Ensemble and calibration multiply robust estimation for quantile treatment effect.

Area of Science:

• Biostatistics
• Causal Inference
• Health Economics

Background:

• Estimating causal effects of treatments on health outcomes is crucial.
• Existing methods for quantile treatment effects often rely on strict model assumptions (propensity score or conditional density) that are hard to validate.
• This poses a challenge in real-world biomedical and health intervention evaluations.

Purpose of the Study:

• To develop novel, robust estimators for quantile treatment effects.
• To overcome the limitations of existing methods that require correct specification of single models.
• To provide more reliable causal inference in health outcome studies.

Main Methods:

• Developed calibration estimators using multiple imputation and inverse probability weighting via empirical likelihood.
• Constructed ensemble estimators to reduce computational load while maintaining robustness.
• Allowed for multiple models for propensity scores and conditional density vectors.

Main Results:

• Proposed estimators are "multiply robust," meaning they are consistent if at least one of the specified models is correct.
• Ensemble estimators offer a computationally efficient way to achieve multiple robustness.
• Simulations demonstrated the good finite sample performance of the new estimators.

Conclusions:

• The new calibration and ensemble estimators offer a more flexible and reliable approach to causal inference for quantile treatment effects.
• These methods are particularly valuable in biomedical research where model uncertainty is common.
• Applications to birthweight and clinical trial data illustrate the practical utility of the proposed techniques.