Robust estimation with latin hypercube sampling: a central limit theorem for Z-estimators

Abstract: Latin hypercube sampling (LHS) is a stratified samplingmethod widely used in computer experiments. In thiswork, we extend convergence results on the sample meanwith Latin hypercube sampling to the class of Z -estimators,gathering all estimators that can be written as zeros of asample mean function. In particular, the asymptotic vari-ance of this estimate is obtained. This asymptotic vari-ance is shown to be lower using LHS than using classicindependent and identically distributed sampling. A Cen-tral Limit theorem for Z -estimators under LHS is alsogiven.