Causal Inference Approach to Matching in Two-Sided Marketplaces
ABOUT OUR SPEAKER
Data scientist, social scientist, statistician, and software developer. Sean mostly specializes in methods for solving causal inference and business decision problems, and particularly interested in building tools for practitioners working on real-world problems. He is a generalist, who likes to hang out with people from many fields and borrow as many ideas as possible. He has collaborated with computer scientists, economists, political scientists, statisticians, machine learning researchers, and business school scholars. It’s fun for Sean to jump around a bit, continue learning new things and make connections between fields.
Background: Ph.D. student at NYU’s Stern School of Business, ex-research scientist and manager on Facebook’s Core Data Science Team and currently head of the Rideshare Labs team at Lyft. To know more about Sean check his page.
ABOUT THE TALK
Improving two-sided marketplaces like Lyft's ridesharing business is accomplished when we implement better matching algorithms that create more efficient allocations. However, marketplaces generate complex causal graphs which make counterfactuals difficult to reason about and estimate. I describe the causal effects of matching we must estimate to allocate drivers to requests, and the potential confounding problems that bias estimates from more standard machine learning approaches. We will then discuss how causal inference ideas can be applied in related settings like search, ranking, and recommender systems.