Analyzing Web-based simulation data

Abstract

The article deals only with simulation models that have stochastic, or random input. Classical statistical methods for independent observations assume that each observation carries the maximum information, and therefore they compute the smallest confidence interval. Since stationary simulation output data carries less information, the confidence interval resulting from applying classical statistical computations using autocorrelated observations would be too small. This would lead one to conclude the parameter estimate is much more precise than is actually the case. To get around this problem, several methods have been suggested in the output data analysis literature. Two of the most widely accepted methods are: 1) the method of independent replications; and 2) the method of batch means. Both methods try to avoid autocorrelation by breaking the data into "independent" segments. The sample means of these segments are considered i.i.d, and used to calculate confidence intervals. In the first method, several independent runs are executed. In the second method, a long simulation run is executed and divided into several "nearly uncorrelated" batches. The article specifically examines the Java Simulation (JSIM) Web based environment which has evolved to incorporate component based technology. If component based technology succeeds, the long hoped for gains in software development productivity may finally be realized.