Study on outliers in factorial design of experiments

Abstract

In this dissertation, we discuss the sensitivity of optimum balanced 2m fractional designs of resolutiun V to outliers using the measure suggested by Box and Draper (1975). In addition, we define a new measure of robustness of design with respect to outliers , which we called the "0ut- lier-insensitivity factor". Various properties of these two measures of robustness of designs (with respect to outliers) will be outlined. The optimum balanced resolution V plans are known in the literature for practical values of m and the number of observations. For Srivastava-chopra optimum balanced resolution V plans for 2m factorial design we first simplified expressions for these two measures of insensitivity. We then find their values for optimum designs when 4<m<8, and it will be shown that sensitivities to outliers of these optimum designs are low,