Leave a comment

5 Comments

1. This will sound like a loaded question, but it is totally sincere: How much time do you spend cycling in the city? I tried re-taking up cycling as an adult in Seattle and it was terrifying, even in areas with clear bike lanes. In fact, I only cycled in areas with clear bike lanes….I had this vision that I would get comfortable enough that I would be able to cycle to my stable about 40 miles away via the trails paved over the old railway lines. My life changed and I ended up moving down there to shorten my commute, but cycling out here is very much an at-your-own-risk activity. But my experiment in Seattle failed. I was very much not comfortable on a bike, despite my grand vision of joining cycling culture and going all over the city without a car.

2. The Longreads assertion is not statistically correct. We can start to draw confidence intervals around the fatality rates now. More importantly, we can expand and improve the analysis massively by looking at injury rates, which, because they are far more common, provide much more statistical certainty than only looking at fatalities. The other aspect here is that the Longreads phrasing casts all AVs as having one monolithic level of safety (and implicitly behavior). Not the case, I would be far more trusting of the cutting edge Waymo algorithms from yesterday, than I would be of Uber algorithms earlier this year. Estimating safety is going to a be a moving target where you need to disaggregate by algorithm and equipments. You need a Bayesian approach and you need to look at a uniform definition of injury.

   • Also, scanning the Longreads article, it’s just a bad piece made to look authoritative by lots of quotes and serif font. The author doesn’t really know what they are talking about, and it shows in their inability to hold their sources accountable.

   • Hey Alex, Thanks–I miss your statistical expertise down here in LA.

Comments are closed.