IDENTIFICATION OF DRIVERS OF OVERALL LIKING - DETERMINATION OF RELATIVE IMPORTANCES OF REGRESSOR VARIABLES

This paper introduces and utilizes three new methods, which appear in statistical and machine learning literature, to identify drivers of consumer overall liking. These methods include the Lindeman, Merenda and Gold (LMG) method, Breiman's Random Forests (RF) and Johnson's relative weight. The methods provide valid measures of relative importances of attributes to consumer overall liking in the situation of multicollinearity. The methods are flexible and can be used for both aggregated data (averages) of multiple products and raw data of one or more products. Breiman's RF and Johnson's methods can be used for both the data sets with more observations than variables and the data set with more variables than observations, which is the typical situation in sensory panel descriptive analysis data. The LMG and Breiman's RF methods are applicable to both continuous attributes and categorical (including binary) attributes. Numerical examples are provided to illustrate the applications of the methods using freely available R programs.