Scholarlink

<p>In Lauritzen-Wermuth-Frydenberg (LWF) chain graphical models (CGMs), conditional independence collapsibility means that the structure of a marginal model over a subset of variables can be exactly captured by the corresponding induced subgraph. Collapsibility helps reduce model complexity and improves interpretability. To describe collapsibility using graphical criteria, prior work has explored various interaction models. In this paper, we study conditions and criteria for collapsibility in LWF chain graphs. We introduce the concept of CI-removable node, which offers an equivalent condition for collapsibility when marginalizing over one variable. Extending this idea, we define a sequentially CI-removable set as a sufficient condition for collapsibility over multiple variables. We also use this condition to find minimal I-maps of marginal models in LWF chain graphs. This approach allows the efficient recovery of structural information from marginal models while preserving the original conditional independence relationships.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Conditional Independence Collapsibility in LWF Chain Graphical Models

  • 期刊论文

  • 发表日期：

    2026年1月29日

查看全文
  • Yi Sun,
  • Cai-qin Zhao

摘要

In Lauritzen-Wermuth-Frydenberg (LWF) chain graphical models (CGMs), conditional independence collapsibility means that the structure of a marginal model over a subset of variables can be exactly captured by the corresponding induced subgraph. Collapsibility helps reduce model complexity and improves interpretability. To describe collapsibility using graphical criteria, prior work has explored various interaction models. In this paper, we study conditions and criteria for collapsibility in LWF chain graphs. We introduce the concept of CI-removable node, which offers an equivalent condition for collapsibility when marginalizing over one variable. Extending this idea, we define a sequentially CI-removable set as a sufficient condition for collapsibility over multiple variables. We also use this condition to find minimal I-maps of marginal models in LWF chain graphs. This approach allows the efficient recovery of structural information from marginal models while preserving the original conditional independence relationships.