Verifying a dominant cause of output variation

Abstract

Finding the dominant cause(s) of variation in process improvement projects is an important task. Before trying to reduce variation in the dominant cause or mitigate the effect of variation in the dominant cause to reduce output variation, it is strongly recommended that we verify we have identified the true (dominant) cause. This article is about how best to verify we have correctly identified a dominant cause, as the existing literature does not properly answer this question. Although it may seem that a randomized controlled experiment is sufficient for this purpose, we show that experimental studies alone cannot provide all the required information. An experiment identifies whether a suspect is a cause of variation; however, we also require additional information (i.e., from observational studies) to determine whether it is dominant and not just significant. This article lists some viable composite study designs, assesses their relative merits, and recommends proper sample sizes. We also investigate how to systematically conduct a verification study in the era of smart manufacturing. Moreover, we provide a tangible example to illustrate our proposed procedure.

Keywords:

• observational study
• red $X$^®
• study design
• variation reduction
• verification study

Notes

1 Note that we treat $X$ as a random variable. Therefore, we implicitly assume that even if $X$ is deliberately changed on occasion (e.g., due to process controllers), we look at the process over a long enough time to justify modelling $X$ as a random variable.

Additional information

Notes on contributors

Mahsa Panahi

Mahsa Panahi is Ph.D. candidate in Statistics at the University of Waterloo, Canada. Her email address is [email protected].

Stefan H. Steiner

Dr. Stefan H. Steiner is Professor of Statistics at the University of Waterloo, Canada, in the Department of Statistics and Actuarial Science. His email address is [email protected].

Jeroen De Mast

Dr. Jeroen De Mast is Professor of Data-Driven Business Innovation at the University of Amsterdam, The Netherlands. He is also adjunct Professor of Statistics at the University of Waterloo, Canada. His email address is [email protected].