Matching in Labor Marketplaces With Experiential Learning

Abstract

Online labor marketplaces match workers to short-term jobs. Though the quality of hired workers directly impacts the quality of matches, marketplace reputational mechanisms are often inadequately informative, prompting buyers to assess and screen workers directly. We study the platform intermediary's problem of matching workers to jobs when buyers learn worker quality experientially. As a basic trade-off, platforms can either explore new matches to expedite buyers' experiential learning or maximize short-term match quality. Additionally, over-exploring incurs efficiency losses, because every new match incurs new setup costs as workers tailor services to buyers. We develop a structural estimation method to infer hidden worker quality from buyers' hiring decisions and improve platform matching. Our empirical analysis of 1.2M hiring decisions on a major online freelancer platform demonstrates that, in contrast to visible ratings, experiential learning explains approximately 87% of the variation in applicants' utility to buyers. Based on our estimates, we propose improved platform matching policies that importantly calibrate between promoting existing matches and exploring new workers. They increase buyer welfare by up to 45-47% of gross revenue. We observe high value from exploration: in the two markets we study, greedy policies under-explore and therefore underperform revenue-wise by 18.9% and 8.7%.

Keywords: Choice modeling and estimation, Empirical operations management, Information friction, Market intermediaries, Marketplace design, Matching with costly screening, Moment inequalities, Online labor markets