非獨立群規模資料之有效推估法

Abstract

本論文主要針對參數模型在非獨立群規模之資料下提出參數有效估計法，在所考量之資料結構下，群內重覆抽樣 (WCR) 及群加權廣義估計式 (CWGEE) 方法無充分利用群內相關性訊息。從過去的研究可發現， WCR 及 CWGEE 方法所得的估計式有相同的大樣本性質，然 CWGEE 在計算上比 WCR 更為簡便快速，當最小群規模大於一及群內相關性訊息可用下，吾等所提之估計式較 WCR 及 CWGEE 估計式有效。 在此研究，吾將利用模擬探討估計式之有限樣本性質， 並進一步與 CWGEE 估計式比較。

In this study, we propose two estimation methods for considered marginal models under the cluster data setting with informative cluster size. The information of within-cluster correlation is appropriately used through the minimum cluster size in our approach, which is not fully considered in the within-cluster resampling (WCR) and cluster-weighted generalized estimating equation (CWGEE) methods. It is known in the former works that the approaches of WCR and CWGEE are asymptotically equivalent but the WCR estimation procedure is computationally intensive. When the within-cluster correlation is available and the minimum cluster size is greater than one, our estimatiors improve the inefficiency of the both estimators. The finite sample properties of the proposed estimators are examined through a Monte Carlo simulation. Meanwhile, a comparison with the CWGEE method is made in the numerical study.