View Article - PDF     View Article - Webpage     Dataset (text)

Narrated PowerPoint presentations on Capability Analysis are available here via the American Society for Quality (ASQ)

Title: Capability Analysis with Non-Normal Data
Author: Keith M. Bower
Publication Source: American Society for Quality, Six Sigma Forum, March 2005

Abstract

This article addresses the assumptions and methodology behind obtaining valid capability estimates when the assumption of sampling from a Normal distribution is invalid.

Notes

For this paper I wanted, primarily, to address the concept of  reasonably assessing any distributional form for a given process. Using the Phase I/II philosophy as described in the referenced paper provides a reasonable framework.

Regarding whether it is "better" to use a particular probability model (e.g. the Weibull) or transforming the data to obtain, say, approximate Normal results seems a moot point. As George Box would say, all models are wrong - some are useful. The procedure of finding a "useful" model for the dataset (when a stable distribution has been witnessed) seems sensible.

To download Adobe Acrobat (for free) click here.