Empirical Comparison of Three Multiple Correlation Techniques

THE selection of the most valid composite of subtests from a battery may be accomplished by any of a number of multiple correlation techniques. It is the purpose of this study to make an empirical comparison of the efficiency of three techniques in predicting success in specific areas. I . The multiple correlation and regression weights for the whole battery were computed by the Cowles-Crout technique ( I ) , which is just a convenient way of solving a set of simultaneous linear equations. This technique is similar to one of the methods of single division presented by Dwyer (2, p. 106). A single large work sheet will accommodate a matrix of twenty variables. A summation column is carried as a check on each row of the forward solution. A column is used for the regression weights and another for the check on accuracy of these weights, The best composite of subtests was then selected by choosing the variables with the largest positive regression weights. The decision to disregard variables with negative weights was an arbitrary one based on previous experience with Air Force data. The omission of a good suppressor variable, if one occurred, would result in a significant loss in validity. A suppressor test is one that correlates negligibly with an external criterion, but does correlate substantially with another predictor which, in turn, shows a t least a moderate relationship with the criterion. The role of the suppressor is to suppress or partial out from the predictor (for which it is a suppressor) some of its variance that is invalid for the criterion a t hand, thus making the predictor a purer or more pertinent measure of the criterion. * The assistance of hIrs. Iris H. hlassey in supervising theCowles-Crout and WherryDoolittle computations is gratefully acknowledged. The views expressed in this article are those of the author and do not necessarily represent the official views of the United States Air Force.