Applying mathematics to police killing data

114 Views Asked by At

I want to analyze, albeit loosely, the rate at which black and white offender populations are killed by police (because often people argue disproportionate killings by police are explained by disproportionate crime rates, so I want to investigate this). Now, there aren't any published data on the statistics here so I feel compelled to come up with my own.

Say we pick the year 2017. According to Statisa, 457 white people were killed and 223 black people were killed by police. According to census, that same year white people composed 77% of the population and black people composed 13% of the population. Now I can find the proportionate value that should have been killed, $X$, assuming both populations are killed at an equal rate \begin{align*} \cfrac{0.77}{457}&=\cfrac{0.13}{X} \\ 0.77X &= 59.41 \\ X &= 77.15 \\ \therefore X &=77 \end{align*}

So assuming an equal rate, the real amount killed, say $K$, should have been 77. But this wasn't the case. Instead it is 223. So $\frac{223}{77}=2.89$, meaning black people are killed at a rate 2.89 times greater than white people are. But at this point, this is when people argue black people commit homicide at a disproportionately greater rate, typically using the FBI's data table 6 in 2017 where 47% of homicide offenders are white and another 47% are black.

I think it makes sense to take 47% of the total amount killed by police, and only look at those populations by race in order to consider only offender populations. 457 turns into 215, and 223 turns into 105. I used the same proportion equation and found $X=36$ while $K=105$, $K/X=2.91$, and so it follows black homicide offenders are still killed 2.91 times more than white homicide offenders. This revelation should dispel racist counter-arguments used to justify a pandemic of police brutality, yes? If a disparity dependent on race did not exist, then $K/X$ should be 1 or very close to 1 in both offender and non-offender populations killed by police.

Are there any flaws with my math or anything I can do to make it better? If my approach is correct, then I could apply it to data that spans say, over a decade to get an average rate over the years.

2

There are 2 best solutions below

5
On

No, this is not the way to go about this. You got the same proportion because you reduced both numbers by the same proportion. You’re not using the data that would be required (at a minimum) to draw the sort of conclusions you want to draw.

You have no data about the prevalence of homicide. For all you know from the data you cite, anything from none to all of the people killed by police could be killers. If none were, the homicide data disaggregated by race would be irrelevant, and the conclusion that black people are being killed disproportionately would stand. If all were, the conclusion would be that the police are actually disproportionately killing white people, since equal proportions of the homicidal population were white and black and more white killers got killed.

And that’s just the beginning. As Osama Ghani rightly warned, there are lots of contextual aspects that would need to be taken into account. For instance, the FBI data on homicide offenders doesn’t necessarily reflect actual proportions of homicide offenders; it could be biased against black people because homicides by black people tend to be investigated and prosecuted more thoroughly.

1
On

Your math is wrong. It seems like you want to compare "the number of individuals killed by the police" to "the number of individuals the FBI believes committed homicide". To do this, you would just take the numbers $457$ and $223$ and divide each by the number of individuals committing homicide in the FBI's tables. From your numbers, it seems like the divisors are roughly equal, so this quotient for white people would be about twice that for black people.

There's a problem here: I had to describe what you're investigating as "this quotient" because it's really not anything more than a quotient. It doesn't mean anything on its own - as far as math is concerned, it's just this number divided by that number, and it would be foolish to ascribe more meaning to it.

We're not bound to stay in the world of pure math though: we can further ask what it would mean to believe this number is meaningful. Well, we'd have to first believe that the numbers we put into the division are accurate - so we're assuming the FBI's data on homicide rates is an unbiased reflection of homicide rates and that the numbers of people killed by the police is accurate. Moreover, we'd need to believe that the division is meaningful - which is to say: we'd need to think that rates police violence have something to do with homicide rates.

Any one of these assumptions is a whole can of non-mathematical worms - for instances, as joriki points out, the FBI's analysis of homicides is absolutely subject to the possibilities of racial disparity in prosecution or investigation or in what the FBI views as justifiable. The division in particular is a real problem because it implicitly assumes that the police are using violence in response to violence - which, of course, is the very question we were trying to address in the first place (and which has, regardless of one's intentions, incorporated a narrative of the police and tried to ascribe the objectivity of a number to it - and frankly, this is an assumption that conflicts with recorded evidence).

Of course, if it is your purpose to argue that the police are using violence along racial lines, you would realize that this statistic isn't at all helpful. I don't think any of the requisite assumptions have much merit, though you could easily hide this by presenting a number and not discussing it properly. If you are trying to show that the police killings in the USA are racist, you might have reacted with surprise or disbelief at the original number not coming out the desired way - but if we are to seek the truth, we have to ask whether we would have reacted with the same scrutiny towards a statistic that did, on its face, support our view. Essentially: one has to ask that, if this number is so meaningless, why should we believe any other number to be any better? (And this question should give us pause - deriving meaning from numbers is a hard task and not one where mathematical knowledge is at all useful)

To be honest: math isn't likely to help here. Statistics will tell you anything you want to hear, if you just choose the right ones to believe. A number hides so many assumptions under a veneer of objectivity - and this "objectivity" is so often deployed to minimize marginalized voices or to promote the assumptions inherent in the number without the investigation these assumptions deserve. I see mathematicians far too often wrongly assume that if they can bring numbers to a discussion then they must have something important to say - and then to, somehow, take a role in defining the truth, even when they have no business doing so and when doing so hurts the abilities of others to be heard.