| Summary: | 碩士 === 國立清華大學 === 統計學研究所 === 81 === In parametric estimation, the maximum likelihood principle has
played an important role for the recent years. Here, we
introduce a different approach by transforming the model
considered into a linear model, based on which,one then
constructs the generalized least squares GLS estimator. It is
shown that in one-sample parametric model, the GLS estimator is
asymptotically equivalent to the {\em MLE}. We also show, for
the Weibull case,the {\em GLS} estimator is also asymptotically
equivalent to the {\em MLE}. After that, we discuss the
robustness property of the GLS estimator. And then, we make a
suggestion for how to choose the regression point for the
transformed linear model.
|
|---|
