The author contends that model misspecification can occur even after researchers have selected the generally most appropriate class of methods, or general linear model techniques. It is suggested specifically that canonical correlation analysis may provide more meaningful results, as compared with regression, particularly if analysis is augmented by the computation of structure coefficients. It is also suggested that contemporary analytic practice reflects some improvements over more traditional practice. Researchers are increasingly investigating multivariate problems with multivariate methods. Greater use of the multivariate general linear model, or canonical correlation analysis, augmented by the calculation of appropriate coefficients, including structure coefficients, is proposed for future research. (DWH)