Earlier this year, I wrote a post about the relationship between UberX and alcohol consumption, based on this recent article. The key takeaway was that the presence of UberX was associated with an increase in alcohol consumption. As I wrote then:
Decreasing the full cost of drinking at bars, by making it less expensive to travel to and from bars, increases alcohol consumption. That is likely to be associated with increased harm, including harm to health from over-intoxication, or violence. However, before we conclude that UberX has an overall negative effect, we need to consider that the availability of UberX might also offset some harm by reducing the incidence of drunk driving. Teltser et al. didn't look at the effects of UberX on harm, or on drink driving, which is something we would need more information on before drawing a firm conclusion.
Does Uber increase, or decrease, alcohol-related traffic accidents? That is essentially the question addressed by this recent NBER Working Paper by Michael Anderson and Lucas Davis (both University of California, Berkeley). They match normalised rideshare data [*] obtained directly from Uber with fatal traffic accident data from the Fatality Analysis Reporting System (FARS), from July 2012 to January 2017.
Interestingly, Anderson and Davis first show that the 'event study' extensive margin analysis that has been used by many other studies (which essentially use a difference-in-differences approach to compare accidents before and after the introduction of Uber, like the study linked above) leads to highly inconsistent and non-robust results. Most studies have used that approach because they lack access to actual rideshare data. Instead, Anderson and Davis look at the relationship between the (normalised) number of Uber trips and traffic fatalities, and find that:
A one unit increase in ridesharing activity reduces the probability of an alcohol-related fatal crash by 0.038 percentage points (t = -3.2). This corresponds to approximately a 4.8% decrease in alcohol-related fatalities.
Anderson and Davis then show that the effect is concentrated during nights and weekends (as you would expect), and that the results extend to all fatal crashes, in addition to the alcohol-related fatal crashes that their main analysis is based on. Based on some back-of-the-envelope calculations, they show that:
Our estimates of the effects on alcohol-related fatalities imply that Uber saved 214 lives in 2019, or a reduction of approximately 6.1%.
As noted in my earlier post, driving and using Uber are substitutes. The presence of Uber should reduce the instance of driving, including drink-driving. This appears to be borne out by the analysis of Anderson and Davis. However, their analysis falls a bit short of convincing causal evidence, lacking exogenous variation in the number of rideshare trips. And, it would be interesting to know what effect (if any) omitting New York (in particular) and Seattle had on the analysis. Take this as suggestive evidence for now of the positive effects of Uber (in contrast with the negative effects on traffic congestion).
[HT: Richard Holden at The Conversation]
*****
[*] The normalization process leads to data that is expressed as a proportion of the number of rides originating in a specific Census tract in San Francisco. They have to omit Seattle and New York from their data, because the actual number of Uber rides in those cities are published, which would allow them to back out the number of rides in every other Census tract.
Read more:
No comments:
Post a Comment