Variance reduction in the inverse probability weighted estimators for the average treatment effect using the...

Area of Science:

• Medical Research
• Biostatistics
• Epidemiology

Background:

• Propensity methodology is crucial for comparing treatments in nonrandomized studies.
• Inverse probability weighted (IPW) estimators are standard for average treatment effect but suffer from high variance.
• High variance of IPW estimators hinders their practical application in medical research.

Purpose of the Study:

• To propose a smoothed IPW estimator inspired by Rao-Blackwellization.
• To reduce the variance of IPW estimators while maintaining accuracy.
• To enhance the reliability of treatment effect estimation in observational studies.

Main Methods:

• Developed a smoothing technique by replacing original weights with their mean over potential treatment assignments.
• Applied the smoothing method to standard IPW estimators.
• Investigated the application of smoothing to locally efficient and doubly robust estimators.

Main Results:

• The smoothed IPW estimator demonstrated substantial variance reduction compared to the original IPW estimator.
• Variance reduction ranged from two-to-sevenfold for tested IPW estimators.
• The smoothed estimator exhibited only a small increase in bias, preserving overall accuracy.

Conclusions:

• The proposed smoothing method effectively reduces variance in IPW estimators.
• This technique offers improved reliability for estimating average treatment effects in nonrandomized comparative studies.
• The method provides added protection against model misspecification when applied to doubly robust estimators.