User Guide

This user guide provides a comprehensive walkthrough on how to use the main features of the mcf. It includes several example scripts to help users, regardless of their technical expertise, to navigate and use the mcf package effectively.

Modified Causal Forest

• 1. Estimation of treatment effects
• 2. Data cleaning
• 3. Sampling weights and clustering
• 4. Common support
• 5. Feature selection
• 6. Post-estimation diagnostics
• 7. Experimental features
• 8. Computational Speed and Resources for Effect Estimation

Optimal Policy

• 1. Learning an optimal policy

Example scripts

We provide several example scripts in our example folder on GitHub. Below you also find the direct links to these scripts.

Modified Causal Forest

• Minimal example

• Example predicting on the retained filling sample from training

• Full example with all parameters used

• Example to show how to obtain prediction data on common support when multiple training data are used

• Example to show how the mcf with BGATE estimation can be implemented

• Example to show how QIATEs can be computed

• Instrumental-variable example (see the short IV overview)

Optimal Policy

• Minimal example

• Minimal example that jointly estimates a Modified Causal Forest and an Optimal Policy Tree

• Full example with all parameters used

• Example to show how to use the adjustments for policy score uncertainty

• Example to show how fairness adjustments can be implemented

• Treatment-version policy example