The robustness of Stein's two-stage procedure (Q796203)

A two stage confidence interval procedure for estimating a normal mean was suggested by \textit{C. Stein} in Ann. Math. Stat. 16, 243-258 (1945; Zbl 0060.304). The author is interested in so called ''criterion robustness'' of this procedure. More precisely, he shows that this procedure is robust even under the Edgeworth series model, i.e. when the probability distribution can be approximated by the first four terms of Edgeworth series. To illustrate the results, a Monte Carlo study was performed which involves sampling from the gamma population. To study the behaviour of Stein's procedure for non-normal populations, the coverage probabilities and ASN have been computed. The numerical results confirm the idea that Stein's procedure is remarkably robust against the moderate departures from normality.