Semi-parametric Inference for a Class of Varying-Coefficient Models

Abstract

A class of scaled link models extends the scope of generalized linear models for heterogeneous data. If the scale is a simple parametric function, then the estimation may be achieved by an alternating regression algorithm. Generalizing further leads to a useful class of models that allow semi- parametric scaling of some variables and additive effects of others. For the situation in which the functional form of the scaling is unknown, but "smooth", efficient semi-parametric estimates and confidence intervals are developed for the parametric components of the model, and consistent estimates are obtained for the nonparametric components. The approach is illustrated on data from collaborative research.