On adaptive propensity score truncation in causal inference.

Area of Science:

• Causal inference
• Statistical modeling
• Biostatistics

Background:

• The positivity assumption is crucial for causal inference but can be practically violated.
• Violations lead to extreme propensity score (PS) estimates, impacting causal estimator performance.
• Current methods often truncate PS estimates to mitigate these issues.

Purpose of the Study:

• To propose a novel adaptive truncation method for propensity score estimation.
• To improve the finite sample performance of causal estimators.
• To introduce the Positivity-C-TMLE method based on collaborative targeted maximum likelihood estimation (C-TMLE).

Main Methods:

• Developed an adaptive truncation strategy for propensity score estimates.
• Integrated this strategy within the collaborative targeted maximum likelihood estimation (C-TMLE) framework.
• Conducted extensive simulations to compare the proposed method with existing estimators.

Main Results:

• The novel Positivity-C-TMLE estimator demonstrated superior performance in simulations.
• Achieved better point estimation accuracy compared to other commonly studied estimators.
• Provided improved confidence interval coverage, indicating enhanced reliability.

Conclusions:

• Adaptive truncation of propensity scores using Positivity-C-TMLE effectively addresses practical violations of the positivity assumption.
• The proposed method offers a robust approach for causal inference, enhancing both estimation and inference.
• Positivity-C-TMLE represents a significant advancement in handling propensity score challenges in causal effect estimation.