Split distribution method for constructing confidence intervals for simulation output analyses

Abstract

A new procedure to construct a confidence interval for the mean of a simulated process is proposed. The new procedure uses approximate partial variance and it is capable of adjusting itself to compensate for the skewness of the underlying population. A Monte Carlo experiment compares the proposed procedure with some other procedures using different sample sizes and underlying distributions. The result shows that the proposed procedure generally provides a higher probability of coverage with a shorter confidence interval width than the other procedures.