頁籤選單縮合
| 題 名 | 迴歸分析自變數離群值的一種評估方法=Evaluating the Covariate Outliers in Regression Analysis--A New Approach |
|---|---|
| 作 者 | 黃素貞; | 書刊名 | 中國統計學報 |
| 卷 期 | 35:1 1997.03[民86.03] |
| 頁 次 | 頁27-39 |
| 分類號 | 319.51 |
| 關鍵詞 | 離群值; 槓桿; 頻譜分解法; 影響值; Outliers; Leverage; Spectral decomposition; Influential observations; |
| 語 文 | 中文(Chinese) |
| 中文摘要 | 對迴歸分析自變數離群值的診斷,文獻多以槓桿或槓桿的函數值來評估。由於槓 桿是一種綜合性的指標,對自變數某些方向上的離群現象,它不見得有很好的偵測作用。本 文提出頻譜分解法來克服槓桿此方面的缺點。經由例示,我們發現頻譜分解法可以找出一些 具有影響值身份的離群值,而這些離群值卻不被槓桿法所發現。是故,我們所提的方法在自 變數離群值的偵測上較槓桿法敏銳。 |
| 英文摘要 | People use leverage or functions of leverage to diagnose the outliers of indpendent variables in regression problems. Leverage, being a summarizing index, cannot correctly identify the outliers in some directions of the factor space. We suggest a spectral decomposition method to overcome this difficulty. We show, through illustrations, that some outliers found by the spectral decomposition method, while not possible by the leverage method, are influential observations. Therefore, the method suggested is sharper than the classical leverage approach in diagnosing the outliers. |
本系統中英文摘要資訊取自各篇刊載內容。