Comparison of check-all-that-apply product evaluations in different conditions using an L1-norm framework
In sensory evaluation, consumers often evaluate products by answering check-all-that-apply (CATA) questions. These questions instruct the consumer to check all terms in a list that describe the product, where checking a term is called a citation. When all products are evaluated by all consumers under two conditions, the CATA data can be organized into a four-dimensional array (assessors, products, terms, conditions) with responses coded 1 for checked and 0 for not checked. The proposed data analyses are based on a unified L1-norm testing framework. Following the “one citation, one vote” principle, we use permutation tests to investigate whether the two conditions differ (1) overall, (2) in product citation percentages for term t, (3) in product p citation percentages for term t, (4) in the difference in citation rates across terms for each pair of products, (5) in the difference in citation rates for term t for each pair of products, and (6) in product citation percentages for product p across terms. Next, we use L1-norm principal component analysis to explore multivariate differences in the two conditions for the products and terms. The methods are illustrated using CATA data from a sensory study in which 100 consumers evaluated 11 beverage products under different conditions with common terms.
Castura, J.C., Greenacre, M.J., & Chaya, C. (2025). Comparison of check-all-that-apply product evaluations in different conditions using an L1-norm framework. Royal Statistical Society International Conference 2025. 1-4 September. Edinburgh, UK.