High Resolution Treatment Effects Estimation: Uncovering Effect Heterogeneities with the Modified Causal Forest

Area of Science:

• Causal inference
• Machine learning
• Statistical software

Background:

• High demand for inferring causal effect heterogeneity.
• Need for accessible, open-source statistical software for practitioners.
• Modified Causal Forest (mcf) as a causal machine learning approach.

Purpose of the Study:

• To introduce and demonstrate the utility of the open-source Python package 'mcf'.
• To showcase the package's ability to estimate aggregate treatment effects and causal effect heterogeneity.
• To provide inference for treatment effects at all identifiable resolutions.

Main Methods:

• Replication of three established studies from epidemiology, medicine, and labor economics.
• Implementation of the Modified Causal Forest (mcf) algorithm within an open-source Python package.
• Validation of the package's performance against existing research.

Main Results:

• The 'mcf' package successfully replicates aggregate treatment effects from previous studies.
• The package reveals novel insights into causal effect heterogeneity.
• Inference is provided for all identifiable resolutions of treatment effects estimation.

Conclusions:

• The 'mcf' package is a practical and comprehensive tool for modern causal heterogeneous effects analysis.
• It addresses the demand for open-source software in causal inference.
• Facilitates deeper understanding of treatment effect variations across different fields.