Model misspecification and robustness in causal inference: comparing matching with doubly robust estimation

Area of Science:

• Causal inference
• Statistical modeling
• Econometrics

Background:

• Propensity scores are crucial for reducing bias in observational studies.
• Matching and doubly robust estimators are common methods for causal effect estimation.
• Understanding estimator robustness to model misspecification is vital for reliable inference.

Purpose of the Study:

• To compare the robustness properties of matching and doubly robust estimators.
• To investigate the impact of propensity score model misspecification on causal effect estimation.
• To evaluate the finite sample performance of these estimators.

Main Methods:

• Described robustness properties of matching and subclassification estimators.
• Analyzed conditions under which propensity score model misspecification yields consistent estimation.
• Conducted simulations comparing matching, inverse probability weighting, and doubly robust estimators.

Main Results:

• Matching estimators are robust to misspecification if the incorrect model is a covariate score.
• In simulations, the matching estimator had a smaller mean squared error than inverse probability weighting and doubly robust estimators.
• Matching offers robustness advantages when the propensity score model is misspecified.

Conclusions:

• Matching estimators can be robust to propensity score model misspecification.
• The matching estimator showed better finite sample performance in the simulated scenarios.
• Researchers should consider the robustness properties of estimators when choosing a method for causal inference.