Data-Driven Adjustment for Multiple Treatments

2025-03-12

AI Summary

This paper addresses causal effect identification under multiple treatments without experimental data or prior knowledge of the causal graph. We propose a purely data-driven method for automatically discovering valid covariate adjustment sets. Departing from reliance on a pre-specified causal graph, our approach introduces— for the first time—two novel adjustment-set construction pathways grounded solely in data-dependent relationships rather than graphical structure. By generalizing the notion of c-equivalence and establishing a sufficiency criterion for adjustability under multiple treatments, we overcome the restrictive single-treatment assumption. Our algorithm integrates conditional independence testing, d-separation principles, and nonparametric dependence analysis. We provide theoretical guarantees of sufficiency and empirically validate the method on synthetic and benchmark datasets, demonstrating high accuracy and robustness of the discovered adjustment sets, along with substantial improvements in the precision of multi-treatment causal effect estimation.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Constraint Satisfaction and Optimization: Other Foundations of Constraint Satisfaction

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Covariate adjustment is one method of causal effect identification in non-experimental settings. Prior research provides routes for finding appropriate adjustments sets, but much of this research assumes knowledge of the underlying causal graph. In this paper, we present two routes for finding adjustment sets that do not require knowledge of a graph -- and instead rely on dependencies and independencies in the data directly. We consider a setting where the adjustment set is unaffected by treatment or outcome. Our first route shows how to extend prior research in this area using a concept known as c-equivalence. Our second route provides sufficient criteria for finding adjustment sets in the setting of multiple treatments.

Problem

• Finding adjustment sets without causal graph knowledge
• Extending prior research using c-equivalence concept
• Providing criteria for multiple treatments adjustment

Innovation

• Data-driven adjustment without causal graph
• Extends prior research using c-equivalence
• Sufficient criteria for multiple treatments