Demasking large dummy effects approach in revealing important interactions in Plackett-Burman experimental design

Abstract

This paper presents a newly developed demasking large dummy effects (DDE) approach for identification of significant factor interactions in Plackett-Burman experimental design. Dummy factors, as imaginary factors, are not expected to contribute to the model, so the appearance of their high effect values could indicate the significance of the interactions they are confounded with. DDE approach consists of the screening of the large estimable effects, the calculation of the interactions that contribute the most to the large dummy effects from the alias matrix, and a single pass of all subsets regressions including preselected interactions and main factors for the final model tuning. Tested on experimentally obtained data from a robustness analysis of a method for the determination of raloxifene and its four impurities, DDE approach showed to create the models that have greater statistical power than the ones described with the main factors only, which makes them more reliable and accurate for the application. Additionally, because of the interactions selection criterion and final tuning phase of the strategy, DDE approach allows the creation of models that are likely to better describe the real systems, which increases the practical value. Copyright