ASSESSING THE PREDICTIVE RESILIENCE OF COMPOSITE EXPERIMENTAL DESIGNS UNDER MISSING OBSERVATIONS: THE INFLUENCE OF Α ALPHA VALUE VARIATIONS

The field of robust missing observations has become a significant study area in statistical research. Regarding Response Surface Methodology, a well-established and practically applied domain in experimental design, missing observations in an experiment can make parameter estimation chaotic and challenging. In Response Surface Methodology, three types of second-order composite designs are primarily used. These are the Central Composite Design, Small Composite Design, and the Minimum Run-Resolution V. This study employs a range of alpha values to assess the efficiency of the design and the predictive capabilities in the context of missing data. These alpha values encompass rotatable, spherical, practical, and a novel value termed by practical efficiency alpha, intended to rival the practical alpha value. The design’s efficiency is gauged through the minimax loss criterion, while the predictive capability is assessed via a Fraction of Design Space plot. The research findings suggest that αPE performs effectively across various design types, even without specific observations.