Monday, September 14, 2015

How to Lie With Statistics

I recently saw an article from Mother Jones, a publication not known for its impartiality, titled "A Majority of Cop Killers Have Been White." This comes as no surprise. Setting aside for a moment the predominant whiteness of anti-government, "sovereign citizen" militias and their propensity toward cop-killing, the fact is that most people in the United States are white (presently 77.7%), so even if all races killed cops at an equal rate, a majority of the cop-killers would be white. That's how proportionality works--if you have a majority in an overall group, they will also be the majority of all proportionally represented sub-groups. If you have a pitcher of Kool-Aid made of 99.8% water, then any size cup of that Kool-Aid will also be 99.8% water.

Sub-groups aren't always proportional, though. For example, although 77.7% of people in the United States are white, only 69.8% of all persons arrested in the US in 2013 were white. Studying such disparities can lead us to important insights. The problem is that people pushing their pet political agendas often present the disparities as if they were insights in and of themselves. Such presentations rely on shared assumptions and prejudices. For example, an audience that shares the view that non-white Americans are oppressed victims would see this disparity as prima facie evidence that the American criminal justice system is disproportionally hard on non-whites and easy on whites. An audience that is biased in favor of whites and against non-whites, however, would see the disparity as being proof that whites are more law-abiding, while non-whites are more prone to criminality. When such an assertion is presented to a like-minded audience, it tends to slide by unchallenged and serves only to bolster the audience's confidence in their belief that their prejudices are correct. Psychologists call this phenomenon "confirmation bias."

Notice, for example, that when left-leaning journalists and bloggers speak of the number of persons arrested, convicted, incarcerated, or killed by police, they tend to speak in terms of rates rather than percentages. The reason is that in all of these cases, whites make up the majority, but the majorities are less than proportional. For example, while black Americans make up 13.2% of the population, 23.8% of all people arrested in 2013 were black. So if you want to confirm the bias that black people are treated unfairly, you could present this disparity without further data and interpret it (without supporting evidence) as being proof that the preconceived notion is correct. "Of course black people are treated unfairly! Just look at the numbers!" The unquestioned presumption is that the extra 10.6% of black people arrested didn't deserve to be.

Mother Jones points out that the majority of cop killers are white, and this is true--in 2013, out of 28 known cop killers, 15 of them were white and 11 were black (race was not reported for the other two offenders). Clearly, the majority were white. Bad! Bad, bad white people! But that also means that only about 53.57% of the cop killers were white, while whites are 77.7% of the population overall. On the other hand, about 39.29% of cop killers were black, while blacks are only 13.2% of the population overall. So whites are underrepresented among cop killers, while three times as many blacks are cop killers as would be proportional.

Given the tiny numbers we're talking about, none of this says anything about the relative characters of these races. The character of millions of white and black Americans is not defined by a couple dozen cop killers. But when someone wants to promote one group as good or bad, they'll enhance or diminish the appearance of the numbers regarding that group by shifting to discussion of either percentage or rate, accordingly. When Mother Jones wants to make white people sound worse than black people (as they do), they'll present numbers that show (gasp!) that the majority group commits the majority of a particular type of heinous crime, while glossing over the fact that members of the group they're trying to lift up are more than three times as likely to commit that same crime. I'd presume Fox News would do the same thing in reverse.

In fact, I've seen this done by the other side often enough. When right-wing publications, which see poverty as symptomatic of laziness rather than victimization, want to represent minorities as being lazier than white people, they'll point to the higher rate of blacks receiving food stamps. When left-wingers want to counter those facts, they simply point out that the vast majority of food stamp recipients are white...which tells us nothing, given that whites are in the majority generally.

But even aside from whether we're talking about percentage or rate, there's the issue of assuming what the data implies, as I alluded to on the food stamp issue. Are a higher rate of blacks than whites receiving food stamps because they're lazier, or more likely to be discriminated against in hiring for better-paying jobs, or because their families weren't allowed to accumulate wealth in generations past, or any number of any reasons? It's not evident from the data. But we'll pretend it is, if it seems like the numbers support our conclusions.

As such, you can use data to lie without even changing the type of figure (percentage vs. rate) used. For example, pro-black groups will speak of the higher rate of blacks arrested and convicted. But feminists will point out that the vast majority of people arrested and convicted are men. Bad, bad men! The problem is, these are often the same people. A lefty feminist can believe that more men being arrested is proof that men are inherently more criminal than women, while simultaneously believing that more black people being arrested is proof that the system is racist. We'll move the goal post without even realizing we're doing it. If we used arrest rates as a benchmark of criminal nature, then we'd have to conclude that both men and blacks are more criminal than women and whites, respectively. If we instead use arrest rates as a benchmark of injustice (again, with the implied assumption that the excessive number of arrestees didn't deserve to be arrested), then we must conclude that the system is rigged against both men and blacks. Neither of those views fit neatly into the political boxes as they've been drawn by the two major political parties, however, so people will change the standards from one group to the next as it suits their purposes.

tldr; Data is not analysis. When you hear "most of" or "the majority" or "at a higher rate," look for the rest of the story and carefully study the underlying causes before jumping to any conclusions. If you're seeing the world through any chosen ideology, the facts are probably not always going to support your point of view.

No comments:

Post a Comment