A model averaging approach for estimating propensity scores by optimizing balance.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Propensity score estimation is crucial for causal inference in observational studies.
• Existing methods like parametric modeling and machine learning have limitations in achieving covariate balance.

Purpose of the Study:

• To propose a new model averaging approach for propensity score estimation.
• To improve covariate balance and enhance the performance of inverse probability weighting (IPW) estimators.
• To evaluate the causal effect of infant feeding practices on childhood obesity.

Main Methods:

• Developed a model averaging technique combining parametric and nonparametric propensity score estimates.
• Conducted simulation studies across scenarios with varying interactions and nonlinearities.
• Applied the approach to real-world data on infant feeding and BMI Z-scores at age 4.

Main Results:

• The proposed propensity score estimator demonstrated reduced bias and smaller standard errors compared to existing methods under IPW.
• Model averaging focused on minimizing the Kolmogorov-Smirnov statistic yielded the best IPW estimator performance.
• Analysis indicated that formula or mixed feeding is associated with a higher likelihood of childhood obesity at age 4.

Conclusions:

• The novel model averaging approach offers a superior method for propensity score estimation, enhancing causal inference.
• Minimizing the Kolmogorov-Smirnov statistic is an effective objective for model averaging in IPW.
• Formula or mixed feeding may increase the risk of obesity in children by age 4.