Mathematics · Statistics and Probability
On the accuracy of empirical likelihood confidence intervals for<i>M</i>-Functionals
See on the mapAbstract
In this article, we consider the coverage accuracy of the empirical likelihood confidence interval for a class of M-functionals. It is shown that the coverage error in the chi-squared approximation to the distribution of the empirical likelihood ratio statistic is of order O(n −1), where n is the sample size. We further show that this coverage error can be reduced to the order O(n −2) by using the Bartlett correction. To implement the Bartlett correction in practice, the empirical Bartlett correction is given. A small simulation study is presented to demonstrate the efficacy of the Bartlett correction for the case of a Huber's functional.