ABSTRACT
This chapter also considers adding additional predictors that are functions of the existing predictors like quadratic or cross-product terms. Transformations of the response and/or predictors can improve the fit and correct violations of model assumptions such as non-constant error variance. The best approach is to try different transforms to get the structural form of the model right and worry about the error component later. Regression coefficients will need to be interpreted with respect to the transformed scale. There is no straightforward way of back-transforming them to values that can be interpreted in the original scale. The Box–Cox method is a popular way to determine a transformation on the response. It is designed for strictly positive responses and chooses the transformation to find the best fit to the data. Broken stick regression is sometimes called segmented regression. Allowing the knotpoints to be parameters is worth considering, but this will result in a nonlinear model.
