Service Quality on Online Platforms: Empirical Evidence about Driving Quality at Uber

with Susan Athey and Juan Camilo Castillo

Revise and Resubmit, Management Science
Last Updated May 2023

Online marketplaces have adopted new mechanisms for quality control that can accommodate a flexible pool of providers, with unclear effects on overall service quality. We focus on ride-hailing: pre-screening, which prevailed in taxi markets, has been diminished in favor of automated quality measurement, incentives, and nudges. Using telemetry data, an objective measure of quality, we show that UberX drivers perform better than UberTaxi drivers in multiple dimensions, including according to a score of quality that reflects the preferences of UberX riders. We then explore a variety of mechanisms that affect driver behavior, establishing that UberX drivers respond to user preferences, nudges, and information about driving quality. We use data from a randomized experiment to show that informing drivers about their past behavior improves quality, especially for low-performing drivers.

Tipping Is More About Who Is Giving Than Who Is Receiving: Evidence from a Nationwide Study

with Uri Gneezy, John A. List, and Ian Muir
Last Updated December 2022

Even though social preferences affect nearly every facet of life, there exist many open questions on the economics of social preferences in markets. We leverage a unique opportunity to generate a large data set to study social preferences through the lens of a nationwide tipping study on the Uber platform. Our study generates data from more than 40 million trips, which allow us unique insights into the underlying motives for tipping. Even though tips are made privately, and without external social benefits or pressure, more than 15% of trips are tipped. Importantly, we find that rider effects account for about three times more of the observed tipping variation than driver effects, such that the demand-side explains much more of the observed tipping variation than the supply-side. Our data allow us to classify riders into three categories, showing that nearly 60% of people never tip, and only 1% of people always tip.

Online Appendix – Demand effects and interactions between home ZIP demographics

Design and Analysis of Cluster-Randomized Field Experiments in Panel Data Settings

with Ali Hortacsu, John A. List, Ian Muir, and Jeffrey Wooldridge

Reject and Resubmit, Journal of Econometrics
Last Updated July 2019

Field experiments conducted with the village, city, region, or even country as the level of randomization are becoming more common in the social sciences. Subsequent data analysis may be complicated by the constraint on the number of clusters in treatment and control. Through a battery of Monte Carlo simulations, we examine best practices for estimating unit-level treatment effects in cluster-randomized field experiments, particularly in settings that generate short panel data. In most settings we consider, unit-level estimation with unit fixed effects and cluster-level estimation weighted by the number of units per cluster tend to be robust to potentially problematic features in the data while giving greater statistical power. Using insights from our analysis, we evaluate the effect of a unique field experiment: rolling out tipping across markets on the Uber app. Beyond the import of showing how tipping affects aggregate market outcomes, we provide several insights on aspects of generating and analyzing cluster-randomized experimental data when there are constraints on the number of experimental units in treatment and control.

Quadratic Voting in Finite Populations

with E. Glen Weyl
Last Updated December 2019

We study the performance of the Quadratic Voting (QV) mechanism proposed by Lalley and Weyl (2016) in finite populations of various sizes using three decreasingly analytic but increasingly precise methods with emphasis on examples calibrated to the 2008 gay marriage referendum in California. First, we use heuristic calculations to derive conservative analytic bounds on the constants associated with Lalley and Weyl’s formal results on large population convergence. Second, we pair numerical game theory methods with statistical limit results to computationally approximate equilibria for moderate population sizes. Finally, we use purely numerical methods to analyze small populations. The more precise the methods we use, the better the performance of QV appears to be in a wide range of cases, with the analytic bounds on potential welfare typically 1.5 to 3 times more conservative than the results from numerical calculation. In our most precise results, we have not found an example where QV sacrifices more than 10% of potential welfare for any population size. However, we find scenarios in which one-person-one-vote rules outperform QV and also show that convergence to full efficiency in large populations may be much slower with fat tails than with bounded support. The results suggest that in highly unequal societies, 1p1v or QV with artificial currency may give superior efficiency to QV with real currency.

Work in Progress

Coercive Sterilization in India

with Ruchi Mahadeshwar