MIMS Final Project 2020

Causal Forests: A practical guide for understanding and implementation

Randomized controlled experiments allow us to uncover the effect of an action or intervention on a subsequent outcome. Long standing methods estimate this effect by measuring the average difference between people who experienced the intervention and those who are similar to them but did not experience the intervention. While the application of machine learning techniques to most estimation problems is trivial, the fundamental problem of causal inference means more sophistication is required. Causal Forests is one such method which modifies the Random Forest model to estimate causal effects. Additionally, it can exploit the large feature space characteristic of big data and abstract inherent heterogeneity in treatment effects. Our research objectives include an audit of existing literature employing forest based learning methods for causal inference and a white paper with tools and theory for quick easy application of the Causal Forest algorithm.

Last updated:

May 12, 2020