地理加权回归模型的多重共线性诊断方法

The Diagnostic Approach of Multicollinearity in Geographically Weighted Regression Model

地理加权回归是常用的空间分析方法,已广泛应用于各个领域,但利用此方法进行回归分析前,往往忽略了对设计矩阵进行局部多重共线性的诊断,从而导致对模型的估计不准确。因此,本文在引入了全局模型的多重共线性诊断方法的基础上,对这些方法进行了改进,改进后构建了加权方差膨胀因子法和加权条件指标方法——分解比法,用于诊断地理加权回归模型设计矩阵的多重共线性问题。实验结果表明,多重共线性不存在于全局模型,而可能存在于局部模型中,构建的两种方法能够有效地诊断地理加权回归模型的多重共线性问题,且加权条件指标方法——分解比法比加权方差膨胀因子法在诊断多重共线性问题上更有优势。

Geographically weighted regression is one of the most widely adopted spatial analysis approaches and has been used in many applications. However, the problem of muhicollinearity could prevent the exact estimation of model estimation. Therefore, this paper proposed the modified weighted VIF and CIVDP approaches to diagnosis the multicollinearity of GWR models. The simulation experi- ment shows that global model does not exist multicollinearity and GWR model may exist, and verifies the effectiveness of diagnosing multicollinearity in GWR model through these two methods, also can be learned that the weighted CIVDP than weighted VIF has more advantages is the diagnosis of multicollinearity problem.