We have proposed a new confidence interval for R = P(Y < X) when X and Y are two independent
generalized exponential random variables with a common but unknown scale parameter. This interval
is based on a modified signed log-likelihood ratio statistic that has accuracy O(n−3/2), where n is the
sample size. We have compared the performance of this interval with three other known intervals
including one bootstrap based interval. Simulations show that the proposed interval gives the best
performance with respect to coverage probabilities for all n. The bootstrap interval gives the best
performance with respect to coverage lengths but not for all n. We should note however that its
construction is computationally more demanding than others, especially when n is not small.