If rank transformations result in the best specified model, multiple regressions that use
rank transformations of each variable should result in lower mean square error and higher
adjusted R2 as compared to regressions that used untransformed, logarithmic, and squareroot
transformations of the variables in the model