Estimating Heterogeneous Treatment Effects Using Causal Forests: Two Applications Concerning Romanian Education System and Incentives For College Achievement

The causal forests algorithm is widely known as a machine learning technique that is powerful in drawing causal effect inferences of a treatment. Meanwhile, linear regression model as a more statistics-based technique can also estimate treatment effects. In this study, we apply both methods to two different datasets concerning the Romanian Baccalaureate exam out- comes and the students' college achievement measures to answer the following two questions. The first one is do the two estimation methods give similar results for estimating average treatment effects, and the second one is can causal forests generate accurate treatment effect estimates if being trained with few observations. The empirical findings show that the two methods can give accurate and similar estimation results if large enough datatset is utilized. Moreover, when being trained with small dataset, the causal forests algorithm cannot correctly give estimates however, it can still capture potential treatment heterogeneity.