Graphical Views of Suppression and Multicollinearity in Multiple Linear Regression

This article briefly reviews classical suppressor variables, suppression and enhancement, opposing signs of regression coefficients and zero-order correlations, and multicollinearity. A concise and easily understood graphical structure for the study of suppressor variables and enhancementis provided. Classical suppressor variables are shown to be more valuable than other nonsuppressor variables under some conditions. Errors using ratios of correlations in the exposition of suppression are noted. Multicollinearity is shown not to affect standard errors of regression coefficientsin ways previously taught.