Using null effects estimators in the analysis of simulation output

In a simulation experiment there are certain null effects whose true values are known to be zero. This paper investigates various estimators of the null effects. We apply these estimators of zero in two ways. First, null effects estimators are used to construct confidence intervals. We derive a null effects version of the batch means confidence interval that has certain desirable properties. Second, null effects estimators are used as control variates in variance reduction schemes. The achieved variance reductions are modest, but we present some interesting applications; for example, the output from one simulation model can be used to compute control variates for the output of another.