When the Nearest Driver Loses: Order Allocation as Incentive Design in Ride-Hailing

Abstract

This paper studies the hidden dispatch rule using 3.94 million ride-hailing transactions and 538 million GPS pings from a large Chinese city. We reconstruct feasible candidate drivers immediately before dispatch and compare the selected driver with rejected feasible alternatives for the same passenger request. The closest feasible driver receives the order in only 10.45\% of valid events, although pickup distance remains strongly penalized. Assignment also depends on driver's idle/in-trip status. For the platform assigning a given order, a driver's recent activity on that platform raises assignment probability, while recent activity on other platforms lowers it. This contrast identifies platform-specific priority rather than a generic activity premium. The rule is also contestability-dependent: drivers more attached to the assigning platform receive higher baseline access, while multihoming drivers receive stronger marginal rewards for recent activity on assigning-platform; lower-share platforms rely more on this activity margin. Machine-learning validation confirms that distance, driver state, and platform-specific activity improve winner ranking in hold-out choice sets. The findings provide direct order-level evidence on allocation primitives that ride-hailing models typically assume and show that dispatch algorithms allocate priority over earning opportunities by combining pickup efficiency, driver engagement, and platform attachment.