Causal inference for treatment effects in partially nested designs.

Area of Science:

• Social Sciences
• Psychology
• Intervention Studies

Background:

• Partially nested designs (PNDs) are prevalent in intervention research.
• Existing methods for PND data analysis are well-developed, but causal inference, especially with nonrandomized treatments, remains under-researched.

Purpose of the Study:

• To address the research gap in causal inference for PNDs.
• To define and identify average causal treatment effects within PNDs using the expanded potential outcomes framework.

Main Methods:

• Utilized the expanded potential outcomes framework to define and identify causal treatment effects in PNDs.
• Formulated outcome models for causal interpretation and developed an inverse propensity weighted (IPW) estimation approach.
• Proposed a sandwich-type standard error estimator for IPW estimates.

Main Results:

• Simulation studies confirmed that both outcome modeling and IPW methods provide satisfactory estimates and inferences for average causal treatment effects.
• The proposed approaches were successfully applied to a real-world pilot study.

Conclusions:

• The study offers guidance and insights for causal inference in partially nested designs.
• Introduces new tools for researchers to estimate treatment effects in PNDs, enhancing analytical capabilities.