|
Abstract:
|
Classical design of experiments are not applicable at ride-hailing platforms, like Uber, Lyft and DiDi, since such two sided markets are highly spatiotemporal, and correlations are everywhere. Experiment units are not independent with each other, riders share the same pool of drivers, and drivers share the same pool of riders. In this talk, we will talk about the novel designs of experiments for key platform strategies (order dispatch, driver repositioning, riders' promotions, and drivers' incentives), and how we analyze the results using causal inference.
|
