Nonparametric and semiparametric dynamic additive regression models

Dynamic additive regression models provide a flexible class of models for analysis of longitudinal data. The approach suggested in this work is suited for measurements obtained at random time points and aims at estimating time-varying effects. Both fully nonparametric and semiparametric models can be fitted and supplied with (asymptotic) standard errors. The focus is on the cumulative regression functions since they are the ones that can be estimated at the ususal root-$n$ rate. We improve previously suggested estimators with respect to efficiency and show that these new estimators are efficient in special cases. We investigate the finite sample properties of the estimators and conclude that the asymptotic results are valid for even samll samples.