B0717
Title: A sparse estimation method for sensory evaluation data with taking individual scaling differences into account
Authors: Hironori Satomura - Osaka University (Japan) [presenting]
Abstract: In the sensory evaluation field, response styles, especially individual scaling differences, are frequently of concern. The traditional ways of analyzing sensory data, such as two-way ANOVA model comprising test stimuli as fixed effect and assessor related terms as random effects, are not capable of handling this scaling heterogeneity when investigating the differences among the test stimuli, which is of sensory scientists' interest. Assessor model, in which a multiplicative term introduced as an extra term in the two-way ANOVA model is, therefore, often utilized in the field. However, there still exists a limitation that the difference between each stimulus is examined via posthoc multiple comparisons, which does not necessarily produce a non-overlapping grouping of the stimuli. In order to tackle this problem, we propose a penalized likelihood method for this assessor model that encourages exact clustering of stimulus by taking the scaling difference into account. The model parameters are estimated through the EM algorithm with alternating direction method of multipliers. The usefulness of the proposed method is demonstrated through numerical examples.
