[gojomo]

I'm suspicious. It seems the study may only be regressing on 'availability' of Uber – the single bit flag of whether it has launched in a city. If adoption is gradual, as the habits of drinkers/drivers change over months (or years), beneficial effects might not be seen by such a study, or remain hidden by all the other controls applied.

The best analysis would likely need to use ride volume data; there's no hint in this study's abstract they've done that, and the paper is paywalled.

[officemonkey]

Uber doesn't launch in areas where there isn't a market. If there is no need for extensive taxi and mass transit services, Uber won't be there either.

In places where Uber has launched, they're probably just picking up riders that would normally take a cab or a train if Uber wasn't there.

[gojomo]

Of course they only enter markets where there's demand. But it takes a while for people to understand and adopt something new as their preferred transit. Some initially view Uber/etc as weird, before later becoming big fans.

In San Francisco, it appears Uber/Lyft have massively increased the total number of paid-rides taken. They're not just shifting trips from taxis or public transit, but also from private car usage – and creating new trips where people would've just stayed in or walked.

That points out another stat a 'gold standard' study should try to identify: fatalities per trips (or ride-miles) taken, rather than just absolute number of fatalities. If cities with Uber have the same number of fatalities, but spread over twice as much travel, that's giant safety and welfare win, too.