Quasi generalized pseudo maximum likelihood method based on the linear exponential family

Abstract

PML1 allows estimation of the correctly specified mean structure either when the assumed distribution includes no nuisance parameter or when an addi-tional nuisance parameter, such as the covariance matrix from the normal distribution, is fixed. For applications, this assumption might be unrealistic because the a priori guess for the nuisance parameter is rarely good. Fur- thermore, the more similar the chosen nuisance parameter is to the optimal nuisance parameter, the more efficient the estimator will be.