Causal Effect Analysis of Extracurricular Tutoring Based on Random Forest Propensity Score Matching: Evidence from Student Academic Performance

The prevalence of extracurricular tutoring has sparked ongoing debates regarding its causal impact on student academic performance. Traditional assessment methods often fail to address selection bias and complex nonlinear relationships inherent in educational data. This study proposes a machine learning enhanced approach, Random Forest Propensity Score Matching (RF-PSM), to overcome the limitations of conventional propensity score methods when analyzing high dimensional observational data. By leveraging random forests for propensity score estimation, the method captures intricate interactions among student characteristics while maintaining robust covariate balance. The analysis utilizes a nationally representative student performance dataset, incorporating demographic, socioeconomic, and prior academic achievement variables. Key findings reveal significant heterogeneous treatment effects: tutoring demonstrates the strongest positive impact on median performing students, whereas effects diminish for both high and low achievers. The methodological contribution lies in demonstrating RF-PSM's superior performance over logistic regression based matching through reduced bias in effect estimation. Practically, these results inform targeted educational policies by identifying student subgroups that benefit most from supplemental instruction. The study underscores the potential of combining machine learning with causal inference frameworks to derive more nuanced insights from educational big data.