Toward Causal Inference With Interference.

Area of Science:

• Statistics
• Epidemiology
• Social Sciences

Background:

• Causal inference typically assumes no interference between individuals.
• This assumption is violated in many real-world scenarios, such as infectious disease spread and social interventions.
• Interference occurs when one individual's outcome is affected by another's treatment assignment.

Purpose of the Study:

• To develop methods for causal inference in the presence of interference within groups.
• To define and estimate direct, indirect, total, and overall causal effects in clustered populations.
• To provide unbiased estimators and variance calculations for these causal effects.

Main Methods:

• Proposed new estimands for direct, indirect, total, and overall causal effects.
• Utilized a two-stage randomization experimental design (group-level then individual-level).
• Developed unbiased estimators and derived their variances.

Main Results:

• Established relationships among the proposed causal effect estimands.
• Demonstrated that the total causal effect equals the sum of direct and indirect effects.
• Presented unbiased estimators for the defined causal effects.

Conclusions:

• The proposed methodology effectively handles interference within groups in causal inference.
• The methods are applicable to diverse settings, including vaccine efficacy and housing voucher impact.
• This work extends causal inference techniques to more realistic, interdependent social and biological systems.