Finding Transformations for Regression Using the ACE Algorithm

The ACE algorithm of Breiman and Friedman (1985) estimates the transformations giving rise to the maximal multiple correlation of a response and a set of predictor variables. A study of these transformations can give the data analyst insight into the relationships between these variables. Using the methodology of Box and Cox (1964), we show how to find familiar closed form approximations for the optimal ACE transformations. This can lead to substantially improved empirical models for social phenomena.