Loading Events
  • This event has passed.

Jackknifing Methods for Simulation Output Analysis

Abstract:

The output from a steady-state simulation is often difficult to analyze – for example, consecutive customer waiting times are never independent or normally distributed and analysis is not amenable to standard statistical methods. This issue presents challenges when it comes to reporting confidence intervals (CIs) for the corresponding parameters of interest, e.g., the mean waiting time. The key to obtaining valid CIs for the mean is to use good estimators for the sample mean’s variance. To this end, we discuss new variance estimators that incorporate various jackknifing tricks. We show via analytical and Monte Carlo methods that the new estimators have smaller mean-squared error than their predecessors – sometimes significantly smaller.

Bio:

Dave Goldsman is a Professor in the School of Industrial and Systems Engineering at Georgia Tech. He received his Ph.D. from Cornell, and has held visiting positions at Cornell, Northwestern, Oklahoma, Boğaziçi, Özyeğin, and Sabancı Universities. His research interests include simulation output analysis, statistical ranking and selection methods, and medical and humanitarian applications of operations research. Dave received the INFORMS Simulation Society’s Distinguished Service Award in 2002. He also received a Fulbright fellowship in 2006 to lecture at Boğaziçi and Sabancı Universities in Istanbul, Turkey. Dave is a Fellow of the Institute of Industrial Engineers.

View on site | Email this event