Doubly robust estimation of causal effects.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Traditional methods like outcome regression and propensity score analysis rely on correct statistical model specification for unbiased causal effect estimation.
• Both methods, when used individually, are susceptible to bias if their underlying statistical models are misspecified.
• This highlights a critical limitation in accurately determining causal relationships in observational studies.

Purpose of the Study:

• To introduce and explain the concept of doubly robust estimation for causal inference.
• To demonstrate the practical application and benefits of this advanced statistical technique.
• To provide researchers with a more reliable method for estimating causal effects from observational data.

Main Methods:

• Doubly robust estimation integrates outcome regression with propensity score modeling.
• This combined approach requires only one of the two models (outcome or exposure) to be correctly specified for unbiased results.
• The study includes a conceptual overview, a worked example, and a simulation to evaluate performance.

Main Results:

• The doubly robust estimator offers unbiased causal effect estimates when either the outcome regression model or the propensity score model is correctly specified.
• Simulation results examined the performance of estimated and bootstrapped standard errors for this method.
• Supplementary materials provide a demonstration of the doubly robust property and a SAS macro for implementation.

Conclusions:

• Doubly robust estimation provides a more flexible and robust approach to causal inference compared to traditional methods.
• It mitigates the risk of bias stemming from model misspecification in either the outcome or exposure model.
• This method enhances the reliability of causal effect estimates in epidemiological and other research fields.