Validity and Nonlinear Heteroscedastic Models

Summary T he validity of tests commonly is expressed in terms of the Pearsonian coefficient of correlation. The use of this coefficient implies that linear homoscedastic relationships hold between test scores and criteria. Therefore, the position is taken that the accuracy of prediction of criteria from test scores is the same throughout the entire range of scores and that both success and failure on a job are the result of the same factors. Examination of the relationships between test scores and criteria for three groups—executives, office workers, and autobus repairmen—indicated that descriptions of the predictive power of the tests and of the traits important in job success and failure as given by the Pearsonian validity coefficient were not entirely adequate. Use of a different index, theta, was shown to give much more meaningful and useful descriptions.