Optimal nested cross-over designs in sensory analysis

摘要

Designing sensory trials when only a subset of products can be tasted by each subject has been widely studied. In particular, trials can be successfully planned for any combination of number of subjects, number of products and number of products tasted per session, by using algorithms for the construction of optimal designs. These algorithms allow the construction of incomplete crossover designs as balanced as possible for the order and the first-order carry-over effects. We propose here an adaptation of such algorithms when the number of subjects is not precisely known before the evaluation. Two strategies are presented, based on forward and backward sequential algorithms. Each produces a sequence of nested designs that may be used for various numbers of subjects. Simulations show that the nested designs obtained through the forward algorithm have very satisfying statistical properties, and always prove to be better than those from the backward one.