几个条件独立性检验统计量的构造与比较分析  

Construction and Comparative Analysis of Several Conditional Independence Test Statistics

作　　者：聂思玥[1] 李梦花[1] 

机构地区：[1]南开大学经济学院

出　　处：《数量经济技术经济研究》2014年第2期137-147,共11页Journal of Quantitative & Technological Economics

摘　　要：变量间的条件独立性可视为在概率空间上对其因果关系的一种描述,可以通过检验变量之间的条件独立性来检验因果关系。通过详细介绍几个条件独立性检验统计量的构造方法和基本原理,包括线性模型假设下的Fisher-z检验统计量,在非线性模型下或无法确定变量之间的模型时,使用的3个非参数的条件独立性检验统计量,对这几个不同的条件独立性检验统计量的检验效率进行对比分析。The conditional independence between variables can be regarded as a description of its causality in the probability space, So one can test conditional inde- pendence between variables to test causality. This paper describes the construction principles of the several conditional independence test statistic in detail, including Fisher-z test statistic under linear model assumption and three non-parametric con- ditional independence test statistic under the circumstance that the model is non-linear or a model cannot be determined between the variables used. At last the efficiencies of the above test statistic in conditional independence tests are compared and analyzed.

关 键 词：条件独立性 因果分析 检验效率 

分 类 号：F224.0[经济管理—国民经济]