Augmented Weighted Estimators Dealing with Practical Positivity Violation to Causal inferences in a Random...

Area of Science:

• Causal inference
• Biostatistics
• Epidemiology

Background:

• The inverse probability of treatment weighted (IPTW) estimator is crucial for causal inference.
• IPTW relies on ignorability and positivity assumptions.
• Practical violations of the positivity assumption, such as sampling zeros, can bias IPTW estimates.

Purpose of the Study:

• To propose a novel method to address sampling zeros in IPTW estimation when treatment assignment is clustered.
• To improve the accuracy and robustness of causal effect estimation in the presence of positivity assumption violations.

Main Methods:

• Augmenting the IPTW estimating function with estimated potential outcomes for clusters lacking treatment or control observations.
• Utilizing a random coefficient model for potential outcomes within clusters.
• Addressing practical violations of the positivity assumption in clustered data.

Main Results:

• The augmented estimating function converges in expectation to zero, yielding consistent causal estimates.
• The proposed method demonstrates good performance in both simulated and real-world datasets.
• Successful application in a teacher preparation evaluation study.

Conclusions:

• The developed method effectively resolves the sampling zero problem in IPTW for clustered data.
• This approach enhances the reliability of causal inferences derived from observational studies.
• The method is implementable in existing statistical software.