體驗區
|
| 篇名 |
為何多重共線性診斷不宜採用VIF=10為決斷值
|
| 並列篇名 |
Why a Variance Inflation Factor of 10 Is Not an Ideal Cutoff for Multicollinearity Diagnostics
|
| 作者 |
鄭承昌(Cheng-Chang Jeng)
|
| 中文摘要 |
進行迴歸分析時，常用變異數膨脹因子（Variance Inflation Factor, VIF）當作自變項是否具共線性的判別準則。過去研究者所建議的VIF決斷值，從2到10皆有，標準並不一致。本研究採用R模擬，針對不同數量的自變項（n）及自變項間的相關係數（r）計算VIF，得以下結果：（1）VIF和自變項的個數以及自變項間的相關係數有關；（2）當自變項間的相關係數為固定值，且自變項的個數趨近於無窮多時，VIF會趨近於一特定的極限值；（3）共線性的判別不應該採行固定的VIF值為決斷值，應考量自變項個數及所能容忍的相關係數來判斷。基於結果，本研究提出VIF決斷值表，協助研究者能依據其迴歸分析中自變項個數及自變項相關程度來尋找VIF的決斷標準。最後，文末提出本研究的限制與建議。
|
| 英文摘要 |
In linear regression analysis, the variance inflation factor (VIF) is often used to determine whether multicollinearity exists among independent variables (IVs). Despite its frequent use, no consensus has been achieved regarding a VIF threshold that reliably indicates multicollinearity. Although researchers have historically indicated that cutoff values ranging from 2 to 10 should be used, no single value has gained universal acceptance. To address this problem, this study used an R-based platform to calculate VIF values under various conditions, including different numbers of IVs (denoted as k) and paired correlation coefficients between IVs (denoted as r). The study discovered that VIF values are influenced by both the number of IVs and the degree of correlation between them. Moreover, when the correlation coefficient is held constant and the number of IVs increases infinitely, the VIF tends to converge at a limit. The study also asserts that employing a universal VIF cutoff for multicollinearity detection is impractical because the cutoff must be determined with consideration of both the specific number of IVs in a linear regression model and the correlation coefficients researchers deem to be acceptable. The study developed a table of VIF cutoff values to aid researchers in identifying suitable cutoff values for their linear regression analyses. The study concludes by discussing its limitations.
|
| 頁次 |
067-093
|
| 關鍵詞 |
線性迴歸
、多重共線性
、變異數膨脹因子
、Linear Regression
、Multicollinearity
、Variance Inflation Factor (VIF)
|
| 卷期 |
57:2
|
| 日期 |
202310
|
| 刊名 |
教育研究學報
|
| 出版單位 |
國立臺南大學
|
| DOI |
10.53106/199044282023105702004
|
|
相關文獻 :
客服信箱