EVALUATION OF TESTS AND GROUPING OF CULTURES BY A TWO-STAGE PRINCIPAL COMPONENT METHOD

Abstract

A two-stage multivariate procedure for analysis and condensation of large sets of +/− data has been tested on data derived from application of 75 tests to 59 named cultures of Aerobacter, Aeromonas, Bacillus, Escherichia, and Pseudomonas. Seventeen clusters of associated attributes (attribute-complexes) were found by Adansonian R-analysis. As a result of successive separate principal component analyses, 47 principal component vectors were found which cumulatively accounted for 88% of aggregate variance and collectively defined an attribute-complex subspace. Four leading principal component vectors were extracted from attribute-complex space and accounted for 43% of the original variance; cultures were allotted 4-part ordination scores in respect of these vectors and these scores were treated as coordinates of item-points for purposes of cluster analysis. Cultures were separated into appropriate major categories but minor groupings were not entirely satisfactory; deficiences in the data and geometrical distortion during condensation were implicated. Major discriminants were defined in terms of important individual attributes, of combinations of attributes, and of attribute complexes.