Causal effect estimation for multivariate continuous treatments.

Area of Science:

• Causal inference
• Observational studies
• Econometrics

Background:

• Covariate balancing is crucial for estimating causal effects in observational studies.
• Existing methods like one-dimensional entropy balancing have limitations with multiple covariates.

Purpose of the Study:

• To extend one-dimensional entropy balancing to a multidimensional approach for improved covariate balancing.
• To develop and evaluate parametric and nonparametric methods for estimating causal effects with multivariate continuous treatments.

Main Methods:

• Proposed a multidimensional entropy balancing method to address covariate balancing challenges.
• Developed both parametric and nonparametric estimation techniques for multivariate continuous treatments.
• Provided theoretical properties for the proposed estimation methods.

Main Results:

• Simulation studies demonstrated the superiority of the proposed multidimensional method over existing approaches.
• Applied the method to analyze smoking duration and frequency's impact on medical expenditure.
• Parametric results: smoking frequency increases medical costs; duration does not.
• Nonparametric results: a short-term decrease followed by a long-term increase in medical costs with increased smoking duration and frequency.

Conclusions:

• The multidimensional entropy balancing method offers a robust approach for causal inference with complex treatments.
• Smoking frequency significantly impacts medical expenditure, while duration shows a more complex, non-linear relationship.
• The proposed methods provide valuable tools for analyzing real-world causal relationships in observational data.