Causal K-Means Clustering.

Area of Science:

• Causal inference
• Machine learning
• Statistical modeling

Background:

• Population-level causal effects can mask important subgroup variations.
• Identifying unknown subgroup structures is crucial for accurate treatment effect evaluation.
• Existing methods struggle with uncovering hidden heterogeneity in treatment effects.

Purpose of the Study:

• To propose a novel method, Causal k-Means Clustering, for identifying unknown subgroups with heterogeneous treatment effects.
• To develop and analyze estimators for subgroup effect identification.
• To extend the framework for clustering with various pseudo-outcomes.

Main Methods:

• Leveraging the k-means clustering algorithm to uncover unknown subgroup structures.
• Developing a plug-in estimator and studying its convergence rate.
• Creating a bias-corrected estimator using nonparametric efficiency theory and double machine learning.

Main Results:

• The proposed Causal k-Means Clustering effectively identifies unknown subgroup structures.
• The bias-corrected estimator achieves fast root-n rates and asymptotic normality.
• The framework is extensible to various pseudo-outcomes and applicable to complex studies.

Conclusions:

• Causal k-Means Clustering provides a powerful tool for uncovering hidden heterogeneity in treatment effects.
• The developed estimators offer efficient and statistically sound methods for subgroup analysis.
• The approach is particularly valuable for modern studies with multiple treatment levels and complex outcomes.