Performance comparison of MSER-5 and N-Skart on the simulation start-up problem

• DOI: 10.5555/2433508.2433624
• Publication Date: 2010-12-05
• Authors: Anup C. Mokashi; Jeremy J. Tejada; Saeideh Yousefi; Ali Tafazzoli; Tianxiang Xu; James R. Wilson; Natalie M. Steiger
• Keywords: Truncation (statistics); Statistic

We summarize some results from an extensive performance comparison of the procedures MSER-5 and N-Skart for handling simulation start-up problem. assume a fixed-length simulation-generated time series which point confidence-interval (CI) estimators steady-state mean are sought. uses data-truncation that minimizes half-length usual batch-means CI computed truncated data set. randomness test to determine beyond spaced batch means approximately independent each other simulation's initial condition; then using nonspaced means, exploits separate adjustments account effects on distribution underlying Student's t-statistic arising skewness autocorrelation means. In most problems, N-Skart's estimator had smaller bias than MSER-5; moreover in all cases, outperformed MSER-5.