Abstract
It is proposed that when a dependent variate y is classified into N sets of values with different predictive statistical relations for each classification, the associated correlation coefficients rn (n = 1 − N) can be usefully combined into a single correlation coefficient by a weighted average of the rn2. The weighting factor is the variance of y for each n. In assessing the combined predictability of y by the set of N relations an additional weighting factor is Jn, the number of predictions to be made for each n.
