Two-Stage Estimator for Generalized Linear Models with Nonconstant DispersionParameter

摘要

A two-stage estimator is proposed to estimate beta in generalized linear modelswhere the dispersion parameter is a linear function of covariates. Asymptotic normality is proved not only for an exact likelihood of the dependent variables, instead expressions for expectation and variance are presumed together with some regularity conditions. In the first stage square root of n-consistant estimators are constructed. The second stage consists of a single iteration step which suffices to obtain the desired asymptotic covariance matrix for (beta circumflex).