A comparison study of two tests for detecting initialization bias in simulation output

This paper presents a comparison study of two statistical tests for detecting initialization bias in discrete-event simulation output. These two tests are the so-called optimal test and rank test developed and presented by Schruben et al. in 1983 and Vassilacopoulos in 1989 respectively. In the comparison experiments, artificially generated stochastic sequences are used as input to the tests. Such a sequence is obtained by embedding specified initialization bias into a unbiased stationary stochastic sequence. The results of the experiments have shown that both tests do perform satisfactorily in a similar way. It is noted that the optimal test is more sensitive to initialization bias, and therefore has consistently outperformed the rank tests when significant initialization bias is present in the tested sequences. However, since the rank test is simpler to implement and easier to use than its counterpart, it is expected that it will also find wider acceptance and application in the future.