Improving causal inference with a doubly robust estimator that combines propensity score stratification and weighting

Area of Science:

• Health research methodology
• Biostatistics
• Epidemiology

Background:

• Observational studies are crucial when randomized controlled trials are not feasible for estimating treatment effects.
• Conventional methods include regression adjustment, propensity score weighting, and doubly robust (DR) estimators.
• Doubly robust estimators offer advantages by modeling both outcome and treatment assignment.

Purpose of the Study:

• Introduce a novel doubly robust estimator utilizing marginal mean weighting through stratification (DR-MMWS).
• Evaluate the performance of the new DR-MMWS estimator against existing treatment effect estimation methods.
• Assess estimator accuracy under varying degrees of model misspecification.

Main Methods:

• Monte Carlo simulations were employed to compare estimators.
• The DR-MMWS estimator was compared to regression adjustment, propensity score weighting, and other DR methods.
• Propensity score and outcome models were intentionally misspecified to test robustness.

Main Results:

• Doubly robust estimators generally outperformed single-model approaches.
• The DR-MMWS estimator demonstrated superior performance when both propensity score and outcome models were misspecified.
• DR-MMWS performed comparably to other DR estimators when only the propensity score model was misspecified.

Conclusions:

• The DR-MMWS estimator is a promising new strategy for causal inference in observational studies.
• Researchers should consider DR-MMWS for its robust performance, particularly under model misspecification.
• This estimator offers a reliable alternative for estimating treatment effects from observational data.