The CER-optimal bandwidth in a [[Design Pattern II - Regression Discontinuity (RD)|regression discontinuity]] design minimizes the coverage error rate of the confidence interval, rather than the mean squared error of the point estimate as the [[MSE-optimal bandwidth]] does. It chooses $h$ so that the actual coverage of the nominal 95% CI is as close to 95% as possible. Confidence-interval coverage is more sensitive to bias than the point-estimate MSE is, so the CER-optimal bandwidth is typically **narrower** than the [[MSE-optimal bandwidth]]. We can use it when inference is the focus and the width and coverage of the confidence interval matter more than the precision of the point estimate. The canonical implementation resides in `rdrobust` as `bwselect = 'cerrd'` (Calonico et al., 2017).[^1] [^1]: Calonico, S., Cattaneo, M. D., Farrell, M. H., & Titiunik, R. (2017). rdrobust: Software for regression-discontinuity designs. *The Stata Journal*, 17(2), 372–404. > [!info]- Last updated: May 13, 2026