Optimal discrimination between mixed models through nature-inspired algorithms

C. Tommasi, V. M. Casero Alonso, J. López Fidalgo, S. Pozuelo Campos, W. K. Wong

Random effects models are widely used across all disciplines, particularly in the life sciences and clinical studies. It is well known that if the used model is misspecified, the statistical inference can be misleading or become invalid.
This work assumes there are several plausible random effects models and uses the Kullback-Leibler divergence criterion to find a design that optimally discriminates among the competing models.
This optimization problem is complex, because it is a multi-level nested optimization problem over very distinct types of domains and furthermore the design criterion is non-differentiable. A theoretical result that simplifies the computational burden has been developed and a nature-inspired metaheuristic algorithm to search for an optimal discrimination design has been implemented.
Two applications are given: the first concerns fractional polynomials with one continuous variable, and the second relates to multi-factor random effects models.

Keywords: Design efficiency; KL-optimality; Particle Swarm Optimization

Design of Experiments I
June 11, 2025  3:30 PM
MR 1

C. de la Calle-Arroyo, J. López Fidalgo, P. Urruchi Mohino

C. de la Calle-Arroyo, L. Deldossi, C. Tommasi

Á. Cía Mina, L. Deldossi, J. López Fidalgo, C. Tommasi