Six Sigma Rolled Throughput Yield

Click to increase image sizeClick to decrease image size

ACKNOWLEDGMENTS

The author wishes to thank Søren Bisgaard, Chuck Cox, Betsy Wolf-Graves, Chet Harmer, and Xavier Tort-Martorell for helpful comments that contributed to the development of the ideas discussed in this article.

Notes

^*Some people have claimed that designed experiments are only useful for processes in statistical control. This is not accurate. For processes that may lack statistical control, it is important to block the experiment and to randomize within blocks. In such cases, blocking allows effects to be judged relative to the smaller variability within blocks whenever blocks can be made sufficiently small that most of the process instabilities occur between blocks. Meanwhile, if runs are not randomized within blocks, there is a substantial risk that any lack of process control within blocks will be confounded with a main effect of interest. This, in turn, means that attempted use of the apparent results of the experiment may not yield the anticipated effect. Conversely, if runs are randomized within blocks, a lack of process control within blocks will likely be more appropriately identified either as noise or as an implausible high-order interaction.