ESTIMATION OF HEDONIC RESPONSES FROM DESCRIPTIVE SKIN SENSORY DATABY CHI-SQUARE MINIMIZATION I.F. ALMEIDA 1 , A.R. GAIO 2,3 and M.F. BAHIA 1 1 Department of Pharmaceutical Technology Faculty of Pharmacy, University of Porto Rua Aníbal Cunha 164, 4050-047 Porto, Portugal 2 Department of Pure Mathematics Faculty of Science, University of Porto Rua do Campo Alegre 687, 4169-007 Porto, Portugal Accepted for Publication October 25, 2005 ABSTRACT Six topical formulations were evaluated by a trained panel according to a descriptive analysis methodology and by a group of consumers who rated the products on a hedonic scale. We present a new approach that describes the categorical appreciation of appearance, texture and skinfeel of the formula- tions by the consumers as a function of related sensory attributes assessed by the trained panel. For each hedonic attribute, a latent random variable depending on the sensory attributes is constructed and made discrete (in a nonlinear fashion) according to the distribution of consumer-hedonic scores in such a way as to minimize a corresponding chi-square criterion. Standard partial least squares (PLS) regression, bootstrapping and cross-validation techniques describing the overall liking of the hedonic attributes as a function of associated sensory attributes were also applied. Results from both methods were compared, and it was concluded that chi-square minimization can work as a complementary method to the PLS regression. INTRODUCTION Skin-care products can be clinically effective and yet unacceptable to consumers. Besides presenting a biological effect, they are also expected to convey a sensory message to the consumer. It is often difficult to understand why some products are not perceived to be as acceptable as others in the 3 Corresponding author. TEL: (351) 220 100 715; FAX: (351) 220 100 708; EMAIL: argaio@fc.up.pt Journal of Sensory Studies 21 (2006) 2–19. All Rights Reserved. © 2006, The Author(s) Journal compilation © 2006, Blackwell Publishing 2