Model-assisted sensitivity analysis for treatment effects under unmeasured confounding via regularized calibrated...

Area of Science:

• Statistics
• Econometrics
• Epidemiology

Background:

• Unmeasured confounding poses a significant challenge in estimating causal effects from observational data.
• Existing methods for sensitivity analysis often rely on strong assumptions or lack robustness.

Purpose of the Study:

• To develop novel statistical methods for sensitivity analysis in the presence of unmeasured confounding.
• To derive new population bounds and robust estimators for average treatment effects.

Main Methods:

• The study proposes new population bounds based on weighted linear outcome quantile regression.
• It introduces doubly robust point estimators and model-assisted confidence intervals for relaxed population bounds.
• Methods involve regularized calibrated estimation with Lasso penalties for model fitting.

Main Results:

• New representations for sharp population bounds and doubly robust estimating functions are provided.
• Relaxed population bounds are derived, offering greater flexibility.
• The developed confidence intervals are valid under certain model misspecifications and are doubly robust for linear outcome mean regression.

Conclusions:

• The proposed methods enhance the estimation of average treatment effects under unmeasured confounding.
• The R package RCALsa implements these novel techniques for practical application.
• This work contributes to more reliable causal inference from observational studies.