Partial Generalized Additive Models--An information theoretical approach to avoid the concurvity
by Hong Gu
Abstract: We develop a new procedure called partial generalized additive models (pGAM). Instead of back-fitting, pGAM sequentially maximizes the mutual information (MI) between the response variable and covariates and selects the optimal covariate to enter the model. Introduced by Shannon, MI provides a good measure of nonlinear dependence between variables and can be viewed as a generalized, nonlinear version of Pearson's correlation coefficient. At each step, pGAM also removes any functional dependencies between remaining covariates and the ones already in the model, thereby avoiding any potential problems of concurvity and allowing the interpretation of the resulting model to be much more precise. With a number of examples, we show that pGAM is a reasonable and meaningful variable selection procedure and gives much better estimates of the covariates' functional effects.
For More Information: Contact: Guoqi Qian firstname.lastname@example.org