Exact Confidence Intervals in the Presence of Interference

Area of Science:

• Statistics
• Experimental Design
• Causal Inference

Background:

• Two-stage randomized experiments are common in fields like public health and social sciences.
• Estimating treatment effects with partial interference presents statistical challenges.
• Existing methods for exact confidence intervals may lack precision.

Purpose of the Study:

• To propose novel exact confidence intervals for treatment effects in two-stage randomized experiments.
• To address the challenge of partial interference in these experimental settings.
• To improve the precision of treatment effect estimation.

Main Methods:

• Development of new exact confidence intervals for binary outcomes.
• Utilizing a two-stage randomization design.
• Comparison with intervals based on the Hoeffding inequality.

Main Results:

• The proposed exact confidence intervals are demonstrated to be valid.
• Empirical studies show the new intervals have a narrower width than previous exact intervals.
• This narrower width indicates increased precision in treatment effect estimation.

Conclusions:

• The new exact confidence intervals provide a more precise tool for analyzing treatment effects in two-stage experiments with partial interference.
• These findings offer practical improvements for researchers in various scientific disciplines.
• The proposed method enhances the reliability of causal effect estimation.