A couple years ago Uber claimed that it had reduced DWIs in Seattle. I analyzed that claim here (and an interesting extension by Lindsay Pettingill here). This claim is now coming back and there’s a pretty compelling test-ground for it: both Uber and Lyft withdrew from Austin, TX early in 2016. A number of media outlets have made the claim that this has increased DWIs in Austin.
So: new data! Maybe it’s time to update my priors! I went and grabbed monthly DWI arrests for 2010-2016 from the Austin Police Department (previous years are not available AFAIK). You can download the data here. Data is missing for August 2010 and 2011 and I have added a dummy variable for months where Uber was operating in Austin (June 2014 to April 2016).
One year is driving the results
That’s the headline finding. First, let’s look at a couple of descriptive graphs. Here’s the year by year total for DWIs in Austin (all graphs are made with my open-source graphing software for making presentation-ready graphs, Playfair!):
So DWIs were low in 2010 and 2011, spiked in 2012 and 2013, and have been declining since then. This should already be triggering warning sirens already. All of those articles I linked above? They use 2013 as a comparison point, but it sure looks to me like 2013 is an outlier. Maybe enforcement was up in 2013. Maybe the people of Austin, TX were just particularly disorderly that year. Moreover, DWIs in 2016 are very low despite the fact that Uber and Lyft left in early May, leaving nine months where Austin, TX had little formal ride-sharing (some substitutes have sprung up).
Even with 2013, there’s no clear evidence that ride-sharing is reducing DWIs
Here are the raw monthly totals with the period where Uber was active shaded:
Boy howdy is this a noise time series! Obviously month-to-month variation is pretty critical to understanding all this so let’s try something else. Here’s all 12 months with the average number of DWIs in each month when Uber was in Austin and when Uber wasn’t in Austin:
With a few exceptions, the averages are pretty close. If you get a percent difference for each month between with Uber and without Uber and average them, that average is 2% in Uber’s favor. I don’t want to trivialize that – with an average close to 500 DWIs per month that means a reduction of a bit less than 10 DWIs each month. But That result is entirely driven by 2013. If 2013 is removed from the dataset, the time with Uber and without Uber is basically identical. Without knowing more about why there were so many DWIs in 2013 I don’t think we can say anything definitive.
If you want to go the regression route here, you might try regressing DWIs on the Uber dummy and dummies for years and months. This will show no effect from the Uber dummy.
Two final thoughts:
1) From a theory standpoint, I don’t understand this either! It certainly seems like lower transportation costs should translate into less drunk driving. Maybe we need more data or maybe we need better analysis. These trends are noisy and the possibility of temporary changes in enforcement are too plausible to be confident about differences in differences designs, for example.
2) This probably isn’t the specific policy impact we should care about the most anyways. Even if ride sharing does decrease DWIs, Uber and Lyft are currently engaged in a price war that has both companies losing truckloads of money. Eventually a ride sharing company will have to be profitable, prices will rise, and the price differential between old fashioned cabs and ride sharing will decrease, making it less likely that they can have an impact on DWIs.