论土地利用变化统计分析中的多重共线性问题  

Multicollinearity Problems and Their Treatment in Statistical Studies of Land use Change

作　　者：罗湘华[1] 倪晋仁[1] 罗华铭[1] 

机构地区：[1]北京大学环境科学中心.水沙科学教育部重点实验室,北京100871

出　　处：《应用基础与工程科学学报》2001年第4期297-306,共10页Journal of Basic Science and Engineering

基　　金：国家自然科学基金委;水利部联合资助项目 (编号 5 9890 2 0 0 )

摘　　要：针对土地利用变化统计分析中普遍存在的自变量个数不多而多重共线性严重这一类问题 ,本文在分析共线性识别方法和当前处理共线性问题时常用的逐步回归法优缺点的基础上 ,引入全子集法来解决多重共线性问题 .以宝安区土地利用变化的统计分析为例 ,采用全子集法建立了土地利用变化解释和模拟模型 。In the statistical analysis of Land Use/Cover Change (LUCC), it′s very common that the number of independent variables is not quite large yet the multicollinearity among them is very serious. We introduced the All possible regression method to resolve this kind of problem on the basis of summarizing the collinearity diagnosis methods and analyzing the advantages and disadvantages of stepwise regression method, which is now commonly employed among internal scholars to eliminate collinearity. A research on the LUCC driving forces in Baoan District, Shenzhen, was taken as an example. Using the All possible regression method, we successfully avoided multicollinearity among independent variables and established the explaining and simulating model of LUCC in Baoan. The model results are generally fine. Contrast study shows that the All possible regression method is a more effective method to eliminate collinearity in the statistical analyses of LUCC mechanisms.

关 键 词：土地利用变化 多重共线性 统计分析 化解 宝安区 性问题 优缺点 效果 逐步回归法 识别方法 

分 类 号：R167[医药卫生—公共卫生与预防医学] F301[经济管理—产业经济]