Abstract
In this paper a new version of the empirical log-likelihood ratio function for quantiles is presented. It is based on a suitable smooth version of the usual empirical distribution function and yields confidence intervals which are simple to compute and very accurate, even in small samples.