The purpose of this paper is to reiterate the importance of testing for the predictive validity of regression and discriminant solutions in studies devoted to the prediction of athletic success. Two approaches to test for the predictive validity of regression and discriminant solutions, longitudinal and cross-validation, are reviewed. Statistical issues relating to predictive validity, including small case to predictor ratios, multicollinearity, and the limitations of stepwise regression are discussed. The paper also examines alternative methodologies that would enable researchers to obtain better estimates of predictive validity. Several studies in the prediction of athletic success that have reported high squared multiple correlation coefficients (R2) are reviewed. The purpose of this review is to demonstrate how various statistical limitations bring into question the conclusions drawn from these studies.
Ralph Renger is with the Sports Information Group, Inc., 1927 35th St., S.W., Suite 1, Calgary, AB Canada T3E 2X4.