Bagged one-to-one matching for efficient and robust treatment effect estimation.

Area of Science:

• Epidemiology
• Biostatistics
• Statistical modeling

Background:

• Observational studies face challenges with bias from imbalanced baseline confounders.
• One-to-one matching (OOM) reduces bias but can lead to loss of efficiency by excluding subjects.
• Existing methods like OLS, IPW, and TML have limitations in balancing bias-reduction and efficiency.

Purpose of the Study:

• To introduce and evaluate the bagged one-to-one matching (BOOM) estimator.
• To combine the bias-reducing properties of OOM with the variance-reducing capabilities of bootstrap aggregation (bagging).
• To assess BOOM's performance against established methods in simulation and case studies.

Main Methods:

• Development of the BOOM algorithm and provision of R code for implementation.
• Simulation studies comparing BOOM to OOM, OLS, inverse probability weighting, and targeted maximum likelihood estimation.
• Evaluation of performance metrics including mean squared error, bias, variance, standard error estimation accuracy, and confidence interval coverage.

Main Results:

• BOOM achieves comparable bias reduction to OOM with significantly lower variance.
• BOOM's mean squared error is consistently comparable to or better than comparison methods across various simulation settings.
• In a case study, BOOM produced similar estimates to established methods but with narrower confidence intervals.

Conclusions:

• The BOOM estimator effectively reduces bias and variance in observational studies.
• BOOM offers improved efficiency and precision compared to traditional one-to-one matching and other common methods.
• BOOM presents a valuable advancement for robust causal inference from observational data.