Lies, damned lies and statistics on "racism"

African-Americans are about 13% of the US population. African-Americans commit roughly half the homicides in the US. That means African-Americans commit unlawful homicide at a much higher rate than other Americans (5.8 times higher than whites in 2013). Which means that we can expect African-Americans to be arrested and convicted for homicide at a much higher rate than other Americans. So, the mere fact that African-Americans are arrested and convicted for homicide at a much higher rate than other Americans is not an indicator of racism by police, prosecutors, judges or juries.

African-Americans are also far more likely to be victims of homicide (4.9 times higher than whites in 2013). In 2013, 3,005 white Americans were murdered, while 2,491 African-Americans were murdered. Of whom 2,245, or 90% of murdered African-Americans, were murdered by an African-American. If we are going to say blacklivesmatter, then by far the biggest violent death risk to African-Americans comes from other African-Americans.

Homicide is, statistically, a "good" crime because the definition is pretty clear and the reporting can be expected to be reliably high and consistent. Since it is also an extreme crime, one can expect that African-Americans will also disproportionately commit other crimes and disruptive behaviour. (Obviously, only a tiny proportion of African-Americans commit murder, it is just a significantly higher tiny proportion than for other Americans: and that can be expected to follow for other crimes and disruptive behaviour.)

Mere disproportion is not enough
So, any study which (pdf) merely points out that, for example, a police force arrests African-Americans for various crimes at a higher rate than white Americans is not proof of racism by said police force. Without some measure of the comparative rate of actually committing said crimes, it is not even, in any strong sense, evidence of racism. Especially not when what is done to arrested African-Americans matches what is done to white Americans arrested for the same crimes.

Nor are statistics which (pdf) show that African-American students are suspended at about 3 times the rate of white students and 10 times the rate of Asian students. Indeed, these statistics are completely compatible with teachers being biased in favour of black students and against Asian ones if, for example, African-American students are, say, 5 times more likely than white students to be seriously disruptive while Asian students are only half as disruptive as their suspension rate suggests.

In other words, without statistics on how often students in various categories are actually disruptive, or people in various racial categories commit specific crimes, we cannot tell whether there is any racial bias whatsoever in the decisions to suspend or arrest. Merely telling us the rates of suspension or proportion of arrests tells us nothing about whether there is any racial bias.

Not wanting to see
Anyone with any basic statistical knowledge should understand that. So, why bother publishing such statistics on their own? Because a lot of people do not have such basic statistical knowledge, or else refuse to apply it. The background assumption that rates of committing categories of crime, or being seriously disruptive in class, are evenly distributed among racial groups (which we have good reason to think is not the case) is allowed to frame the debate.

Which is a classic example of how signalling Virtue corrupts public debate. Supporting a study which purports to demonstrate racism--no matter how statistically inadequate--is Virtuous. Criticising a study which purports to demonstrate racism--no matter how statistically inadequate--is Not Virtuous: indeed it is condoning, expressing or whitewashing racism. Facts and statistical competence do not matter, only signalling Virtue matters.

What folk so obsessed with their own sense of Virtue demonstrably do not care about is the consequences of getting it wrong. If US public school teachers are systematically racist (a dramatically implausible claim, given the politics of teacher unions), then that is a very different problem than if African-American students are disproportionately disruptive. If the latter is true but the former is blamed, then that will make things systematically worse, since it will be that much harder to deal with disruptive student behaviour.

Similarly, if police in a particular jurisdiction are systematically racist, then that is a very different problem than African-Americans being much more likely to break the law. Once again, if the latter is true, but the former is blamed, then that will also make things systematically worse, since it will be much harder to arrest or cite African-Americans for breaking the law.

Police killings
The sharp end of this, and the focus of blacklivesmatter, is police killings of African-Americans. Though police kill more white Americans than African-Americans, African-Americans are killed by police at a much higher rate than their share of the population, as is pointed out here and here. But they are killed in proportion to the rate that African-Americans commit homicide. Which does not mean there is not a problem with the willingness of US police to use deadly force (fairly clearly there is), but it is not merely a problem of racism: likely not even mostly a problem of racism.

That US politics is so dominated by competing narratives--different languages of politics--does not help. Racism fits neatly into the oppressed/oppressor dichotomy of progressive discourse. Conversely, focusing on much higher crime rates among African-Americans fits neatly into the civilisation/barbarism dichotomy of conservative discourse. Each discourse-community thinks what is the "real" underlying problem is obvious and despises the other for not accepting the "obvious". (A problem which extends to other political issues, but is particularly destructive on the matter of racism and race relations.)

Conversation across the politicised discourse communities is difficult and sadly rare. Constance Rice, an African-American civil rights lawyer engaged by a new LAPD police chief who talked to lots of LA police officers, reported a pervasive fear of black men among police. Given that African-Americans are apparently, according to Peter Moskos, a professor at New York’s John Jay College of Criminal Justice and former Baltimore cop, five times more likely to kill police than white Americans (a very plausible claim, given the differential homicide rates), police fear of black men has a statistical basis. Which makes the problem that much more tangled and difficult. A merely irrational fear would be easier to deal with.

Comparing Asian-Americans
If African-Americans acted more like Asian-Americans (or Jewish-Americans), the issue of racism and race relations in the US would become much less tangled. After all, it is not as if Asian-Americans were not historically subject to vicious racism in the US. Now, they are clearly treated as functionally indistinguishable from white folk. Indeed, a dark-skinned person of South Asian descent can apparently be a very politically successful good ole boy in deepest Louisiana. Merely being physically not-white is clearly not the problem [in these instances].

Which also points out how much racism is socially constructed. This gif set, from this BBC documentary, provides another excellent example of that. As well as a reminder of how vicious white American racism against African-Americans was in the 1940s: lynchings were the homicidal end of a pervasive racism. Racism which manifested--as this essay by writer Ta-Nehisi Coates spells out--in manifold different ways well into our time.

Though I am deeply sceptical that reparations are a useful answer, even without the very messy mechanics of who is supposed to owe whom and why would not the benchmark be a comparison between the current situations of African-Americans and West Africans? I am sceptical because the political structures of the US rather preclude the German Holocaust reparations example Ta-Nehisi Coates cites from being followed.  (It is much harder for the US political class to impose resisted outcomes on the US electorate than it is in Germany.) But, more to the point, a community that waits for someone else to redeem them will wait forever.

Why don't African-Americans collectively behave more like Asian-Americans? Because they do not have the same internal trust networks, the same deeply ingrained reverence for education, the same deeply ingrained traditions of courtesy, the same strong families and kin networks. Nor, unlike Asian-Americans, did the way African-Americans came to the US select for initiative. (It is indicative that Colin Powell was the son of Jamaican immigrants, Barack Obama a university educated East African.) The levels of social capital are vastly different.

Weak social capital
It may be 150 years in the past, but the socially and culturally pulverising legacy of slavery clearly has much to do with that. As the same legacy has much to do with white racism: justifying treating African-Americans as property, and the poisonous human relations slavery creates, was the original poison-pill of anti-black racism. Which Jim Crow then reinvigorated.

There were heroic efforts to rebuild, with the black Churches being at the centre of that. But public policy kept being wielded in ways that undermined those attempts--(re)read Ta-Nehisi Coates's essay with trying to build social capital within African-American communities from a weak base in mind. Add in the use of zoning and highway construction to disrupt African-American communities (particularly in the American South).

Then add in the War on Drugs and the War on Poverty [both of which suffer from Seeing Like A State problems]. Pervasive illegal markets offering potentially high returns for illegal behaviour and property rights defended by private violence are potentially immensely destructive in communities with low social capital. While welfare can also be deeply disruptive in communities with low social capital, as it offers income without effort or social connection. This without considering where public resources went within African-American communities (basically, to those with more social capital).

But folk focused on the conservative civilisation-versus-barbarism language of politics don't want to be told that the war on drugs might have been incredibly, and unevenly, destructive. While folk focused on the progressive oppressors-versus-oppression language of politics don't want to be told that redistribution policies might be seriously counter-productive.

Automatically diagnosing racism offers a corrupting combination of comforting simplicity and effortless virtue. As, of course, does racism itself. Which are just some of the reasons I deeply disliked race analysis--taking folk's racial identity as some sort of ur-identity. It is so very revealing that both mentalities end up endorsing segregation and moral caste systems based on race (I am sorry, "privilege"). They both obsess with race, just from different moral directions.

And thereby avoid difficult questions and complexities. In African-American civil rights activist Robert L. Woodson Sr.'s words:

If race were the primary culprit in injustice and poverty, why are poor blacks no better off in institutions run by their own people, including city governments and public schools? If government safety net programs were the answer, why has $20 trillion spent on the programs over 50 years failed to improve the lot of the poor?

About correlation
On the matter of statistical literacy, I have noticed a tendency for folk, when someone cites correlations they do not like, to parrot off "correlation is not proof of causation" as if that means that correlations have no evidential value at all. Yes, it is true: lung cancer does not cause smoking. But, the correlation between smoking and lung cancer is evidence of a connection (in the sense of being statistically indicative). Not proof, but evidence; and the stronger the correlation, the stronger the evidence.

Correlations are, of course, only evidence, not proof. (And evidence can, indeed, be misleading.) Typically, folk have some model of what is going on and a correlation which points in the direction the model suggests provides evidence (not proof, but evidence) that the model is onto something.  The stronger the correlation, the more reason it provides to have confidence that the underlying model is onto something.

So, no correlations are not proof of causation. But that does not mean they have no evidentiary power at all. Otherwise, folk would never bother with the things in the first place.

It would be nice if people did not sacrifice statistical literacy to Signalling Virtue. But sanctimonious self-righteousness has a lot more emotional power, and cognitively corrupting simplicity, than open-minded concern for facts and evidence and the messy complexities such reveals.


[Cross-posted at Skepticlawyer.]