Design of Experiments (DOE) is a very powerful statistical methodology, especially when used with linear regression analysis. The use of ordinary least squares (OLS) estimation of linear regression parameters requires the errors to have a normal distribution. However, there are numerous situations when the error distribution is non-normal and using OLS can result in inaccurate parameter estimates. Robust regression is a useful and effective way to estimate the parameters of a regression model in the presence of non-normally distributed residuals. An extensive literature review suggests that there are limited studies comparing the performance of different robust estimators in conjunction with different experimental design sizes, models, and error distributions. The research in this thesis is an attempt to bridge this gap. The performance of the popular robust estimators is compared over different experimental design sizes, models, and error distributions and the results are presented and discussed. The results evaluating the performance of the robust estimator with OLS after performing Box-Cox transformation are also presented in this research.
Industrial and Systems Engineering (MS)
Department, Program, or Center
Industrial and Systems Engineering (KGCOE)
Kumar, Pranay, "Experimental Design and Robust Regression" (2017). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus