The effects of batching on the power of the test for frequency domain methodology

The author considers an implementation of frequency domain methodology (FDM) where input factor levels are held fixed in batches. In addition, the simulation output series is batched and the corresponding batch means are used as observations in the FDM factor screening analysis. For a certain class of metamodels, steady-state results are provided. Results are also provided for batch sizes chosen to reduce serial correlation in output series, but not necessarily to achieve steady state. An example is presented which illustrates cases where fixed factor batching (FFB) improves the power of the FDM hypothesis test and cases where it does not. For cases where FFB is beneficial with respect to the power of the test, it is shown that batching to an intermediate batch size may be better than batching to achieve steady state