Combined Ranking and Selection with Control Variates

Nelson and Staum derived R&S procedures that employ control-variate estimators (CVs) instead of sample means to obtain more statistical efficiency. However, control-variate estimators require more computational effort than sample means, and effective controls must be identified. We present a new CV screening procedure to avoid much of the computation cost. We also present a two-stage CV combined procedure which captures the ability to eliminate inferior systems in the first stage and the statistical efficiency of control variates for selection in the second stage. An empirical evaluation is provided