Preference of cauliflower related to sensory descriptive variables by partial least squares (PLS) regression

Two new multivariate techniques for comparing two blocks of variables were tested on sensory data from cauliflower. The techniques are based on Partial Least Squares (PLS) regression on latent variables. The method of PLS2 was used for relating four different preference variables simultaneously to 11 descriptive variables by regression on common factors. The method of PLS1 was subsequently used to relate individual preference variables to the 11 descriptive variables. Two factors were in each case found to be statistically significant by cross‐validation. The first, dominating PLS2 factor was a general quality factor showing that all the descriptive variables were more or less related to all the preference variables. This quality factor agreed well with the producers quality classification of the batches upon harvesting. A second, minor factor distinguished between texture and flavour descriptor's relationships to the preferences. PLS1 regressions showed that texture preference was best predicted by texture descriptois, especially crispness, and flavour preference was best predicted by flavour descriptors, especially fruitiness. Some benefits of the PLS methods over the conventional alternatives of stepwise regression and principle component analysis are discussed.