Obviously the article goes into a lot more depth, though in my opinion things could be a bit clearer if they first established if Autopilot performs any differe

Obviously the article goes into a lot more depth, though in my opinion things could be a bit clearer if they first established if Autopilot performs any differently on freeways or not and then try to see if controlling for any of the other possible confounds has any effect. I'm also not overly fond of their reweighting method, if not done carefully you'll just be amplifying a lot of noise.