The lower confidence limits for response probabilities based on binary response data under the logistic response model are considered by saddlepoint approach. The high order approximation to the conditional distribution of a statistic for an interested parameter and then the lower confidence limits of response probabilities are derived. A simulation comparing these lower confidence limits with those obtained from the asymptotic normality is conducted. The proposed approximation is applied to two real data sets. Numerical results show that the saddlepoint approximations are much more accurate than the asymptotic normality approximations, especially for the cases of small or moderate sample sizes.

The lower confidence limits for response probabilities based on binary response data under the logistic response model are considered by saddlepoint approach. The high order approximation to the conditional distribution of a statistic for an interested parameter and then the lower confidence limits of response probabilities are derived. A simulation comparing these lower confidence limits with those obtained from the asymptotic normality is conducted. The proposed approximation is applied to two real data sets. Numerical results show that the saddlepoint approximations are much more accurate than the asymptotic normality approximations, especially for the cases of small or moderate sample sizes.