Applying the Regression-Discontinuity Design to Infer Causality with Non-Random Assignment