Assessment of adaptive one-factor-at-a-time method vs. fractional factorial methods using reconfigurable paper aircraft

Abstract

Recent research has suggested that under certain conditions, adaptive one-factor-at-a-time (aOFAT) methods outperform more commonly used fractional factorial methods. This study sought to corroborate these claims by analyzing a case study of a real-life experiment. A full factorial experiment was conducted to collect data for simulations of fractional factorial and adaptive one-factor-at-a-time experiments. The experiment used a reconfigurable paper aircraft template with four three-level control factors. Results indicated that the exploitation of control factor interactions by adaptive one-factor-at-a-time occurred at similar rates as predicted by Frey and Wang (2006). AOFAT experiments proved particularly effective at avoiding factor levels that led to poor performance. with rates of avoidance approaching 100% for the worst levels. When bias in the full factorial experiment was eliminated, aOFAT methods even returned a higher (weighted average) leading quality indicator value than full factorial methods.