Estimation in the general linear model when the accuracy is specified before data collection (Q1074276)

The author uses a concept of accuracy saying that an estimator \({\hat \beta}\) of \(\beta\) is ''accurate with accuracy A and confidence c'', \(0<c<1\), if P(\({\hat \beta}\)-\(\beta\in A)\geq c\) for all \(\beta\). Note that A need only be any Borel set having an interior point at zero. Given a sequence \(Y_ 1,Y_ 2,..\). of independent vector-valued homoscedastic normally-distributed random variables generated via the general linear model \(Y_ i=X_ i\beta +\epsilon\), the k-dimensional parameter \(\beta\) is accurately estimated using a sequential version of the well-known maximum probability estimator developed by \textit{L. Weiss} and \textit{J. Wolfowitz} [Ann. Inst. Stat. Math. 19, 193-206 (1967; Zbl 0183.212); see also ''Maximum probability estimators and related topics.'' (1974; Zbl 0297.62015)]. The procedure also generalizes \textit{C. Stein}'s [Ann. Math. Stat. 16, 243-258 (1945; Zbl 0060.304)] fixed-width confidence sets to several dimensions. Some examples are given to illustrate the procedure.