BETTER SADDLEPOINT CONFIDENCE-INTERVALS VIA BOOTSTRAP CALIBRATION

Confidence interval construction for parameters of lattice distributio ns is considered. By using saddlepoint formulas and bootstrap calibrat ion, we obtain relatively short intervals and bounds with O(n(-3/2)) c overage errors, in contrast with O(n(-1)) and O(n(-1/2)) coverage erro rs for normal theory intervals and bounds when the population distribu tion is absolutely continuous. Closed form solutions are also provided for the cases of binomial and Poisson distributions. The method is il lustrated by some simulation results.