[Paper Review] Causal Inference for a Single Group of Causally-Connected Units Under Stratified Interference

Caleb H. Miles, Maya L. Petersen+1 more|arXiv (Cornell University)|Oct 26, 2017

Advanced Causal Inference Techniques20 references3 citations

TL;DR

This paper proposes a doubly-robust estimator for causal inference in a single group of causally-connected units under stratified interference, where each unit's outcome depends only on the total number of exposed units. The method enables valid inference for direct and overall effects using machine learning, validated through simulations and an application to nurse-led triage in Kenyan HIV clinics.

ABSTRACT

The assumption that no subject's exposure affects another subject's outcome, known as the assumption of no interference, has long held a foundational position in the study of causal inference. However, this assumption may be violated in many settings, and in recent years has been relaxed considerably. Often this has been achieved with either the aid of knowledge of an underlying network, or the assumption that the population can be partitioned into separate groups, between which there is no interference, and within which each subject's outcome may be affected by all the other subjects in the group, but only as a function of the total number of subjects exposed (the stratified interference assumption). In this paper, we consider a setting in which we can rely on neither of these aids, as each subject affects every other subject's outcome. In particular, we consider settings in which the stratified interference assumption is reasonable for a single group consisting of the entire sample, i.e., a subject's outcome is affected by all other subjects' exposures, but only via the total number of subjects exposed. This can occur when the exposure is a shared resource whose efficacy is modified by the number of subjects among whom it is shared. We present a doubly-robust estimator that allows for incorporation of machine learning, and tools for inference for a class of causal parameters that includes direct effects and overall effects under certain interventions. We conduct a simulation study, and present results from a data application where we study the effect of a nurse-based triage system on the outcomes of patients receiving HIV care in Kenyan health clinics.

Motivation & Objective

• To address causal inference in settings where interference occurs between all units in a single group, violating the no-interference assumption.
• To model interference such that each unit's outcome depends only on the total number of exposed units, not individual exposure patterns.
• To develop a doubly-robust estimator that incorporates machine learning for improved efficiency and robustness.
• To enable valid statistical inference for direct and overall causal effects under interventions in such interference structures.
• To apply the method to real-world data, specifically nurse-based triage in Kenyan HIV clinics, to assess impact on patient outcomes.

Proposed method

• The paper introduces a causal model where interference is stratified by the total number of exposed units in a single group, not by individual exposure patterns.
• It proposes a doubly-robust estimator that combines outcome regression and propensity score models, allowing for machine learning in both components.
• The estimator ensures consistent estimation if either the outcome model or the propensity score model is correctly specified.
• Inference is conducted using a robust variance estimator that accounts for the dependence structure induced by interference.
• The method supports estimation of direct effects (e.g., effect of triage on individual patients) and overall effects (e.g., system-wide impact).
• The approach is validated through a simulation study and applied to a real-world dataset on nurse-led triage in Kenyan HIV clinics.

Experimental results

Research questions

• RQ1 How can causal effects be estimated when all units in a population interfere with one another, but outcomes depend only on the total number of exposed units?
• RQ2 What is the performance of a doubly-robust estimator under this interference structure, especially when machine learning is used in modeling?
• RQ3 Can valid inference be achieved for direct and overall effects under interventions in such a setting?
• RQ4 How does the method perform in finite samples compared to standard estimators under interference?
• RQ5 What are the practical implications of this method in real-world public health interventions, such as nurse-led triage in HIV care?

Key findings

• The proposed doubly-robust estimator provides consistent estimates of causal effects under the stratified interference assumption, even if one of the working models (outcome or propensity score) is misspecified.
• The method maintains valid inference for both direct and overall effects, demonstrating robustness in finite samples.
• Simulation results show that the estimator achieves good coverage and low bias, especially when machine learning is used to model complex relationships.
• In the application to Kenyan HIV clinics, the method suggests a significant positive effect of nurse-based triage on patient outcomes, though the exact magnitude is not quantified in the source.
• The approach enables reliable causal inference in settings where traditional methods fail due to unstructured interference.
• The study demonstrates the feasibility of applying this method to real-world health interventions with complex interference patterns.