| DATE: | Wednesday, September 25, 2002 |
| TIME: | 4:40-5:40 pm |
| PLACE: | LN 2205 |
| SPEAKER: | U. U. Mueller (University of Bremen, Germany) |
| TITLE: | Efficient estimators for conditionally constrained models |
We consider a general class of regression models, described by a constraint on the joint law of covariate and response: Some function of covariate, response and an unknown parameter has conditional mean zero given the covariate. This includes nonlinear regression models, in which the conditional mean of the response is a certain function of the covariate and a parameter, and quasi-likelihood models, in which both the conditional mean and the conditional variance are functions of the covariate and a common parameter. We construct efficient estimators for finite-dimensional functionals of the distribution, in particular for the parameter and for expectations of the joint law of covariate and response. Estimators for the parameter are obtained as solutions of weighted estimating equations. The estimators of the joint law are improvements of empirical estimators.
This talk is based on joint work with Wolfgang Wefelmeyer.