The problem of the initial transient (again), or why MSER works

In a comprehensive study of methods for dealing with the problem of the initial transient, Hoad et al. (2008) determined that the MSER (White, 1997) was an efficient and effective truncation rule appropriate for automation. In this paper, we suggest that the MSER works well because it minimizes an approximation to the mean-squared error in the estimated steady-state mean. Using the example of an M/M/1 queue, we provide a clear statement of the problem in both the time and frequency domains, distinguishing between the biasing effects of initialization and autocorrela-tion. We also demonstrate that, as a result of autocorrela-tion, the objective of minimizing initialization bias is not exclusively a matter of determining the most representative initial condition. This observation further argues against the replication/deletion approach to output analysis.