\
 

Start Date

16-8-2018 12:00 AM

Description

When two or more predictor variables are highly correlated and contain measurement error (M+ME), regression and Partial Least Squares (PLS) beta coefficients and t statistics can be inaccurate (Goodhue, Lewis and Thompson, 2017). Corrections to the misleading values can be made by using the application created by Goodhue et al., or by correcting correlations for attenuation before running a regression. We examined research articles in three Management Information Systems journals over multiple years and discovered that, of the regression and PLS papers that reported correlations, about one-half (48%) were operating in the danger zone. In the 10 papers that provided sufficient information to use the Goodhue et al. (2017) application we found that, on average, the t statistics were biased by about 0.70. This suggests that when using regression or PLS, researchers should check for the M+ME bias and correct for it when found.

Share

COinS
 
Aug 16th, 12:00 AM

What Do You Mean, My Results are Incorrect? The Impact of Multicollinearity and Measurement Error in Tests of Statistical Significance

When two or more predictor variables are highly correlated and contain measurement error (M+ME), regression and Partial Least Squares (PLS) beta coefficients and t statistics can be inaccurate (Goodhue, Lewis and Thompson, 2017). Corrections to the misleading values can be made by using the application created by Goodhue et al., or by correcting correlations for attenuation before running a regression. We examined research articles in three Management Information Systems journals over multiple years and discovered that, of the regression and PLS papers that reported correlations, about one-half (48%) were operating in the danger zone. In the 10 papers that provided sufficient information to use the Goodhue et al. (2017) application we found that, on average, the t statistics were biased by about 0.70. This suggests that when using regression or PLS, researchers should check for the M+ME bias and correct for it when found.