causal-stuff

A playground for experimenting with causal inference and treatment effect estimation methods on open datasets. Built with Claude Code to speed up prototyping and testing different approaches.

What's here

Testing various causal methods (propensity score matching, Mahalanobis matching, double ML, metalearners) on benchmark datasets like LaLonde/NSW to see how well they recover true treatment effects.

Using frameworks like DoWhy, EconML, and CausalML to compare approaches and evaluate bias.

Setup

source .venv/bin/activate
uv sync
jupyter lab

Check out CLAUDE.md for more details on structure and conventions.