The analysis of a set of questionnaires composed of 151 items on salt-, sweet- and fat-liking using LISREL-type and PLS-PM techniques leads to equivalent results

Poster + présentation oraleInternational audienceMeasure of salt-, sweet- and fat-liking is commonly assessed using sensory testing. However, this might become cumbersome with large number of subjects and products to assess. Thus, a questionnaire has been developed and administered through internet (Nutrinet-Sante Study) to a large sample (n=43 807) from the French population. The present study aimed to estimate the structural models underlying each sensory sensation (sweet, salty, fatty and sweet, fatty and salty) and to compute an overall score for each subject over each sensation using structural equation modeling (SEM). Subjects completed 151 questions based on (1) liking towards foods containing high amounts of sugar, or ‘sugar and fat’, or ‘salt and fat’ (9-point scale); (2) favorite level of seasoning with sugar, salt or fat on various foods (6-point scale); (3) favorite dishes (maximum 4) in a restaurant menu; (4) salt-, sweet- and fat-related behavior. The latent variables underlying each sensory sensation were determined using an exploratory factor analysis (principal component solution, varimax rotation; SAS® procedure FACTOR). Second-order models based on these latent variables were subjected to SEM using LISREL-type techniques (SAS® procedure CALIS) and Partial Least Squares-Path Model (PLS-PM, XLStat). Incidentally, PLS-PM (XLStat) consumed much more time than PROC CALIS (SAS®) for second-order models. Nonetheless, both regression coefficient and individual score vectors were highly correlated between the two types of SEM with Pearson’s coefficients greater than 0.75 and 0.95, respectively. In the present study, both types of SEM gave the same interpretation in terms of individual scoring