Association for Public Policy Analysis & Management

While multicollinearity weakens statistical power, the presence of correlation among predictors (multicollinearity) violates no assumptions of the standard linear regression model. This fact has led many scholars to conclude that multicollinearity poses no problems to valid statistical inference when significant results are obtained. While this conclusion is correct when all regression assumptions are perfectly met, multicollinearity can exacerbate problems associated with model misspecification or measurement error. In this paper, I use Monte Carlo simulations to demonstrate the effect of multicollinearity on type I errors (false positives) when nonlinearities are incorrectly modeled and when classical measurement error is present in some of the predictors. I conclude by offering a set of practical suggestions to applied researchers who encounter multicollinearity in their data.