Causal Inference with the Instrumental Variable Approach and Bayesian Nonparametric Machine Learning
This work addresses causal inference challenges for researchers and practitioners by providing a more adaptable method, though it is incremental as it builds on existing instrumental variable and machine learning techniques.
The authors tackled the problem of causal inference with instrumental variables by introducing a flexible framework that uses Bayesian nonparametric machine learning, specifically BART and Dirichlet Process mixtures, to estimate functions and error distributions. The results show minimal loss in linear cases and dramatic improvements in nonlinear scenarios without manual tuning.
We provide a new flexible framework for inference with the instrumental variable model. Rather than using linear specifications, functions characterizing the effects of instruments and other explanatory variables are estimated using machine learning via Bayesian Additive Regression Trees (BART). Error terms and their distribution are inferred using Dirichlet Process mixtures. Simulated and real examples show that when the true functions are linear, little is lost. But when nonlinearities are present, dramatic improvements are obtained with virtually no manual tuning.