Research on Policy Evaluation Based on a Dual Machine Learning Model: Taking the Impact of the “Double Reduction” Policy on Students’ Academic Performance as an Example

Abstract

The “Double Reduction” policy is a major reform in China’s basic education field, and the impact assessment of this change on students’ academic performance is the core basis for strategy optimization. Previous evaluation methods were prone to selection bias and confounding variables, compromising the authenticity of results. This study employs panel data from students at School A (2020–2022) as a sample, treating the “Double Reduction” policy as a quasi-natural experiment. By applying a dual machine learning model to control for confounding variables such as individual student characteristics, family background, and school resources, it precisely identifies the causal relationship between the policy and academic outcomes. The study finds that the strategy has the characteristics of “overall improvement and structural improvement” in students’ academic performance, the average score of core subjects is increased, and the dispersion degree is reduced, the improvement effect on students with medium academic level is the most outstanding, and the influence on excellent students and students with learning difficulties is relatively moderate. Research demonstrates that the dual machine learning model effectively addresses endogeneity issues, providing robust methodological support for educational strategy evaluation. The “Double Reduction” policy enhances academic quality and efficiency through optimized learning environments.