A three-class variance swapping technique for simulation experiments

The 2^k factorial design is widely used in simulation experiments involving several factors in which the mean effect, the main effects, and the interaction effects of these factors are estimated. Previous research has differentiated all effects into effects of interest and no interest. Traditional variance swapping techniques are then used to increase the accuracy of the estimators of the effects of interest. This article allows one to make a finer distinction among levels of interest. We allow the effects to be divided into three groups, say, those of primary interest, secondary interest, and little interest. We do this by dividing the class previously labelled as being of no interest into subclasses of ‘secondary interest’ and ‘little interest’. Then we reallocate the relative variances of the estimators of the effects in the two subclasses. We call this approach a three-class variance swapping technique.