An analysis of five simulation methods for determining the number of replications in a complex Monte Carlo study (Q1091792)

Five simulation methods are summarized and compared for estimating a multiparameter mean \(\mu\) within a given accuracy: a two-stage extension of the one-dimensional procedure of Stein, a three-stage extension of a procedure due to Hall, a purely-sequential procedure due to Finster, a continuously-monitoring procedure also due to Finster, and a purely- sequential maximum-eigenvalue procedure developed by Srivastava. The quality of the methods are judged by the number of replications needed to be within the given accuracy. The comparison is performed by simulation on the assumption that \(\mu\) is the mean of a p-dimensional normal distribution.