Short CommunicationComparison of Correspondence Analysis based on Hellinger and chi-square distances to obtain sensory spaces from check-all-that-apply (CATA) questions

- Correspondence Analysis is used to analyze CATA data.
- Hellinger and chi-square distances were compared when analyzing CATA data.
- Sample and term configurations from both analytical approaches were similar.
- Stability of sample and term configurations did not largely differ in most studies.

Sample configurations from check-all-that-apply (CATA) questions are obtained using Correspondence Analysis (CA). Classical CA is based on chi-square distance, which has been reported to be strongly affected by infrequently selected terms. The Hellinger distance has been proposed as an alternative distance metric, and the aim of the present work was to compare product spaces from CATA questions obtained using CA based on chi-square and Hellinger distances. Data sets from 71 studies (5121 consumers), differing by product category, number of consumers, number of samples and number of terms included in the CATA question, as well as frequency of infrequently used terms, were analyzed. For each of the studies, frequency tables were input to CA based on chi-square and Hellinger distances. Sample and term configurations in the first two dimensions were compared using the RV coefficient. Furthermore, the stability of sample and term configurations for each type of distance was evaluated by simulating repeated experiments using a bootstraping resampling approach. Sample and term configurations obtained using Hellinger and chi-square distances were similar (average RV coefficients for sample configurations = 0.99; average RV coefficients between term configurations = 0.89). The stability of sample and term configurations were not largely affected by the type of distance used to analyze frequency tables. Results from the present work suggest that CA based on chi-square and Hellinger distances provide similar results. Contributing to guidelines for practitioners, this research therefore supports classical CA analysis as an acceptable approach to the analysis of sensory-specific CATA data.