Mirror, mirror on the wall: Race, police and crime
I recently got into a discussion about police killings in the USA. Having not looked at this issue for some time, I decided to examine the available data myself, to see what I could unearth.
The below train of thought should be seen as a ‘back-of-the-envelope’ investigation, rather than a comprehensive analysis.
Deaths by police
The most recent USA census captured deaths resulting from confrontations with police. The graph below shows the absolute numbers broken down by racial category, “White”, “Black” and “Asian”. I have focused here just on these 3 racial groups, because they are most comparable with the groups captured by the FBI (the relevance of this is discussed later).
If we want to get an initial impression of the risk that any given group faces in society, we must look at how many people from a group died, versus their proportion in the wider population. According to the Statistical Atlas, racial proportions in the USA are as follows: white: 62%, black: 12.6%, Asian: 5.2% and “other”: 20.2%. If we now weight absolute numbers of those killed, above, by their proportion in the population, we get the risk of death posed to each group:
The above graph shows that, when correcting for the fact that blacks only make up 12.6% of the population, they are 2.5 times more likely to be killed by police, than whites, and nearly 7 times more likely to be killed by police, than Asians.
Deaths by gender
Out of curiosity I stratified deaths by gender, instead of race, and applied the same method of estimating death risk:
Police kill men, as can be seen above, roughly 23 times more often than they kill women.
Making context assumptions
I found the disparity in gender outcome interesting. Without any information about context, it would be tempting to conclude that the gender-stratified deaths indicate a “war against men”. Intuitively, I reasoned, however, that men are likely disproportionately killed because they themselves commit violence disproportionately and, as such, are more likely to come into violent (and potentially lethal) confrontations with police.
With this hypothesis in mind I looked at FBI data of police killed in the line of duty between 2004 and 2013. Of the 565 police deaths, 98% (551) were caused by men. With the assumption that violence towards police is causal for death, the respective contribution to police deaths gives a completely different picture of who is at more at risk of being killed:
In other words, when the risk people pose to police is taken into account, we find that women are, in fact, killed more frequently than men, than we should expect.
Context assumptions about race
What happens if we weight risk posed to police, by race, as we did with gender? The results, once more, flip things around:
With the weighting of risk to police, blacks are killed approximately half as frequently as whites and Asians.
But what does “risk to police” mean. Here is were we run into difficulty. The event of a police death does not, in and of itself, give context. We have no explicit evidence that deaths from either group result from mutually lethal confrontations. If we assume that there is lethal combat, who is the aggressor? On one hand it’s possible that officers are at greater risk of being shot in certain neighbourhoods and, therefore, are more likely to kill in response. But the converse is also possible: that police are the aggressors, and that they are killed when black people defend themselves. Or, perhaps, other factors that we aren’t (or cannot) be aware of.
Risk of death from the general population
Police deaths aside, I investigated whether black people, as a group, are at disproportionate risk of death from the hands of their fellow civilian countrymen. I looked at national-level FBI homicide data and demographically weighted the numbers to get a blunt first impression of the risk of homicide, within, and between, races.
The numbers show that people are overwhelmingly killed by members of their own group. To my understanding, this is expected because most homicides occur within families, and among people already known to each other. This finding, then, is merely reflects the reality that most people live near, and mix with, people who look like them.
In terms of inter-racial killing, white victims of homicide are 12 times more likely to have had a black assailant, than vice versa.
Lies, damned lies, and statistics
At this stage it is worth pointing out that I don’t trust the above data, or the conclusions that can be drawn from them. Some obvious problems:
- The sample size of cops killed is very low (565). The number killed by women (90) and Asians (36) is so small that including them in the analysis verges on irresponsible. There is also no indication of the size of the police force, so the actual risk of an officer dying cannot be known, because the total number of serving officers, needed to establish that risk, is not known.
- The “contextualised”, relative risk analysis involves multiplying scaling factors, meaning that errors, omissions, or faulty assumptions, are compounded. Similar magnitude values, like murder risk by gender or race, versus risk to cops, should be looked at with caution. Note, also, that scaling assumes that deaths result from mutually lethal confrontations with police, which, while plausible, may not be the case.
- Continuing on point 2, above—The context of “risk posed to police” can, at least in theory, be causally attributed to gender: high testosterone is known to cause aggression and risk-taking behaviour. Men have much more testosterone, so the claim, then, that they pose greater risk to police, at least has some biological plausibility. There is no such link between race and aggression.
- US census data classifies race as Black, White, Hispanic, Native American and Asian. FBI data, in contrast, sometimes uses this classification, but at other times uses White, Black, Other Race, and Unknown Race. Comparing FBI data with Census data therefore potentially cross-contaminates racial pools.
- Race is itself a contentious issue from an analysis point of view. What does it mean to be “black”? Where is the line between black, white, hispanic and “other”. It is conceivable that prior conceptions of what defines race influence classification of police homicides. People, for example, who appear racially ambiguous, or who are “edge” cases, might be classified as “black” because of a prior belief that black people commit crime. Conversely, police might classify a slain person as “other”, or “white” to avoid accusations of racism. Both of these practices would skew the picture.
Further complications and distortions of racial classification
To what extent is race a useful predictor for any one person’s experience? I tried to find data that was stratified in addition to race, to get an idea of the extent to which race is, in and of itself, usefully predictive. Are all people classified as black, equally likely to end up in prison, for example? I found that the risk of incarceration for black foreigners, versus black Americans and white Americans, for high school drop-outs in 2010, showed enormous disparities:
From an incarceration point of view, young black immigrants appear to have far more in common with white people than they do with the black Americans they are categorised with.
This finding led me to question wider race aggregated data: to what extent do racial correlations survive stratification over socio-economics, education level, and other factors? My impression from the geographic data is that very specific postcodes overwhelmingly contribute to crime and police confrontation. Pooling people in these postcodes with people outside of them therefore risks two negative outcomes: First, it dilutes the problem facing the worst postcodes, artificially making them seem less violent than they really are. Second it creates the impression that people in different postcodes, and from different backgrounds, are at higher risk than they actually are. Media stories about black incarceration that do not aggregate data by immigrant status, for example, might easily allow a black immigrant to believe that they’re at special risk of going to prison, when, in fact, they aren’t.
Discussion points
So far in this journey, I get the impression that:
- In terms of police: Establishing what amounts to a disproportionate risk of death at the hands of police is impossible to know from the available numbers. As such, my suggestion would be for an independent panel to investigate each death and assess the circumstances, as best as possible, for both police officer and civilian deaths. Until that happens people can conclude what they want from the data with their own assumptions about what the context is.
- There is no evidence at all that black civilian people are at disproportionate risk of violence or death from white civilian people.
- The only lethal police confrontation I have seen on Youtube demonstrated, in my view, disproportionate police aggression. It showed a man with a chain destroying a car in a blind rage. A police officer drew a fire arm and commanded the man to put his chain down. The man, seemingly agitated, drunk, and potentially mentally unstable, didn’t comply with the instruction, and instead walked towards the officer, swinging his chain. The officer repeated his command for the man to drop the chain, and when the man once more didn’t do so, the officer shot him dead. It seemed to me an extraordinarily disproportionate confrontation. There were no civilians in the picture, so the only risk to life was to the officer himself, who chose, for some reason, not to retreat to a safe distance. Instead he stood his ground and fired lethal shots in a built up residential area, thereby placing innocent civilians at risk in a situation that was, at that time, non-lethal. In this one context, my sense was that the police drove the lethal confrontation. What about all the other cases?
- It bothers me that there is no agreed definition of what race is, or who gets to decide it. Some claim race is “self identified”, but this is problematic for two reasons: first, homicide victims cannot state their race. Second, public reactions to people such as Rachel Dolazel, who identify as black, but who are denied this identity by their peers, suggests that, at the very least, race is something negotiated with society, rather than privately chosen.
- If we go with my working hypothesis that death at the hands of police is restricted to violent, deprived, predominantly black postcodes, then the pooling of affluent black groups and black immigrant groups, with this subset, massively downplays the true scale of the problem. I am left with the impression that, correctly stratified by socioeconomics, geography, and education level, certain neighbourhoods may well prove themselves to be war zones, existing in a parallel world to normal society.
I will continue the journey of examining race-based data. For those interested in the next instalment, you can follow me to receive updates.
I welcome any critical comments and angles that I may have missed.
References
