Benchmarking Estimators for Natural Experiments: A Novel Dataset and a Doubly Robust Algorithm