Outliers In Designed Experiments I-Classical Robust & Resistant Methods Ahmed

Abstract

The robust regression analysis and the designed experiments era two of the fastest growing fields in contemporary Statistics. There has been very little overlap between these fields. In designed experiments, designs were contrived for the efficient use of the least square estimates to maximize response, while in robust regression analysis, robust alternatives to the non-robust conventional least square estimates were developed. This paper, the first in the series of three, is an attempt to bridge the gap. It discusses classical robust, like el •stimators using a iti function developed by Huber, Hempel, & Tukey, and resistant estimators Of these for¬regression-analysis techniques to deal with outliers in the domain of designed experiments. Further, a Monte-Carlo simulation is carried out to appraise the efficiency of these methods in designed experiments and then a factorial experiment, with possible outliers, is reanalyzed. It is revealed that theee techniques, with some precautions and modifications, work excellent in designed experiments.