Estimating Average Treatment Effects With Support Vector Machines.

Area of Science:

• Machine Learning
• Causal Inference
• Statistical Modeling

Background:

• Support Vector Machine (SVM) is a widely used classification algorithm.
• Estimating causal effects requires balancing covariates between treatment and control groups.
• Existing methods for covariate balancing have limitations.

Purpose of the Study:

• To demonstrate the utility of SVM for covariate balancing and causal effect estimation.
• To adapt SVM as a kernel-based weighting procedure for improved balance and sample size.
• To analyze the trade-off between covariate balance and effective sample size controlled by SVM's regularization parameter.

Main Methods:

• Adapting the SVM classifier as a kernel-based weighting procedure.
• Minimizing maximum mean discrepancy between treatment and control groups.
• Maximizing effective sample size using SVM's regularization parameter for balance-sample size frontier computation.

Main Results:

• SVM effectively balances covariates and estimates average causal effects under unconfoundedness.
• SVM provides a continuous relaxation of the largest balanced subset problem, linking to cardinality matching.
• The regularization parameter in SVM controls the bias-variance trade-off in causal effect estimation.

Conclusions:

• The proposed SVM-based methodology is competitive with state-of-the-art covariate balancing methods.
• SVM offers a flexible approach to balancing covariates and estimating causal effects.
• Simulation and empirical studies validate the performance of the SVM procedure.