Analysis of Simulation Experiments Under Normality Assumptions

Abstract

The use of control variables in computer simulations to improve the performance of estimators of model parameters is widely advocated. The technique that usually seems to be suggested is a distribution free approach based on least squares. In this note it is suggested that, frequently, normality assumptions apply which allow a more detailed, as well as simpler, statistical analysis. It is also pointed out that though the analysis is similar in many ways to an ordinary regression analysis, there are certain important differences, especially when variances are to be estimated.