On the selection of significant variables from unreplicated factorial designs

Abstract

Separation of the few control factors with significant effect from the remaining ones at the early phase of experimental studies is a recommended practice for systems development and improvement. The normal probability plot is a widely accepted graphical technique for screening, although it does not provide a unique interpretation of results in some cases. Thus, more formal procedures have been proposed to supplement it. The performance of popular methods along with lesser known and used ones is assessed using four evaluation standards based on a simulation study. The results show that Al-Shiha and Yang’s method stands out as the best performer when compared with other computational methods so it is the best choice to supplement the often used graphical technique. A strategy for analysing unreplicated fractional factorial designs is also outlined.