Generalized nonlinear models and variance function estimation

An iterative estimation method by combining quasi-likelihood estimation related to a nonlinear variance function with the least-squares estimation based on such variance function is proposed. It is shown that the iterative estimation has consistency and asymptotic normality under some mild assumptions. Simulation studies show that for the large samples our method has certain advantages and is different from previous methods. Two examples are also given to illustrate their modelling method and proposed algorithm.