頁籤選單縮合
| 題 名 | 直線迴歸中影響值之偵測 |
|---|---|
| 作 者 | 蘇秀媛; | 書刊名 | 國立臺灣大學農學院研究報告 |
| 卷 期 | 34:2 1994.06[民83.06] |
| 頁 次 | 頁174-185 |
| 分類號 | 319.51 |
| 關鍵詞 | 迴歸診斷法; 影響值; 截斷點; Regression diagnostics; Influential observation; Cutoff point; |
| 語 文 | 中文(Chinese) |
| 中文摘要 | 過去十多年來,有許多統計量被用來偵測直線迴歸模式中之影響觀測 值,本文係利用確定分布與Borferroni氏不等式求得這些統計量之截斷點,進而決 定那些統計量之表現較佳,且與其他截斷點所得之結果再做比較。此研究包括了 (i)一個詳細的電腦模擬研究及(ii)其他學者研究過之資料組之分析。結果顯示,利 用確定分布與Bonferroni氏不等式方法求得截斷點比利用其他截斷點者所得結果 較佳。在各種不同偵測法中,第i個觀測值之AP比值與F比值兩種偵測法表現最 好,Welsch氏統計量及Cook氏距離值之測定結果亦佳。總而言之,要由單一影響 值測定法來偵測出所有影響值似乎是不可能的,但郤可由少數某些測定法來偵測 出所有之影響值。 |
| 英文摘要 | In the past 16 years, many different statistics have been derivedfor detecting the influential observations in linear regression model.In this article, the exact distributions and Bonferroni's inequality an:used to obtain the cutoff points for these statistics. And a thorouglistudy is made for determining which statistics perform better thanothers. Also, a comparison between the results by using differentcutoff points is discussed. This research includes (i) a detailedcomputer simulation study, and (ii) analyses of data set studied by other authors. The results show that the statistics perform better when using exact distributions and Bonferroni's inequality to obtaincutoff points. APi ratio and FVARATIO, statistics perform better.Wi (Welsch's statistic) and C, (Cook's distance) measures are also quite appropriate for detection of influential observations. However, it is not possible to detect all influential observations by using only one measure. But a few measures may be enough for finding out all influential observations. |
本系統中英文摘要資訊取自各篇刊載內容。