Sensory analysis is a very important set of procedures in food industry. It helps us in finding differences between samples and to describe those samples (differences). In the last fifteen years or so, the quick profiling techniques gained popularity. Those methods allow a descriptive analysis without the usage of the same vocabulary between assessors. My actual work is based on the Free Choice Profiling technique, which allows assessors to freely choose the number of descriptors and their meaning to describe a product. GPA (generalized Procrustes analysis) and MFA (multiple factor analysis) are the two most popular statistical techniques used to analyze the obtained data. They are both based on dimensionality reduction of the data (matrices) and unlike the PCA (principal component analysis), they retain individual assessor information. These methods process the data by creating a consensual (average) space of products (samples), which reflects the average perception of all assessors.
My work was focused in comparing the similarity of GPA and MFA outputs by measuring two statistics (RV2 coefficient and ARI-Adjusted Rand index) under different conditions; score variability, number of assessors, number of selected descriptors, GPA/MFA dimensionality output, nature (focus) of assessment. The simulations showed that RV2 is slightly biased under the assumption of independence between the simulated main data matrix and the simulated assessors' ones.
The simulation confirmed that the GPA and MFA are indeed very comparable under all conditions and that the measured statistics are not highly correlated in a direct GPA / MFA comparison. Additionally, I modelled the influence of the selected factors on the averaged (MFA+GPA) measured statistics. The obtained information is very valuable because it provides an insight about the influence of the selected factors. This allows us to properly combine them to optimize the execution of future assessments with the assessed Free Choice profiling technique.