Showing posts tagged statistics

Lying with Statistics: Racism-Denying Edition

The Trayvon Martin story has generated a lot of backlash from conservative commentators. While most of it has been the usual thinly-veiled racist vitriol, some have coyly pointed out things like “only 3.2 percent of black victims were murdered by whites in 2005, the last year for which data are available. The black-on-white rate was higher — 8.8 percent.”

I’ve seen this statistic cited in a couple different places (though I only recall the one offhand), and a quick search tracked it back to this BJS webpage. There’s a problem; while technically correct, this statistic makes a subtle deception.

"8.8 percent" is the answer to a strange question: "given that I am White, and given that I have been murdered, what is the probability that I was murdered by someone Black?” What we really have in mind, however, is “Given that I am White, what is the probability that I will be murdered by someone Black?”

It turns out that probability is much, much lower. This stems from two distortions:

  1. The percentage of the population that identifies as Caucasian for Census purposes is relatively large: roughly 250 million, or about 83%.
  2. As someone White, my chances of being murdered at all are very low.

To create a statistic that answers the question we’re actually interested in, I went to the FBI Supplemental Homicide Statistics for 2009, the last year for which data are actually available. I used this handy tool from the CDC to get some population estimates, and came up with these numbers:

  • In 2009, excluding 'justifiable homicides' like self-defense and police shootings, there were 691 Black-on-White murders and 304 White-on-Black murders.
  • In 2009, the White population (incl. Whites of Hispanic origin) was estimated at 246,978,488 and the Black population was estimated at 40,999,984.

The Results: For 2009, the Black-on-White murder rate was about 0.28 per hundred thousand. The White-on-Black murder rate, by comparison, is about 0.74 per hundred thousand. This means Blacks are about 2.65 times as likely to be murdered by Whites as the other way around, almost exactly flipping the ratio quoted from the deceptive BJS statistic.

Of course, this is hardly a perfect comparison. There’s room for argument about the role of ‘justifiable homicides’ - if I include them in the calculations, that 2.65 ratio rises to 3.77. For the ‘Hispanic deniers’ out there, considering only non-Hispanic Whites lowers the ratio to 1.63. It’s also clear the population is not evenly distributed by race, so the numbers should be weighted to take that into account (the FBI data include region, so doing this is possible). I’ve also only used data from 2009, whereas data are available for about the last forty years. The point, however, is that a few minutes in Excel with some publicly-available data can quickly show just how wrong that 8.8%-3.2% ratio is.

Note: Sometime this summer, I plan on making a miniature project of combining the available data into some more useful form. If there’s sufficient interest, I’d be happy to make this available in something open-source instead of my usual Stata.