Doubly robust proximal synthetic controls

Area of Science:

• Econometrics
• Biostatistics
• Epidemiology

Background:

• Synthetic control (SC) methods estimate treatment effects for single units using panel data by creating a weighted combination of control units.
• Traditional SC methods require accurate modeling of counterfactual outcomes and precise pre-treatment trajectory matching, which can be restrictive.

Purpose of the Study:

• To develop novel nonparametric identifying formulas for the average treatment effect for the treated unit.
• To introduce a new doubly robust estimator that is consistent if either the outcome or weighting model is correctly specified.
• To evaluate the effectiveness of pneumococcal conjugate vaccine using these advanced causal inference techniques.

Main Methods:

• Developed two novel nonparametric identifying formulas for average treatment effect inspired by proximal causal inference.
• Introduced the concept of covariate shift to synthetic control methods for identification conditional on treatment assignment.
• Created two treatment effect estimators using generalized method of moments, including a doubly robust option.

Main Results:

• The proposed methods provide new ways to identify and estimate treatment effects without strict modeling assumptions.
• The doubly robust estimator offers improved reliability by requiring only one of the two models to be correctly specified.
• The application to pneumococcal conjugate vaccine data demonstrated the method's utility in real-world public health evaluations.

Conclusions:

• The novel nonparametric synthetic control methods offer flexible and robust approaches for causal inference with panel data.
• The doubly robust estimator enhances the reliability of treatment effect estimation in the presence of model misspecification.
• These methods provide valuable tools for evaluating interventions, such as vaccination programs, in observational studies.