Comparison between linear regression, causal forest, and Bayesian causal forest models for causal inference in the presence of heterogeneous treatment effects : linear and non-linear data
In research, we often want to make causal claims about the treatment effect on a given population. However, “historically most datasets have been too small to meaningfully explore heterogeneity of treatment effects beyond dividing the sample into a few subgroups” (Wager & Athey, 2015). Furthermore, there is a concern that when looking for heterogeneous treatment effects, subsetting data multiple times may lead to spurious effects. Using simulated data, we will compare how well three different methods estimate the treatment effect given heterogeneous treatment effects in our data. The three methods we chose to examine are linear regression, causal forests and Bayesian causal forests.
Access the full paper at:
https://juinerurkar.com/heterogeneous-treatment-effects-linear-and-non-linear-data/