I love when other people do my writing for me! Thank you and keep these coming:
On the racial disparity stuff, at least the DOJ is claiming they controlled for crime rates. (see p.63). But did they really? The report is full of boneheaded statements like “Despite making up only 41 percent of the Northern District’s population, African Americans accounted for 83 percent of stops in the district.” (p.65, n.72). Gee, how much of the violent crime in the Northern District is being committed in Greenmount [ed note: poor black] as opposed to Roland Park [rich white]?
And people always talk in terms of higher crime rates in certain “communities” — as I just did. But communities don’t commit crime; people commit crime. Saying black people live disproportionately in high-crime neighborhoods is a euphemistic way of saying black people commit a disproportionate amount of crime. (And we’re talking about violent crime here, which is what drives street-level proactive enforcement). [Last year 93 percent of Baltimore homicide were black]
If they tried to control for crime rates by focusing on the rates of crime in different communities rather than the rates of offending by black and white residents, isn’t that problematic? (Genuine questions here; I’m no statistician) [Ed note: Yes]. Doesn’t this all come down to how the DOJ drew the boundaries for the neighborhoods they studied? What did they use? Police districts? It seems like it. Look at p.63 (referring to disproportionate stops of African Americans despite “significant variation in the districts’ demographic characteristics and crime rates”).
Let’s say the fictional North Central district is 100% white, and the fictional South Central district is 100% black. The crime rate is twice as high in the South Central, and people in the South Central are twice as likely to get stopped by police. I assume that would pass muster with the DOJ. But, of course, real districts in Baltimore don’t look like that. They’re mixed and largely segregated. They look like the Northern District, with lots of crime in black neighborhoods like Greenmount and hardly any crime in white neighborhoods like Roland Park. They look like the Southeast district, with far less crime in largely white Sector 1 than in Sector 2 (bordering the Eastern), where the population is mostly black. It seems to me this messes up any attempt to control for crime rates if they’re only looking at things at the district level. Better would be to break it down by individual neighborhoods, though that leaves room for error, too. Even better would be to study actual rates of offending by black and white residents. What percentage of robbery suspects were reported as black? What percentage of shooting suspects in cases that had eyewitnesses? What percentage of suspects from “CDS outside” complaints were described by the callers as black?
Wouldn’t it be a powerful illustration to just look at the gender of people stopped? Men presumably account for about 50% of Baltimore’s population. Surely they make up a much higher percentage of those stopped and arrested. And I’m sure the disparity persists even if one attempts to control for crime rates in different “communities.” (I doubt the male/female ratio differs all that much district-to-district). Is that evidence that BPD cops are profiling men, and have an anti-male bias? Or are cops focusing on men because, well, men commit most of the crime in the city, and therefore it’s mostly going to be men who exhibit suspicious behavior that prompts officers to stop them?
In geography, the term for some of these issues is the Modified Unit Area Problem (MAUP). Basically its an issue when the geographic unit you have data for might not be the best to analyze whatever it is you're trying to study, or when the results of your study will vary based on the geographic unit you choose. The writer makes a good point about how the heterogeneity of crime and race within districts would affect analysis. If the data are aggregated to the district level though, there isn't much you can do about it unless you can go back through the reports and disaggregate them to the block or neighborhood level. I doubt DOJ had the resources to do that and identifying neighborhood boundaries has its own set of issues.
Where I think the writer starts going wrong is when they start talking about controlling for crime rate by race. You'd have to break down racial crime rates in each district by race rather than use city wide racial crime rates. You also need to choose a crime rate that can't be subject to reporting bias. Basically, you can't use anything where officers have a degree of discretion about enforcing the law. So loitering, trespassing, and drug possession are probably out. Called in reports and witness statements are probably also not reliable enough to use. Even things like rapes and assaults probably aren't good for it given the degree of discretion available for arresting people in fights and the allegations about the sexual assault investigations. What you're left with is homicides. But, even though 300 homicides is objectively a lot, when you start breaking it down by race in individual districts, the numbers get too small to draw statistical inferences with a high degree of confidence.
So yeah, its not the best statistical work out there and if it were the only evidence in the report about racial bias, you might want to take it with a grain of salt. But given the other allegations about downgrading citizen complaints and that sort of thing, I think it offers some degree of insight into the priorities and tactics of the BPD.
Aweome. Good post.