Causal Inference Made Easy: A Review of DoWhy Package in Python Programming Language

• In Event: Effect Sizes, Multilevel Models, and Causal inference

Abstract

This review covers an interface to causal inference methods. DoWhy, developed by Microsoft, is a python library for causal reasoning that supports explicit modeling and testing causal assumptions. The methods supported by DoWhy can promote modeling for problems of interest to educational statisticians, given its interpretable output and robustness checks. We provide readers with non-technical descriptions of basic ideas underlying each method. However, each technique is challenged with assumptions. Fortunately, DoWhy tests the robustness of the estimates to violations of those assumptions. We review the core concepts of DoWhy, such as the Bayesian graphical model framework, allowing users to specify what they know and what they don’t know about the data. Then, we apply these concepts to large-scale educational data.

Authors

• Zachary Kendall Collier, The University of Delaware

• Liu Liu, University of Washington - Seattle