A Framework for Estimation and Inference in Generalized Additive Models with Shape and Order Restrictions

Methodology for the partial linear generalized additive model is presented, where components for continuous predictors may bemodeled with shape-constrained regression splines, and components for ordinal predictors may have partial orderings. The estimated mean function is obtained through a projection (or iteratively reweighted projections) onto a polyhedral convex cone; this is key for formally derived inference procedures. Pointwise confidence bands and hypothesis tests for the individual components, as well as a model selection method, are proposed. These methods are available in the R package cgam.