Moving out of the Linear Rut: The Possibilities of Generalized Additive Models

Generalized Additive Models are statistical models in which the usual linear relationships between the response and predictor variables are replaced by non-linear 'smooths'. This paper discusses the context in which this technique has been developed, and provides two illustrations. The first illustration is an analysis of precipitation in California which requires a normal-theory model. The second is the prediction of land-use from satellite imagery; the discrete nature of the response necessitates the fitting of a GAM with a logit link and a binomial random term.