Multiply robust matching estimators of average and quantile treatment effects

Area of Science:

• Causal Inference
• Statistical Modeling
• Biostatistics

Background:

• Propensity score matching is a traditional method for addressing confounding in causal inference.
• This method relies on stringent model assumptions that can be difficult to meet.
• Model misspecification can lead to biased estimates in causal inference.

Purpose of the Study:

• To propose a novel causal inference method, double score matching (DSM).
• To enhance robustness against model misspecification by utilizing multiple candidate models.
• To develop an estimator with multiple robustness properties.

Main Methods:

• Double score matching (DSM) combines propensity score and prognostic score matching.
• Multiple candidate models are used for each score to protect against misspecification.
• Asymptotic distribution is characterized using martingale representations and local Normal experiments.

Main Results:

• The de-biasing DSM estimator demonstrates a multiple robustness property.
• Consistency is achieved if at least one of the score models is correctly specified.
• Simulations show DSM outperforms single score matching and multiply robust weighting estimators, especially with extreme propensity scores.

Conclusions:

• Double score matching (DSM) offers a more robust approach to causal inference than traditional methods.
• The method provides protection against model misspecification through the use of multiple scores.
• DSM is particularly effective in scenarios with extreme propensity scores, improving estimation accuracy.