Chemometrics: The Use of Multivariate Methods for the Determination and Characterization of Off-Flavors

ABSTRACT Flavor is a complex phenomenon, for which statistical analyses of one variable at a time do not often suffice especially in problems of sample classification. Multivariate statistics is a set of tools that can help deal with the complexities and subtleties that must be confronted as researchers try to analyze, and characterize and classify flavors. Treating two or more variables simultaneously requires the use of the mathematical apparatus of matrix algebra, including the use of vectors, eigenvectors, and eigenvalues. When categories are already known and samples are available from each category, techniques such as SIMCA and nearest neighbor analysis can serve to classify unknown samples. When researchers do not know ahead of time what categories exist, techniques such as clique analysis and multidimensional scaling can suggest categories. When the underlying variables are not known to the researcher, then the technique of factor analysis helps provide clues to what these “phantom” variables might be. Reducing the number of variables — clearing out the underbrush, so to speak — is often accomplished by techniques such as (1) principal components analysis, which condenses the variability into just a limited number of variables and (2) discriminant analysis, which focuses on those few combinations of the variables that distinguish among categories. Typically, in multivariate analysis, the individual initial variables are combined to form new variables that are more tractable and more meaningful. The coefficients in these combinations correspond to the relative importance of each of the original variables. Thus, the values for the coefficients provide a filter for removing insignificant variables. Computers do the tedious computations but the researcher must do the deciding, the thinking, and the interpreting. Selected articles on food research given in the last part of this chapter furnish concrete examples of the uses and often, misuses of multivariate statistics. This chapter is intended to help bridge the gap between what the researcher knows and what the researcher needs to know in order to understand and appreciate other people's published results and to apply intelligently and appropriately multivariate statistics to his/her own research.