On exact interval estimation for the odds ratio in subject-specific table

We first show that the infimum coverage probability of a commonly used asymptotic confidence interval for the odds ratio is much less than the nominal level and then propose three exact intervals, which are also valid under mixed-effect models.