,
8 tweets,
3 min read
Read on Twitter
To assess racial disparities in police interactions, we compiled & analyzed a dataset detailing nearly 100 million municipal and state patrol traffic stops conducted in dozens of jurisdictions across the country, the largest such effort to date. 5harad.com/papers/100M-st… (thread)
In partnership with @StanfordJourn we filed hundreds of public records requests in all fifty states and over 100 cities, and spent years to clean and standardize the data.
We analyzed these records in three steps. First we measured potential bias in stop decisions by examining whether black drivers were less likely to be stopped after sunset, when a "veil of darkness" masks one’s race. C.f. rand.org/content/dam/ra…
Because the sun sets at different times during the year, one can examine the racial composition of stopped drivers as a function of sunlight while controlling for time of day. We found black drivers made up a smaller share of stopped drivers after sunset, suggestive of bias. 
Second we examined bias in search decisions. One approach is to look at "hit rates": if searched minorities are found with contraband less often than searched whites, it suggests a double standard. But such outcome tests suffer from infra-marginality. C.f. 5harad.com/papers/thresho…
So we instead applied our "threshold test", which uses info both on hit rates and on search rates. We find evidence that black and Hispanic drivers are often searched on the basis of less evidence than white drivers.
Finally, we examined the effects of legalizing marijuana in CO and WA. We found evidence that legalization reduced the total number of police searches conducted for all drivers, but also found that the bar for searching minority drivers continues to be lower than for whites.
We've released the data for anyone to use, along with tutorials and tools to explore our findings, including an interactive stop map for each city. Check out the website for all of this and more! openpolicing.stanford.edu
