A study on Taxi Drivers

Taxi drivers provide an ideal setting for testing various models of labor supply, however given their long history there is actually very few studies investigating their behaviour (this has probably shifted a bit now that Ridesharing platforms have gone mainstream). I saw this as an opportunity and during my time at the SFMTA Taxi Division, I was able to bring in an economics PHD student and close friend, Vincent Martin, and partner with him to take a deeper dive at the data.

Comparing data between New York and the SFMTA, we conducted an analysis of taxi trip data to reconciles previous inconsistencies in evidence for reference dependence in taxi drivers. By testing for a particular non-linear relationship between shift income and drivers’ hazard of stopping, we identified behavior that is consistent with Prospect theoretic (S-shaped) reference dependence as opposed to the more extensively examined loss aversion model of reference dependence. This particular model of reference dependence in this setting allows me to estimate individual driver reference points without an explicit functional form or ex-ante assumptions about the existence of a reference point.

In this project, I served as a data expert and SME on the underlying data, and taxi operations.

Journal of Economic Behavior & Organization

Volume 144, December 2017, Pages 166-187

When to quit: Narrow bracketing and reference dependence in taxi drivers

Highlights
  • I reexamine the question of whether taxi drivers are reference dependent in daily labor supply.
  • Previous work focused on rejecting a model of loss aversion.
  • Evidence is consistent with reference point in sensitivity (S-shaped preferences).
  • Results are consistent across large datasets from San Francisco and New York City.

Abstract

Taxi drivers provide an ideal setting for testing various models of labor supply. Despite this, the literature studying driver behavior has produced mixed evidence of which labor supply model best applies to driver labor supply. Using a novel analysis and large datasets of taxi trips from two different cities, this paper provides a unifying analysis which reconciles previous inconsistencies in evidence for reference dependence in taxi drivers. By testing for a particular non-linear relationship between shift income and drivers’ hazard of stopping, I identify behavior that is consistent with Prospect theoretic (S-shaped) reference dependence as opposed to the more extensively examined loss aversion model of reference dependence. This particular model of reference dependence in this setting allows me to estimate individual driver reference points without an explicit functional form or ex-ante assumptions about the existence of a reference point.