Estimating Causal Effects in Partially Directed Parametric Causal Factor Graphs

TL;DR

This paper extends lifted causal inference methods to partially directed parametric causal factor graphs, enabling efficient and exact causal effect estimation in models with mixed causal and undirected relationships.

Contribution

It introduces PPCFGs, a generalization of parametric causal factor graphs, allowing lifted causal inference on partially directed graphs with less prior causal knowledge.

Findings

PPCFGs enable lifted causal inference in more general models.

Causal inference can be performed efficiently at a lifted level.

The approach maintains exactness in causal effect estimation.

Abstract

Lifting uses a representative of indistinguishable individuals to exploit symmetries in probabilistic relational models, denoted as parametric factor graphs, to speed up inference while maintaining exact answers. In this paper, we show how lifting can be applied to causal inference in partially directed graphs, i.e., graphs that contain both directed and undirected edges to represent causal relationships between random variables. We present partially directed parametric causal factor graphs (PPCFGs) as a generalisation of previously introduced parametric causal factor graphs, which require a fully directed graph. We further show how causal inference can be performed on a lifted level in PPCFGs, thereby extending the applicability of lifted causal inference to a broader range of models requiring less prior knowledge about causal relationships.