Comparison of a new correspondence analysis for check-all-that-apply data with existing exploratory data analysis procedures
Check-all-that-apply(CATA) questions are widely used in sensory and consumer testing. A CATAquestion presents a list of predetermined responses along with an instructionto check all responses in the list that apply. In product testing, consumers arepresented with the same CATA question for each product sample, where the termsoften relate to perceptions (e.g. appearance, aroma, flavour, texture), but mayalso relate to emotions or conceptualizations. Often, assessors will evaluatemultiple products or concepts using such a CATA question. Researchers applyvarious multivariate data analyses to explore CATA data. The most popularapproaches include simple correspondence analysis, correspondence analysisusing the Hellinger distance, multiple-response correspondence analysis,principal component analysis, and L1-norm principal component analysis. As willbe shown, these methods do not fully account for data dependencies in CATAdata. A newly proposed correspondence analysis that accounts for these data dependenciesis recommended on theoretical grounds, but how different are its results inpractice? The presentation will give highlights from analyses of three CATAdata sets: (i) data from 114 Uruguayan consumers who evaluated fruit from 6strawberry cultivars using 16 sensory terms, (ii) data from 161 Canadianconsumers who evaluated 6 whole-grain breads using 31 sensory and emotionterms, and (iii) data from 100 UK consumers who evaluated 11 blackcurrant beveragesusing 34 emotion terms. Bootstrap-derived confidence ellipsoids were constructedto investigate pairwise differences between products and show that the proposedcorrespondence analysis is well aligned with results from conventionalinference tests for CATA data. Although the analyses often reveal similarassociations between products and terms, there are obvious differences. Resultsfrom the proposed correspondence analysis are similar to results from multiple-responsecorrespondence analysis but emphasize different terms. Reasons are given and recommendationsare discussed.
Castura, J.C. (2026). Comparison of a new correspondence analysis for check-all-that-apply data with existing exploratory data analysis procedures. Associazione per la Statistica Applicata (ASA) Conference 2026. 16-18 September. Teramo, Italy. (Oral Presentation).