Importance of adjusting for correlated concomitant variables in psychiatric research

The potential importance of adjusting for correlated concomitant variables in psychiatric research to avoid possible erroneous or misleading results is demonstrated through examples of "suppression effects" and "spurious associations." Special reference is made to correlational studies and unbalanced factorial analysis of variance. Methods of adjusting for correlated covariates are noted, and the usefulness of multiple regression for this purpose in certain situations is shown. Simulated and real data examples are provided. Part I discusses these issues in relation to correlational studies, and also lays a general groundwork for understanding them in terms of simple path analysis models. Part 2 relates the problems of correlated covariates to factorial analysis of variance (ANOVA). As in Part 1, "suppression effects" and "spurious associations" are discussed, this time as they result from confounded main effects and interaction effects in ANOVA. Confounding in this case is synonymous with cell sample size imbalances in the factorial cross-classification. A useful method for dealing with these problems without loss of information is discussed. This method involves viewing the ANOVA from a "general linear model" perspective and performing the analysis as a multiple regression.