ABSTRACT
The statistical models we studied so far (linear regression, nonlinear regression, nonparametric smoothing and additive models) have a common feature: to properly use these modeling techniques, we need to know which predictors to use in a model. Knowing which predictor variables to use is often the most important part of a study. However, a variable selection result is always model-specific. For example, using a linear regression model, variables selected will likely be those having a linear relationship with the response variable. Even when the additive model is used, the additive assumption will affect the variable selection results. In environmental and ecological studies, the additive assumption is rarely realistic. As a result, both linear regression and the additive model are inefficient in selecting variables that may have strong interaction effects.
