Generalized linear models for quality-improvement experiments

Since the early 1980s, industry has embraced the use of designed experiments as an effective means for improving quality. For quality characteristics not normally distributed, the practice of first transforming the data and then analyzing them by standard normal-based methods is well established. There is a natural alternative called generalized linear models (GLMs). This paper explains how GLMs achieve the intended goal of transformation while at the same time giving a wider class of models that can handle a range of applications. Moreover, the same iterative strategy for data analysis that has been developed for normal data over the years, namely, the alternation between model selection and model checking, extends easily to analyses with GLMs. The paper illustrates the ability of GLMs to handle many different types of data by the re-analysis of three quality-improvement experiments.