摘要

Constructing confidence intervals with high-order coverage probability precision is more difficult for one-sided intervals than for two-sided intervals. Many existing methods can achieve precision of order n(-2) for two-sided intervals but only n(-1/2) for one-sided intervals. Through a creative use of adjusted empirical likelihood, we develop a new procedure that attains coverage precision of order n(-3/2) for one-sided intervals while retaining order n(-2) precision for two-sided intervals. We provide detailed comparisons of the asymptotic properties of the new method and those of representative existing methods. Simulation results show that the new method offers many advantages.