A seminar at Center for Mathematical Sciences, July 29, Wednesday, at 4:30pm

Location：Center for Mathematical Sciences, Room 813（创新研究院恩明楼813室）

Speaker:Weizhen Wang, Wright State University, USA

Title:Asymptotic versus Exact Confidene Intervals

Abstract:Confidence interval is used to capture the parameter of interest. The probability that the interval includes the parameter is called the coverage probability and it is a function of parameters. The coverage probability is required to be at least the prespecified confidence level at all parameter configurations. In statistical practice, confidence interval is obtained as soon as the parameter estimator is shown to be asymptotically normally distributed, and as a result the “validity” of confidence interval is claimed. In this talk, we first evaluate the coverage probability of some commonly used asymptotic intervals and show that their infimum coverages are below the give level even with large samples.Then we introduce some optimal exact confidence intervals with a coverage probability function that is always at least the prespecified confidence level.