SUMMARYTwo methods of discriminant analysis, Chi‐square tests, “t” tests and analysis of variance were applied to gas‐liquid chromatographic (GLC) data to compare GLC measurements of food volatiles with organoleptic evaluation of flavor. Model systems composed of ethyl butyrate, ethyl heptylate and benzaldehyde were used for one set of trials. A second system consisted of blends of Pepsi‐Cola® and Coca‐Cola®. The stepwise discriminant analysis (SDA) procedure described by Powers et al. (1968) was only moderately successful in classifying the mixtures when applied to the GLC data for the model systems, but the cola blends could be classified readily. From GLC data, cola blends differing from each other by only 7–10 could be distinguished. A new discriminant analysis program, designated MUDAID by its originator (Bargmann, 1967), is more useful for predictive purposes. The analysis yields the correlation coefficients for each peak area with sample differences and the weighting factors. Discriminant equations can thus be written. By substitution of measurement values for unknowns, their identity may be predicted. Organoleptic subthreshold additive effects could not be demonstrated with binary mixtures of flavor substances, but judge‐compound interaction effe
展开▼
