Sunday, June 18, 2017

The effect of Uber on traffic congestion

Early this year, a paper in PNAS using a computer model estimated that car sharing services like Uber and Lyft could reduce the number of taxi vehicles on roads by ~76% without significantly impacting travel time. As Joe Cortright has said, the authors are overly optimistic. 

There is another study from last year that analyzed what actually happened to congestion levels when Uber entered the market in some US cities (abstract below). The results of this study are not really comparable to the the paper in PNAS, though. The methods are sound but I have the impression the authors pay too much attention to the statistical significance of the results and do not really discuss the magnitude of the effects of Uber entry on congestion. In any case, it's a good read. 


Li, Z., Hong, Y., & Zhang, Z. (2016). Do Ride-Sharing Services Affect Traffic Congestion? An Empirical Study of Uber Entry. Available at SSRN: https://ssrn.com/abstract=2838043

Abstract:
Sharing economy platform, which leverages information technology (IT) to re-distribute unused or underutilized assets to people who are willing to pay for the services, has received tremendous attention in the last few years. Its creative business models have disrupted many traditional industries (e.g., transportation, hotel) by fundamentally changing the mechanism to match demand with supply in real time. In this research, we investigate how Uber, a peer-to-peer mobile ride-sharing platform, affects traffic congestion and environment (carbon emissions) in the urban areas of the United States. Leveraging a unique data set combining data from Uber and the Urban Mobility Report, we examine whether the entry of Uber car services affects traffic congestion using a difference-in-difference framework. Our findings provide empirical evidence that ride-sharing services such as Uber significantly decrease the traffic congestion after entering an urban area. We perform further analysis including the use of instrumental variables, alternative measures, a relative time model using more granular data to assess the robustness of the results. A few plausible underlining mechanisms are discussed to help explain our findings.

A good-looking video of the computer simulation model of the PNAS paper.


1 comment:

Laurent Franckx said...

I do have one or two questions about the study. First, I am not convinced by the choice of the instruments. I would expect that unemployment and the share of elder people is affecting the congestion in the city: unemployed people will for instance typically not drive during peak hours (to the extent that they can afford a car at all). In the case of elderly people, the situation is more complicated: as they may be driving their grandchildren to school (at least, that's quite common in Belgium), they can contribute to peak congestion. However, they generally have more discretion in the timing of their travel than the people of working age. A second issue, which is common to all urban models of course, is the delimitation of the area of study. If the study is limited to the inner cities, it may underestimate the impact of Uber drivers riding from their house in the suburbs in order to be available to pick up passengers. This being said, this is a very interesting study on a highly topical subject.