Sequentual first-crossing look-ahead procedure for selecting a population with the largest mean in normal-normal model (Q2010176): Difference between revisions

The article concerns a significant contribution to the problem of statistical selection of a population with the largest mean value. Let \(p\) independent normal populations \(\xi = (\xi_1,\ldots,\xi_p)\) be considered. They have, by assumption, random means \(\vartheta = (\vartheta_1,\ldots ,\vartheta_p)\) such that \(\xi_i\sim {\mathcal N}(\vartheta_i, \sigma^2)\), where \(\vartheta_i\sim {\mathcal N}(\mu,\tau^2)\) for \(i = 1,\ldots , p\), where \(\sigma^2\), \(\tau^2\) are known populations and prior variances, respectively; \(\mu\) is known prior mean and \(\{\vartheta_i\}\) are mutually independent. It is assumed a Bayesian setting where \(\{\vartheta_i\}\) are drawn before the current experiment, after which they are fixed until the experiment ends. The populations \(\xi_i\) are observed only. The problem is to determine the population \(\xi_i\) with the largest mean \(\vartheta_i\geq \vartheta_j\), \(j = 1,\ldots, p\) based on as few observations as possible. A sequential selection procedure, called the first-crossing look-ahead (FCLA) is proposed and investigated. A numerical performance comparison (in terms of sample size) with the Kao-Lai [\textit{S. C. Kao} and \textit{T. L. Lai}, Commun. Stat., Theory Methods A9, 1657--1676 (1980; Zbl 0454.62075)] efficient selection procedure for an indifference-zone setting of the selection problem is presented.