
Jeff Kessler
2.4K posts

Jeff Kessler
@KesslerJeff






Waymo blocking a bus and a whole ass intersection right outside the opera house.






The most likely interaction any of us will have with the police is a traffic stop. That simple fact accentuates the import of the following question: when an officer decides to pull someone over, does race play a role? It sounds like a straightforward question. It is not. And the reason comes down to a measurement problem that has haunted this literature, and many others, for decades. What do you need to properly explore this question? Three numbers: i) how many minority drivers were stopped, ii) how many minority drivers were on the road, and iii) how many minority drivers were actually speeding. Knowing all of these is a tall task, especially knowing who was speeding among those who were never pulled over. I call this the denominator problem and measuring discrimination critically relies on solving it. Prior research has a bunch of creative: veil-of-darkness designs, daytime versus nighttime comparisons, benchmark approaches using census data. All clever. Yet, they do require certain assumptions that everyone is not comfortable making. A paper that I just talked about at our student visitation day this past Friday (and I will talk about this week in my Economics for Everyone course) solves it differently. When I was back at Lyft, we leveraged records on the GPS location of every driver every few seconds. That gives us actual driving behavior — speed, location, time — for over 200,000 drivers before any police interaction occurs. We matched those records to official Florida speeding citations secured through a Freedom of Information Act request (kudos to Florida, still the only state to adhere to our FOIA). As far as I know, this is one of those rare occurrences where we have solved the denominator problem: who was on the road and exactly how fast are they driving? Guess what we found? You can find out here (it is depressing so I am not going to recount it): ideas.repec.org/p/feb/natura/0… But the broader lesson I want to make with my students is that the denominator problem is not unique to discrimination research. It shows up everywhere in social science. Hiring discrimination studies rarely observe the full applicant pool and almost never observe underlying qualifications. Health disparities research cannot see who never sought care. Criminal justice research fights this at every turn. The general insight is this: selection into measurement is itself the phenomenon you are trying to study. In certain cases, partnerships with organizations can help to solve that key issue.








Instead of a dying service that needs constant bailouts because it lets junkies and creeps ride for free, BART should install fare gates everywhere and be a thriving public transit system that working class people can use to get around without a car.


Oakland - one of 'most corrupt cities in America' - considers huge pay bumps for council members trib.al/zUpe8YV


There’s like maybe a maximum of 8 jobs where I understand what the job is. Shopkeeper. Farmer. Teacher of some kind. Priest. Novelist. Journalist. Private detective. Chef. That’s it. What is a data engineer. I don’t know.



I think the NBA should eliminate the 3 pointer. It would be a vastly more attractive sport.



wait this graph is crazy BART installed anti-fare-hopping gates and the amount of station maintenance and cleanup they had to do went to basically zero strong evidence that the poor condition of public transit is fairly easy to fix + caused by a very small group of people







