ABSTRACT
Transformations of the response and/or predictors can improve the fit and correct violations of model assumptions such as non-constant error variance. We may also consider adding additional predictors that are functions of the existing predictors like quadratic or cross-product terms.
