Bipartite Causal Inference with Interference.

Area of Science:

• Causal Inference
• Environmental Health
• Statistical Methodology

Background:

• Evaluating interventions is complex due to interconnected units and interference, where one unit's outcome depends on others' treatments.
• Existing causal inference methods struggle with interference, especially when treatments and outcomes are measured on distinct sets of units.

Purpose of the Study:

• Introduce and define bipartite causal inference with interference.
• Develop causal estimands for settings with distinct treatment and outcome units and interference.
• Adapt existing methods for empirical application in this novel setting.

Main Methods:

• Formulate definitions and causal estimands for bipartite causal inference with interference.
• Adapt an inverse probability of treatment weighted estimator.
• Apply the adapted estimator to a real-world dataset.

Main Results:

• The study formulates novel causal estimands for bipartite interference settings.
• An adapted inverse probability of treatment weighted estimator is proposed.
• The methods are applied to assess air pollution interventions' effects on cardiovascular hospitalizations.

Conclusions:

• Bipartite causal inference with interference provides a framework for complex intervention evaluation.
• The proposed methods and estimators offer a way to address interference in distinct treatment-outcome unit settings.
• Empirical application demonstrates the utility of these methods in environmental health research.