Propensity scores based methods for estimating average treatment effect and average treatment effect among treated: A...

Area of Science:

• Statistics
• Epidemiology
• Biostatistics

Background:

• Propensity score methods are crucial for estimating average treatment effect (ATE) and average treatment effect among treated (ATT) in observational studies.
• Commonly, propensity scores are estimated using logistic regression, but model misspecification can lead to biased results.
• Generalized Boosting Method (GBM) offers an alternative for propensity score estimation, capturing complex covariate relationships.

Purpose of the Study:

• To conduct a comparative analysis of popular propensity score-based statistical methods.
• To evaluate the performance of these methods when propensity scores are estimated using logistic regression versus GBM.
• To assess the accuracy of Average Treatment Effect (ATE) and Average Treatment Effect Among Treated (ATT) estimates under different scenarios.

Main Methods:

• Utilized propensity score matching, regression, stratification, inverse probability weighting (IPW), and doubly robust (DR) estimating equations.
• Estimated propensity scores using both traditional logistic regression and the Generalized Boosting Method (GBM).
• Performed extensive simulations to compare the performance of various methods and propensity score estimators.

Main Results:

• Estimates for ATE and ATT varied significantly depending on the statistical method employed.
• Regression-based methods showed limitations in estimating both ATE and ATT, irrespective of the propensity score estimation technique.
• Inverse Probability Weighting (IPW) and stratification yielded reliable ATT estimates when the propensity score model was correctly specified.
• Stratification, IPW, and DR methods produced ATE estimates close to true values when propensity scores were accurately specified by logistic regression or estimated via GBM.

Conclusions:

• Regression methods are generally unsuitable for estimating ATE and ATT in observational studies.
• IPW and stratification are recommended for reliable ATT estimation, particularly with correctly specified propensity score models.
• Stratification, IPW, and DR methods demonstrate robustness in ATE estimation when propensity scores are well-specified or estimated using GBM.