Selecting the better binomial population under a budget constraint (Q1058237)

Consider the problem of choosing the better of two binomial populations based on fixed numbers of observations from each, where the better population is taken to be the one with the largest probability of success. In the earlier paper ''Selecting the better binomial population with unequal sample sizes'', Commun. Stat., Theory Methods 14, 123-158 (1985)], the author has shown that the intuitive decision rule, choose the population corresponding to the largest observed relative frequency of success, although minimax for equal sample sizes, possesses poor properties for samples of unequal size. In the present paper, the author extends the problem to include a budget constraint. It is assumed that each observation made on each population has a fixed, known cost, which can be different for the two populations. It is further assumed that the total cost of all observations made cannot exceed a budgeted amount B. The problem of optimal allocation of observations on the two populations, in order to maximize a measure of the probability of correct selection, is considered. The author proposes a certain class of decision rules, and within this class he tries to find the minimax solution to the selection problem. Solutions are given for small samples, and large sample approximations are also given. It is shown that a 70\% reduction of the probability of incorrect selection is possible by using an unequal rather than equal allocation.