1: \begin{abstract}

2:   We consider the problem of constructing honest confidence intervals (CIs) for

3:   a scalar parameter of interest, such as the regression discontinuity

4:   parameter, in nonparametric regression based on kernel or local polynomial

5:   estimators. To ensure that our CIs are honest, we derive and tabulate novel

6:   critical values that take into account the possible bias of the estimator upon

7:   which the CIs are based. We show that this approach leads to CIs that are more

8:   efficient than conventional CIs that achieve coverage by undersmoothing or

9:   subtracting an estimate of the bias. We give sharp efficiency bounds of using

10:   different kernels, and derive the optimal bandwidth for constructing honest

11:   CIs. We show that using the bandwidth that minimizes the maximum mean-squared

12:   error results in CIs that are nearly efficient and that in this case, the

13:   critical value depends only on the rate of convergence. For the common case in

14:   which the rate of convergence is $n^{-2/5}$, the appropriate critical value

15:   for 95\% CIs is 2.18, rather than the usual 1.96 critical value. We illustrate

16:   our results in a Monte Carlo analysis and an empirical application.

17: \end{abstract}