Sequential and two-stage procedures for selecting the better exponential population covering the case of unknown and unequal scale parameters (Q797238): Difference between revisions

Let \(\pi_ 1\) and \(\pi_ 2\) be two exponential location-scale populations; i.e., their densities are of the form \(f(x;\mu,\sigma)= \sigma^{-1}\exp(-\sigma^{-1}(x-\mu))I(x>\mu).\) The parameters \(\mu_ i\), \(\sigma_ i\), \(i=1,2\), are unknown. Based on samples from \(\pi_ 1\) and \(\pi_ 2\) it is desired to select the population with the largest value of \(\mu\) subject to the condition that the probability of correct selection has a pre-chosen lower bound provided \(| \mu_ 1-\mu_ 2| \geq \delta^*\), where \(\delta^*>0\) is also pre-chosen. When \(| \mu_ 1-\mu_ 2| =\delta^*\) the parameters are said to have ''least favorable configuration'' (LFC). Both the case \(\sigma_ 1=\sigma_ 2\) and the case of completely unknown \(\sigma_ i\) are considered. In the former, let \(\sigma\) be the (unknown) common value of the \(\sigma_ i\), and let \(C^*=C^*(\sigma)\) be the required fixed sample size if one would know \(\sigma\). A two-stage procedure is proposed, involving a random sample size N, with the property that under the LFC, \(EN/C^*\to 1\) as \(\delta^*\to 0.\) Next, a purely sequential procedure is proposed with the stronger property that \(EN-C^*\) converges to a constant as \(\delta^*\to 0\) (under LFC). Similar procedures and results are obtained in the case that the \(\sigma_ i\) are completely unknown.