摘要
目的结合肺癌危险因素研究中变量的筛选过程，探讨在涉及较多自变量的大型多元回归分析中，变量间多重共线性的诊断和处理方法。方法首先将经单因素分析筛选的变量进行相关分析，得出相关系数矩阵Ｒ的特征值，用主成分分析法判定自变量间是否存在多重共线性以及存在几个多重共线性关系。然后将这些自变量进行正交旋转，取得旋转后公因子所对应的自变量及其多重共线性关系，结合专业知识和以往研究的经验加以去除。结果将去除多重共线性的自变量引入多元回归模型，即可取得比较满意的结果。结论在大型多元回归分析中用上述方法进行多重共线性的诊断和处理是可行的。
Objective It is very difficult to discover and handle the collinearity in multiple regression.We try
to find a way to deal with it in the casecontrol study of risk factors in lung cancer. Methods First,
correlative analysis is conducted to form the correlative coefficient matrix R of the explanation
variables.Then the eigenvalues in the matrix are used to determine if there are collinear
relationships among the variables by principal companent analysis.If so,using factor
analysis,promax rotation is done to find the factors and the corresponding collinear variables.
Results Collinear variables are excluded according to the abovementioned result and academic
experience. Conclusion the combination of principal component analysis and factor analysis
can cotribute successfully to the discovery and handling of collinearity.
出处
《中国卫生统计》
CSCD
北大核心
1999年第3期136-138,共3页
Chinese Journal of Health Statistics
关键词
肿瘤
危险因素
筛选
多重共线性
诊断
处理
CollinearityMultiple regressinPrincipal component analysisFactor analysis
