CaSPECT: Discovering Causally Homogeneous Subgroups via Directed Spectral Clustering

2026-07-03

AI Summary

This study addresses the challenge that existing covariate-based clustering methods struggle to identify subpopulations with homogeneous causal mechanisms. The authors propose a novel framework integrating causal discovery with spectral clustering: individual-level causal similarity is defined via the topological structure of directed acyclic graphs (DAGs), which are constructed using Bootstrap-stabilized PC and DirectLiNGAM algorithms. To ensure robust edge orientation, a newly introduced Orientation Validation Score (OVS) is employed, while edge weights are calibrated by backdoor-adjusted average treatment effects, thereby avoiding reliance on pre-specified propensity score models. Spectral clustering is then performed in the Chung directed Laplacian embedding space to preserve consistency in causal propagation pathways. Experiments demonstrate that the method accurately identifies causally homogeneous subgroups exhibiting significantly positive treatment effects across synthetic data and real-world benchmarks—including LaLonde CPS1, IHDP, and 401(k)—effectively mitigating severe confounding bias.

Abstract

We propose \textbf{CaSPECT}, a causal spectral clustering framework for discovering causally homogeneous subgroups from observational data. Rather than clustering in covariate space, CaSPECT defines similarity through the topology of a learned directed acyclic graph (DAG); a bootstrap-stabilised PC algorithm recovers the causal skeleton; a novel \emph{Orientation Validation Score} (OVS) combines PC bootstrap evidence with DirectLiNGAM to orient edges robustly; directed edges are weighted by backdoor-identified average treatment effects estimated via OLS or double machine learning. Chung's directed Laplacian provides a spectral embedding in which individuals close together share the same causal propagation pathways. We establish almost-sure consistency of the full pipeline and validate the method through a controlled simulation study and on LaLonde CPS1, IHDP, and 401(k) datasets, where CaSPECT recovers a positive and statistically significant treatment effect within the causally comparable subpopulation and corrects for severe confounding without requiring a pre-specified propensity score model.

Problem

• causal homogeneity
• subgroup discovery
• observational data
• confounding bias
• treatment effect

Innovation

• causal spectral clustering
• directed acyclic graph
• Orientation Validation Score
• backdoor adjustment
• treatment effect heterogeneity