论文标题
与自然反事实互动效应的概念的总效应分解
Decomposition of Total Effect with the Notion of Natural Counterfactual Interaction Effect
论文作者
论文摘要
调解分析是基于定向的无环图获得因果推理的至关重要工具，该图已被广泛用于生物医学科学，社会科学，流行病学和心理学领域。总效应的分解提供了深刻的见解，以完全了解每个路径和相互作用项的休闲贡献。由于提出了四向分解方法来识别反事实框架中介导的相互作用效应，因此该想法已扩展到更复杂的情况，并具有非序列的多个介体。但是，该方法表现出局限性，因为因果结构包含介体之间的直接因果边缘，例如对依赖性和非识别性的不适当建模。我们提出了自然反事实互动效应的概念，发现总效应的分解可以与我们提出的概念一致地实现。此外，自然反事实相互作用效应克服了缺点，并具有清晰而重要的解释，这可能在很大程度上可以提高研究人员分析高度复杂的因果结构的能力。
Mediation analysis serves as a crucial tool to obtain causal inference based on directed acyclic graphs, which has been widely employed in the areas of biomedical science, social science, epidemiology and psychology. Decomposition of total effect provides a deep insight to fully understand the casual contribution from each path and interaction term. Since the four-way decomposition method was proposed to identify the mediated interaction effect in counterfactual framework, the idea had been extended to a more sophisticated scenario with non-sequential multiple mediators. However, the method exhibits limitations as the causal structure contains direct causal edges between mediators, such as inappropriate modeling of dependence and non-identifiability. We develop the notion of natural counterfactual interaction effect and find that the decomposition of total effect can be consistently realized with our proposed notion. Furthermore, natural counterfactual interaction effect overcomes the drawbacks and possesses a clear and significant interpretation, which may largely improve the capacity of researchers to analyze highly complex causal structures.
