A comparison of causal inference methods for evaluating multiple treatment groups.

Area of Science:

• Statistics
• Machine Learning
• Causal Inference

Background:

• Causal inference uses counterfactuals to investigate causal questions.
• Machine learning integration offers flexible models but adds complexity to statistical inference.

Purpose of the Study:

• To systematically assess methods for causal inference with multiple treatment groups.
• To evaluate outcome regression, inverse propensity score weighting, double-robust estimators, super learners, and targeted maximum likelihood estimator (TMLE).

Main Methods:

• Numerical studies with complex data-generating models were conducted.
• Comparison of various causal inference estimators, including those using machine learning.

Main Results:

• Double-robust estimators combined with machine learning demonstrated lower biases.
• This approach yielded a valid variance estimator and improved 95% confidence interval coverage probabilities.

Conclusions:

• The double-robust estimator with machine learning is a favorable approach for causal inference in multiple treatment scenarios.
• This method enhances the reliability and accuracy of causal effect estimation.