Regression discontinuity estimates are sensitive to the choice of bandwidth and polynomial order. While there are data-driven methods for deciding bandwidth, I wonder if specific criteria help determine the optimal polynomial order for regression discontinuity (continuity-based approach).