Discriminant Functions with Covariance

Abstract

A brief review of the literature includes uses of discriminant functions. A theoretical example is then introduced with a plan of procedure. With an initial measurement assumed to average the same in 2 populations, and later "discriminator" measurements, an analysis of covariance is carried out. Adjusted values or deviations from this analysis* are used to calculate the final discriminant. A numerical example is presented using data published on effect of insulin on blood sugar in rabbits. The initial blood sugar level is used in connection with levels at 2 time periods after dosage. The object is to get the best linear combination of effects at 2 periods, which would be helpful in utilizing both or in deciding to drop one. Coeffs. are selected to maximize the ratio of the discriminant to its standard.error. Methods for calculation, using a dummy variate, and tests of _ significance, are illustrated. In later sections, theory is treated more fully and extended to more complex cases, and relation to other work on the same subject is shown.