Handling missing data in randomized experiments with noncompliance.

Area of Science:

• Statistics
• Prevention Research
• Causal Inference

Background:

• Treatment noncompliance and missing outcomes are frequent challenges in naturalistic field experiments.
• Existing statistical methods often fail to address both issues simultaneously in prevention research.
• This complicates accurate causal effect estimation in real-world settings.

Purpose of the Study:

• To demonstrate accessible statistical methods for estimating causal treatment effects considering both noncompliance and missing outcomes.
• To review and apply missing data assumptions within the complier average causal effect (CACE) framework.
• To introduce sensitivity analysis using alternative missing data assumptions for robust causal effect estimation.

Main Methods:

• Review of missing data assumptions, including latent ignorability.
• Application of parametric complier average causal effect (CACE) estimation.
• Sensitivity analysis using a range of alternative missing data assumptions.

Main Results:

• Identification and estimation of causal effects are feasible even with noncompliance and missing outcomes.
• Alternative missing data assumptions yield a range of causal effect estimates, facilitating sensitivity analysis.
• The proposed methods offer a more reliable approach to causal inference in prevention research.

Conclusions:

• Statistical methods can effectively handle treatment noncompliance and missing outcomes in field experiments.
• Using a range of missing data assumptions enhances the robustness of causal effect estimates.
• This approach improves the validity of findings in prevention research by addressing common data complications.