A comparative study between latent class binomial segmentation and mixed-effects logistic regression to explore between-respondent variability in visual preference for horticultural products

A methodological concept is proposed to study between-respondent variability in visual preference for horticultural products using quantitative imaging techniques. Chicory, a typical Belgian vegetable, serves as a model product. Eight image sequences of high-quality chicory, each representing a different combination of two factor levels of length, width and ovality, were constructed to satisfy a 23 factorial design by using quantitative imaging techniques. The image sequences were pair-wise visualized using a computer-based image system to study visual preference. Twenty respondents chose which of two samples was preferred in all 28 pair-wise combinations of the eight constructed image sequences. The consistency of the respondents and the agreement between respondents was evaluated. The poor fit of a traditional binomial logit model that relates preference with quality descriptors was due to the low agreement in preference between respondents. Therefore, latent class binomial segmentation is compared to mixed-effects logistic regression. Both approaches relax the traditional assumption that the same model holds for all respondents by recognizing the typical between-respondent variability inherent in preference studies. Where the latent class model simultaneously estimates different logit models for different consumer segments, the mixed-effects model recognizes between-respondent variability by incorporating random effects varying by respondent in model formulation.