How Uber affects public transit ridership is a relevant policy question facing cities worldwide. Theoretically, Uber's effect on transit is ambiguous: while Uber is an alternative mode of travel, it can also increase the reach and flexibility of transit's fixed-route, fixed-schedule service. We use a difference-indifferences design to measure the effect of Uber on public transit ridership. The design exploits variation across U.S. metropolitan areas in both the intensity of Uber penetration (as measured using data from Google Trends) and the timing of Uber entry. We find that Uber is a complement for the average transit agency. This average effect masks considerable heterogeneity, with Uber being more of a complement in larger cities and for smaller transit agencies. Comparing the effect across modes, we find that Uber's impact on bus ridership follows the same pattern as for total ridership, though for rail ridership, it is a complement for larger agencies.

No abstract

This paper shows that a judiciously designed toll applied to a portion of the lanes of a highway can generate a Pareto improvement even before the resulting revenue is spent. I obtain this new result by extending a standard dynamic congestion model to reflect an important additional traffic externality recently identified by transportation engineers: additional traffic does not simply increase travel times, but also introduces frictions that reduce throughput. In particular, I show that as long as some rich drivers use the highway at the peak of rush hour, then adding tolls to a portion of the lanes yields a Pareto improvement. To confirm the relevance of this theoretical possibility in practice, I use survey and travel time data to estimate the joint distribution of driver preferences over arrival time, travel time, and tolls, and use these results to estimate the effects of adding optimal time-varying tolls. I find that adding tolls on up to half of the lanes yields a Pareto improvement, and that the social welfare gains of doing so are substantial-up to $1,740 per road user per year. Keywords: congestion pricing, value pricing, Pareto improvement * I am especially grateful for the guidance and support that Gary Becker and Eric Budish have given me. I am also grateful for helpful feedback from

No abstract