept applies in a variety of disciplines,[5] with both the nature of the evidence or justification and the criteria for sufficiency being area-dependent ... in jurisprudence the corresponding term is evidence,[10] with "burden of proof" as a concept common to both philosophy and law. ... Exactly what evidence is sufficient to prove something is also strongly area-dependent, usually with no absolute threshold of sufficiency at which evidence becomes proof

I then showed examples of how people use "proof" for something more like evidence than, say, how it's used in mathematical logic, including cases where the proof can be wrong.

Now it's certainly true that there are many ways to interpret the data. That's why I wrote "Some might say this is evidence which supports multiple hypotheses". You are in that group. And there's nothing wrong with that.

I actually agree that it supports multiple hypotheses, but I haven't looked at the data or any of the research on the topic. I haven't done the analysis. I am making the more pedantic point that it's not outside the bounds of established use to say "prove" here.

Then again, it doesn't sound like you have done the analysis either, and you are reacting on the basis that you don't like the conclusion and you think "proof" only means something like "beyond reasonable doubt".

WP helpfully reminds us that in US law the lowest legal burden of proof is the far "reasonable suspicion". If you believe the data doesn't even mean that low burden of proof, then I certainly disagree.

> What I said was, there was a hypothesis, it made a testable prediction, and measurements support that hypothesis. This is sometimes called evidence.

If the hypothesis is "police are racist against black people when deciding to use force", then no, measurements don't support that hypothesis at all or too weakly to reject the null hypothesis -- we haven't measured the coupling to demographics to get a good prior estimate on the number of deaths. Without that, we need to sort of "average" over possible couplings, and end up with something like 0.5 to 2.5 times "appropriate number of blacks to number of whites being killed by police". Since that range includes 1, until we can refine the coupling between shooting distribution and demographics, we basically have a null result on our hypothesis.

> Then again, it doesn't sound like you have done the analysis either, and you are reacting on the basis that you don't like the conclusion and you think "proof" only means something like "beyond reasonable doubt".

> WP helpfully reminds us that in US law the lowest legal burden of proof is the far "reasonable suspicion". If you believe the data doesn't even mean that low burden of proof, then I certainly disagree.

I don't disagree that there's a "reasonable suspicion" of racism affecting the outcomes of policing -- but I think that's been the case for, oh, the entire lifetime of the country.

I disagree that there's a "preponderance of evidence" (which I'd argue is the lowest standard to reach a conclusion besides "we should check more" or "there isn't enough evidence to tell") that the dominant cause of the high number of black deaths is racism, that we even have a number of black deaths which are in some sense "in excess" of what we'd expect from "fair" or "not racist", and think people are likely a bit more... uh... "reasonable" than I am if they think there's a "reasonable suspicion" that racism is the dominant cause of the discrepancy between black per capita and white per capita death-by-police rates.

I'm not, however, against looking in to it more. People should totally do that, and I bet they'll get their name in some sociology book for proving the racist component exists.

I just think, you know, we should talk about the fact that police kill twenty people per state every year, rather than the few people in the entire country that happens to who happen to have it happen just because they're black. That is, perhaps we should talk about how POLICE ARE JUST BEATING AND SHOOTING TONS OF PEOPLE UNNECESSARILY, rather than about how we should fix it so that they do so to the same number of white and black people, who once you account for confounding factors, seem to get abused at a roughly even rate already (eg, +/- 10%).

I think that the "evidence" backing the claims hasn't even cleared basic Bayesian analysis, and is a case of people jumping on an emotionally charged things that sounds good.

It's not that I need 5-sigma to accept a social issue, it's that I need proponents to at least pass the basic threshold of "entirely statistical artifact". And right now, the rate at which blacks are being killed relative to whites is anywhere from 0.5-2.5x, which because it includes 1, is only a slightly skewed-basically-null result.

So with the police-are-racists claim hovering at "statistical anomaly", but the police-are-violent-psychopaths claim well documented, I think people who are focusing on the "racism of the police" are... well... misguided.

There's a very real, well-documented issue. It's just seems that it's a complex classism one, not a racism one, which is way less sexy to talk about.

I guess for you it's way more sexy to call me a racist against white people.

Well, yes.

But if you want to address one rhetorical flourish in calling you out for assuming a link between police shootings and demographics which supported your position without addressing equally valid and likely links which didn't support your position as "racist" merely because it involved you assuming that one race (by virtue of being that race) was the victim of racism, when the data suggests that it's possible either is the victim of racism, then you are correct, it's not strictly speaking racist.

But you're well in to very technical arguments about why that's not racist.

http://mappingpoliceviolence.org/

mcguire 3550 days ago

According to the Mapping Police Violence site, fewer than 1/3 of the black suspects shot were suspected of a violent crime or armed. Randomly shooting people who match crime demographics seems a poor strategy.

[1] https://www.washingtonpost.com/graphics/national/police-shoo...

tzs 3550 days ago

It was way more than 1/3 according to the Washington Post's database [1]. They have 258 black people shot by police in 2015 (out of 991 people total shot by police). Of those, they categorize 183 as involving "Attack in progress" (71%).

They have 63 categorized as "Other". I'm not sure what counts as "Attack in progress" because looking at those under "Other", it looks like about 40% [2] of them involve the suspect charging at officers with a knife, or threatening officers or a third party with a knife and refusing orders to drop it, or trying to run down officers with a car.

That fits in with what I've seen when I've picked a random sample at killedbypolice.net and sorted them into "justified" and "unjustified" piles. There I got something around 80-90% seemed reasonably justified, at least based on the data initially available.

Did police have to shot all these people who were threatening with knives and such? Probably not. Better training and techniques could probably have handled those situations without anyone dying or getting seriously injured.

[2] Warning: I arrived at 40% by looking at around 25 of them, but it was the first 25, not a random sample of 25 out of the 63. Since they are ordered chronologically, if there is some seasonal variation in the circumstances under which people get shot by police it could affect my estimate.

(FWIW, The Guardian, at https://www.theguardian.com/us-news/ng-interactive/2015/jun/... , lists 1146 people killed by police in 2015, of which 306 were black. Almost identical percentages, and it doesn't change your point.)

The tricky part is to look at how non-black people were treated in similar situations. If use of force would have been justified against white people, but non-lethal means were used instead, then it wouldn't show up in the resources you consulted.

Sure, it's a bad strategy, and we definitely should do something about the police violence.

But being biased towards violence when criminals match the demographics of violent criminals who previously have attacked you or your friends is an understandable human response, and not particularly racist.

My point is this: police are likely just violent douchebags who are actually trying to be a little anti-racist; they're violent psychopaths who escalate to fatal violence when someone who even vaguely matches a previous attacker resists their orders at all, but they actually respond that way slightly less to black people than the raw statistics of who murders who suggests they would, showing a sensitivity and attempt to tune that response.

Of course, this isn't an actual study or theory. It's merely pointing out that the narrative around the data was shaped before any real analysis of the data was performed (and by a group with a horse in the race), and there's perfectly coherent stories that aren't racist. I find it a little strange how quickly the orthodox view around the issue formed, and a little disturbing how viciously it's defended, even against moderate versions of the same view.

I don't think anyone is disputing that US police are unnecessarily violent, though.

tbrownaw 3550 days ago

I think applying statistics to societal issues is a right pain in the ass to get right, and even more of a pain to explain concisely without accidentally becoming misleading.

It's not dishonest to get something wrong, when that something is unfeasible or perhaps even impossible to get right.

You're correct that it's not dishonest to get something wrong, and I regret targeting my comment as directly as I did.

However, I think places like the Guardian are being dishonest: they're telling stories with those numbers that the numbers don't really justify, and presenting those figures as if they're meaningful and carry some sort of utility, rather than just being random mishmash numbers that were easy to cook up.

http://www.huffingtonpost.com/entry/campaign-zero-debunks-po...

anon1385 3550 days ago

>Among this larger statistical breakdown of police violence, Campaign Zero also discovered that there is no direct correlation between police violence and violence that occurs within a given community.

[1] http://mappingpoliceviolence.org/2015/

Just reading their report quickly, it doesn't look like they're controlling for a lot of things, and are measuring non-target cultural factors which skew their correlation. (Of course, they don't even mention their methods, because it likely would be such an obvious source of criticism with their conclusion.[1])

Which isn't to say I'm right either, and certainly raises a bigger question about it, but back to my point: we just can't tell, because there is a lack of detail in their analysis.

Their methods are listed under the "about the data" tab, which links to http://mappingpoliceviolence.org/aboutthedata/ :

> This information has been meticulously sourced from the three largest, most comprehensive and impartial crowdsourced databases on police killings in the country: FatalEncounters.org, the U.S. Police Shootings Database and KilledbyPolice.net. We've also done extensive original research to further improve the quality and completeness of the data; searching social media, obituaries, criminal records databases, police reports and other sources to identify the race of 91 percent of all victims in the database.

It also lists their definitions for "police killing", "unarmed", "vehicle", and "allegedly armed."

They also make the data available for download, so you can verify it. The columns are:

  Victim's name
  Victim's age
  Victim's gender
  Victim's race
  URL of image of victim
  Date of injury resulting in death (month/day/year)
  Location of injury (address)
  Location of death (city)
  Location of death (state)
  Location of death (zip code)
  Location of death (county)
  Agency responsible for death
  Cause of death
  A brief description of the circumstances surrounding the death
  Official disposition of death (justified or other)
  Criminal Charges?
  Link to news article or photo of official document
  Symptoms of mental illness?
  Unarmed

I meant their methods of analysis for that specific report, not gathering their base dataset.

I did, however, already download their dataset to look through it later, when I have more time.