Multiple regression is often used to compare the importance of two or more predictors. When the predictors being compared are measured with error, the estimated coefficients can be biased and Type I error rates can be inflated. This study explores the impact of measurement error on comparing predictors when one is measured with error, followed by a simulation study to help quantify the bias and Type I error rates for common research situations. Two methods used to adjust for measurement error are demonstrated using a real data example. This study adds to the literature documenting the impact of measurement error on regression modeling, identifying issues particular to the use of multiple regression for comparing predictors, and offers recommendations for researchers conducting such studies.
Keywords: measurement error; multiple regression; predictor importance; predictor reliability.
© The Author(s) 2019.