Study of Causal Analysis of Categorical Variables based Information Theory

本研究采用logit、loglinear和广义线性模型对分类变量进行因果分析。通常的路径分析方法是基于线性回归模型，LISREL模型被广泛应用于许多科学领域，如心理学、经济学、社会科学等。另一方面，在分类变量的路径分析中，不能假设变量之间的线性关系，如何评估因果关系是一个难题。第一年研究了基于logit模型的路径分析。因果效应，即直接效应和间接效应，通过基线对数比值比来定义，分配给每条路径的路径系数用Kullaback-Leibler信息来定义，该信息描述了所采用的模型与解释变量和响应变量独立的模型之间的距离。研究结果发表在论文《基于logit模型的分类变量路径分析》中，并提交发表。在最后一年，通过发展本研究第一年的结果，考虑了基于对数线性和广义线性模型的路径分析。在实证研究和观察研究中讨论了解释变量的因果效应。我正在完成论文的研究：广义线性模型中的效应分析。

In this study, causal analysis of categorical variables by using logit, loglinear, and generalized linear models was studied. The usual method for path analysis is based on linear regression models, and LISREL model is widely applied to many scientific fields, e.g. psychology, economics, social science, etc. On the other hand, in path analysis of categorical variables linear relationships among the variables cannot be assumed, and it is a difficulty question how the causal effects are assessed. In the first year, path analyisis based on logit models was studied. The causal effects, i.e. direct and indirect effects, were defined through baseline log odds ratio, and the path coefficient to be assigned to each path was defined by Kullaback-Leibler information that describes the distance between an employed model and the model that explanatory variables and response variables are independent. The results are reported in the paper :Path analysis of categorical variables based on logit models,and the above paper is submitted for publication. In the final year, path analysis based on loglinear and generalized linear models was considered by developing the results in the first year of this research. Causal effects of explanatory variables were discussed in both empirical and observational studies. I am getting the research completed as the paper :Effect analysis in generalized linear models.