Comparisons of Bayes factors for 𝟐𝟒 full, fractional, and reduced factorial designs

Abstract

The effect of factors in full and fractional factorial designs is being studied ubiquitously in all fields of science and engineering. At times, researchers would want to gather additional information than the fractional factorial design provided, there is no restriction to conducting more experimental runs. In this study, we propose a reduced fractional factorial design consisting of all significant factors. This paper illustrates the effectiveness of factors through real data application and simulation by comparing the full factorial, reduced factorial, and fractional factorial designs. The actual weightage of the main/interaction effects in these three designs was found by identifying and quantifying the Bayes factors through the simulation datasets. It is observed that the reduced factorial design produces better results when there are no constraints to select or add factors to the model.