Excited to share research with Ben Feigenberg ( @UICLAS) on racial disparities in motor vehicle searches conducted by highway patrol. TLDR: eliminating these disparities would *increase* contraband yield.

NBER: https://www.nber.org/papers/w27761 
Ungated: https://conrad-miller.github.io/fm_profiling_aug2020.pdf [1/20]

Motivation: during traffic stops, Black and Hispanic motorists are substantially more likely to be searched for contraband (e.g. drugs, weapons) by police than White motorists. Yet searches of Black and Hispanic motorists are equally or *less likely* to yield contraband. [2/20]

There is a debate about how to interpret these facts. Some see this as evidence of biased policing: why search Black/Hispanic motorists more often if those searches are less productive on average? [3/20]

Others (often economists) counter that this is insufficient evidence to prove bias. To know whether police are biased, we would need to know whether police are using a lower standard of suspicion for searching Black/Hispanic motorists, something we can’t measure directly. [4/20]

If racial disparities in search are explained by bias, then everyone can agree that search rates should be equalized across groups. But if police are using the same search threshold for all motorists, then equalizing search rates could reduce contraband yield. [5/20]

(Aside: the fact that there may be an equity-efficiency tradeoff does not imply that efficiency is the right social goal to pursue. E.g., in my view, racial profiling is unjust even where there is an efficiency rationale. In practice, people disagree on this.) [6/20]

We see the latter camp as arguing that racial disparities in searches may be justified on efficiency grounds. We ask: would equalizing search rates across groups in fact reduce contraband yield? This allows us to investigate this premise without (dis)proving bias. [7/20]

Answering this question is hard because it requires knowing a counterfactual. What would contraband yield look like if racial disparities in search rates did not exist? [8/20]

Our approach is to use the fact that officers vary in how often they search motorists. To learn what contraband yield would look like if white motorists were searched twice as often, we can look at officers that search white motorists twice as often as the average. [9/20]

(This approach requires some assumptions I’ll return to at the end.) [10/20]

We apply this approach using data on speeding stops made by Texas Highway Patrol. Black and Hispanic motorists are 200% and 60% more likely to be searched. The rate that searches yield contraband (“hit rate”) is similar for Black and White motorists, lower for Hispanics. [11/20]

The key to our approach is that we need to measure how troopers vary in how they treat essentially identical stops. This can be tricky because troopers patrol different areas, and even in the same area, troopers may differ in whom they decide to stop for speeding. [12/20]

We take three approaches that give us consistent answers: selection on observables (fixing stop location and time), within-motorist design using those stopped multiple times, spatial regression discontinuity using patrol area borders. A lot more details in the paper. [13/20]

Here’s our main finding. These figures plot the relationship between trooper search rates (x-axis) and the percentage of stops (not searches) that yield contraband, by motorist racial group. The key feature to note is that these relationships are approximately *linear*. [14/20]

In other words, troopers that search twice as often find twice as much contraband. (That’s more surprising than it may sound. At some point we would expect declining returns to search--searching more often requires searching those that who less likely to have contraband.) [15/20]

That means that we can use observed hit rates by group (≈ to the slopes) to approximate counterfactual hit rates. If we replace searches of Black/Hispanic motorists with searches of White motorists, we get the same/higher contraband yield (similar/higher hit rates). [16/20]

Hence, we conclude that troopers could equalize search rates across motorist racial groups, conduct the same number of searches, and *increase* contraband yield. [17/20]

Two caveats. First, to construct counterfactuals, our approach requires that troopers that search motorists at high rates aren’t systematically better at figuring out which motorists are carrying contraband. [18/20]

Second, we abstract from deterrence effects. If troopers search white motorists at higher rates, perhaps white motorists would be less likely to carry contraband in the first place. We provide suggestive evidence in the paper that neither of these are first order issues. [19/20]

There’s more in the paper. For example, we find that low income motorists are also more likely to be searched yet those searches are less likely to yield contraband, and that black-white disparities in search are larger in counties with higher 2016 Republican vote shares. [END]