Selecting an economical binary test battery for a set of microbial cultures

Abstract

Selection of diagnostic tests for key construction from binary test-scores was achieved by evaluation of the partitioning effects of sequentially executed dichotomies and by an exploratory process of data-matrix reduction. Corresponding computer programs are described and their results compared. Both procedures yielded similar, acceptable, keys when applied to data of various origins ranging in size from 9 cultures X 8 tests to 308 X 131. The matrix-reduction method was usually more economical of computer time. Some indications were derived of the approximate number of such diagnostic tests likely to be needed in keys for given numbers of individuals. The actual number required in any instance will depend on the extent to which individual tests are correlated in the data. Consequently, overall performance of tests when associated in keys is not predictable from simple numerical indices expressing their divisive power.