[pumblechook]

Honest question: is this not a classic example of the fallacy of claiming causation in an observational study, in the presence of a potentially infinite number of confounding variables? I'm not a statistics expert by any means, but this is one of the first things taught in an intro stats class: observational studies can establish relationships (correlation), but only well designed experiments can establish causation. This is quite obviously an observational study, and a tiny one at that.

Furthermore, wouldn't the fact that marijuana was legalized in Washington violate the parallel trend assumption that has to hold to use difference in differences when comparing with San Francisco?

I understand the powerful incentive for Uber to provide this narrative, and I think anything that encourages people to not drive while intoxicated is a good thing. I just find it disheartening that companies feel the need to start with these powerful marketing narratives (which these types of 'studies' really are, not honest efforts to uncover useful knowledge), and use dubious analysis to give those narratives credibility.

It's not that I'm disputing there is a relationship - they've obviously identified that. I have a problem with statements like these: "We estimate that the entrance of Uber in Seattle caused the number of arrests for DUI to decrease by more than 10%." And, "We can reasonably extrapolate these results and apply them to what we consistently hear from our riders in Chicago and elsewhere." It is quite a stretch to claim that Uber caused anything based on an observational study with a minuscule amount of data, and it is even further of a stretch to claim that you can combine this claim with anecdotal evidence and somehow generalize it into principle.