Research on Description and Utilization of Meta-Information in Psychological data

本研究的目的是构建辅助心理学研究中数据分析活动的工具，特别是使用代表变量之间复杂关系的统计模型。(a)我们开发了一个用Prolog编码的程序来分析有向无环图（DAG）的马尔可夫性质。主要研究成果如下：两个指定相同条件独立性的概率族的dag称为“马尔可夫等价”。在小型PC上，我们成功地将统计等效的DAG族分类到6阶。如果观察到所有相关变量，则确定一个变量对另一个变量的因果影响是直接的。但在某些变量未被观察到的情况下，这个推论就不是微不足道的了。Pearl（1995）给出了推导因果效应估计公式的概率变换规则。这些规则包含了对图的马尔可夫性质的复杂检验。我们的软件能够从图拓扑中自动推导出公式。(b)统计元数据描述的研究趋势。我们的重点是两种基于XMLsed的元数据格式。一种是ICPSR中的DDI（数据文档计划），它为调查代码本指定了广泛的格式。另一个是柴田教授课题组在庆应义塾大学的ddd项目。我们成功地使用基于Prolog的XML解析器从XML格式的码本中灵活地检索信息。

The purpose of the research is to construct tools for assisting data-analysis activities in psychological study, especially usage of statistical models that represent complex relationship between variables.(a) We developed a program coded in Prolog to analyze Markov properties of directed acyclic graphs (DAG). The major research results are as follows. Two DAGs that specify the probability families of the same conditional independences are called "Markov equivalent". We succeeded to classify the statistically equivalent DAG families up to order six on small PC. Identifying causal influence of a variable to another is straight if all the related variables are observed. But in cases where some variables are unobserved, the inference is not trivial. Pearl (1995) gave probability transformation rules that deduce causal effect estimation formula. The rules contain complex inspections of Markov properties of the graph. Our software is able to deduce automatically the formula from the graph topology.(b) we surveyed the research trends in statistical metadata description. Our focus is on two XMLsed metadata formats. One is DDI (Data Documentation Initiative ) in ICPSR, which specifies an extensive format for survey codebook. Another is DandD project in Keio University by Prof. R. Shibata's research group. We used Prolog based XML parser successfully for flexible information retrieval from codebooks in XML formats.