Analysis of 2-Level Factorial Experiments by Bootstrapping GLM for Gamma Distributed Responses

Abstract

In recent years, the use of statistically designed experiments is greatly expanded into many new application areas of engineering, sciences and product and process improvement. Generalized linear models (GLM) unify a class of regression models for categorical, discrete and continuous response variables is popularly used by many researchers for analysing factorial experiments. It was found that GLM performed better than other traditional methods like ANOVA method and traditional Transformation method. The present study discusses some bootstrap methods and applies them for the analysis of 2n full factorial experiments and their regular fractions for Gamma distributed response variables for some specified parameter combinations. The comparison of bootstrap based methods for GLM is based on the coverage and expected length of confidence intervals (CI) for the analysis of 2n full factorial experiments when the response variable is Gamma distributed. The study is extensively simulation based.