Dubious claims of Uber decreasing drunk driving
A recent article that appeared in the Economist blog, citing a working paper by Jessica Lynn Peck of CUNY, suggests Uber may help to curb drunk driving in NYC. Indeed the graphic presented by the Economist does seem to suggest that post introduction of Uber in four boroughs, save Staten Island, drunk driving accidents fall.
The author employs a statistics technique known as “Differences-in-Differences” to create a quasi-experimental setting in which to study the difference between a treatment group and a control group in a natural experiment. She claims that the four boroughs, Manhattan, Brooklyn, the Bronx and Queens, were “treated” with access to Uber while Staten Island was not. Thus, it is claimed, Staten Island provides a valid “control” to compare the effects on drunk driving given access to Uber . The two groups are then compared over time with attention paid to the difference in the two groups “post treatment period”, in this case the dashed red line in the graphic above.
If there was no concurrent event happening at this time, in this case May 2011, that would have differentially affected drunk driving rates across NYC, then we can reasonably conclude that this estimate is reliable. However problems in this design arise even before this assumption can be considered.
Uber announced that it was #superpumped to enter into NYC in 2011 in a press release that touted its efficiency and cost saving new app that enjoyed much success in San Francisco. Uber uses the traditional geographic boundaries of NYC, as comprised of 5 boroughs including Staten Island. Peck sets up the statistical analysis in such a way that assumes that Uber was not rolled out in May 2011 in Staten Island, as if a local regulation kept it out. This does not seem to be the case and the lack of consumer (and driver) uptake in Uber ridership gets confounded as a “control” group.
What is then the reason for the less than enthusiastic uptake of Uber in Staten Island? On the driver side, it could be that Uber gives drivers incentives to keep them in more “desirable areas” such as Manhattan and “Yuppified” regions of Brooklyn. Whereas on the passenger side, it could be that since Staten Island is connected to Manhattan either by a 40 minute drive or a 25 minute ferry, commuters prefer to take the ferry.
A better comparison for difference-in-difference estimation of Uber’s impact on drunk driving crashes is to compare two places, one where Uber was introduced and one where local regulation kept it out completely. With this design we can assume that since consumers in BC were given no option, their preferences will not bias the estimates we generate. Two such areas that can stand in as reasonable “treatment” and “control” are Seattle, WA and Vancouver, BC, respectively.
Uber was rolled out in Seattle in August 2011 with sleak black cars picking up passengers across the city. BC’s government recently voted to introduce ride-sharing services to the province, a month before a re-election bid was to commence. To do a quick diff-in-diff between these two places one would compare drunk driving accidents before and after August 2011 for both places. If there is a marked decrease in Seattle and none in Vancouver, one could reasonably assume that Uber helps decrease drunk driving.
In Seattle, 2011 saw a record low level of fatalities involving impaired drivers in 2011 and then an uptick back to pre-2011 levels.
Impaired driving involved “serious accidents” on the other hand have been decreasing over time since 2008, with a slight increase in 2012.
While in Vancouver, impaired driving accidents as a percentage of total crashes have declined from a peak in 2010.
From the charts above one can draw some preliminary conclusions that Uber probably does not account for the fall in drunk driving that is seen across four of five NYC borough. While Uber may have had an impact in NYC in terms of drunk driving crashes, Peck’s argument implying a causal effect is fundamentally flawed.
From Seattle and Vancouver one can see a marked decrease in drunk driving in both places, thus some other effects perhaps education or stiffer penalties for driving under the influence, may be helping to contribute to that overall decline.
In the mean time, Uber is dealing with accusations in Seattle that the company has ignored passenger complaints of intoxicated drivers in violation of its zero-tolerance policy for driving under the influence.