Comparison of two causal inference methods: linear regression and matching score and introduction of causal forest

Abstract

Under the context of the rapid development of artificial intelligence, the introduction of causal inference will improve the accuracy of machine analysis of data. This work aims to introduce three methods of calculating causal effect, linear regression, propensity score matching and causal forest, illuminate some about the combination of causal inference and matching-learning. Here, we substitute the dataset National Supported Work experiment by Lalonde (1986) into three methods and compare the results. We show experimentally that causal forest Causal forests can minimize data bias and obtain more accurate estimates of causal effects.