Chatterjee, Shoutir Kishore; Chattopadhyay, Gaurangadeb - In: Statistics & Probability Letters 21 (1994) 4, pp. 263-269
It has been shown that there does not exist any uniformly most accurate unbiased or uniformly most accurate invariant confidence set for [theta] at level 1 -- [alpha] where the sample is drawn from a rectangular ([theta], [theta] + 1) distribution. So Pratt's (1961) statement in this connection...