頁籤選單縮合
題名 | 存在群內相關時類別機率估計之研究=A Study for Probability of Classification with Intra-Cluster Correlation |
---|---|
作者 | 蘇聖珠; 陳麗霞; Su, Sheng-chu; Chen, Lschen; |
期刊 | 統計與資訊評論 |
出版日期 | 19990900 |
卷期 | 5 1999.09[民88.09] |
頁次 | 頁1-17 |
分類號 | 319.23 |
語文 | chi |
關鍵詞 | Dirichlet-Multinomial模式; 群內相關; 最大概似估計法; Dirichlet-Multinomial model; Intra-cluster correlation; Maximum likelihood estimation; |
中文摘要 | 當抽樣單位內的相關對象,因處於相同或類似的環境,而對某些議題有同等反應的傾向時,所搜集到的群集類別資料,可採Dirichlet-Multinomial模式(以下簡稱D-M模式),以適當地將群內相關性納入考量。 本文以D-M模式來架構,對於具有群內相關(intra-cluster correlated)的類別資料進行模擬。為求取參數之最大概似估計值,我們採用最陡上升法(steepest ascent method)求解。由於文獻對D-M模式中類別機率之個計多直接採用各類別之出現頻率,故本文乃在不同的群內相關係數、類別數、類別分佈、群數,及群內樣本數之下,比較最大概似估計量及出現頻率之表現。 觀察模擬結果發現,當群集數固定;若群內樣本數越大或群內樣本數固定;群集數越小或群內相關越大時,則採用最大概似估計量來估計群集類別機率較以出現頻率估計之誤差為小。 |
英文摘要 | In many studies, data are collected from differ clusters, and members within the same cluster behave similary. Thus, the responses of members within the same cluster are not independent and multinomial distribution is not the correct distribution for the observed counts. Dirichlet-Multinomial model (D-M model) is widely used when categorical data with intra-cluster correlation. This paper presents a simulation method of analyzing categorical data with intra-correlation by D-M model. Usually, to estimate the classified probability is use appear frequency We will use the steepest ascent method to find maximum likelihood estimation (MLE) method. This article used simulation method to compare the MLE and appear frequency under different intra-cluster correlation, kinds of classified distribution. To observe the simulate result, when clusters is fixed, the more of cluster size, or cluster size is fixed, the less clusters or the more intra-cluster correlation, the root mean square error is smaller by MLE then by appear frequency. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。