Semiparametric models for correlated nominal data

Abstract

In this paper we consider semiparametric modeling for correlated nominal data. The model consists of two components, parametric and nonparametric. We propose generalized estimating equation (GEE)-Smoothing spline as a method to estimate these components. GEESmoothing spline can be seen as an extension of parametric GEE to semiparametric GEE. The parametric component is estimated based on GEE, while the nonparametric is estimated based on smoothing spline method. In this paper we consider the logit link function (nominal logistic regression model).