PROBLEMS IN THE STATISTICAL ANALYSIS OF SIMULATION EXPERIMENTS: THE COMPARISON OF MEANS AND THE LENGTH OF SAMPLE RECORDS

摘要

Statistical analysis of simulation experiments containing autocorreLated time series continues to pose problems for researchers. One important aspect of such experiments is determining the stability of sample means. This Memorandum shows how to estimate the lengths of sample records needed to use certain large sample results in measuring stability. It also describes analogies between autocorrelated data and independent observations, and suggests a way to test the difference of the mean of two simulation experiments.nSection II shows how the variance of the sample mean relates to the spectrum of the generating process.nEstimation of the quantities of interest is described in Sec. III. Stress is placed on the need to obtain a well-resolved estimate of the variance of the sample mean;and the care necessary to avoid significant bias is explained.