New Progress in the Application of Experimental Methods in Educational Economics
Education is a critical public policy issue.Within the constraints of limited public resources,policymak-ers have long been concerned about the extent to which improvements in the quality of the education system can influence human capital accumulation and drive endogenous economic growth.Since the late 20th century,evidence-based educa-tional reforms have increasingly shifted policy making from reliance on experience to reliance on objective evidence.The application and advancement of natural experimental methods have significantly enhanced the scientific rigor of educa-tional research and improved the effectiveness of educational decision-making.This paper first outlines the fundamental principles and potential pitfalls of causal inference in education and reviews the toolkit of natural experimental methods employed by education economists.According to the human capital theory,education is viewed as individual investments to enhance their future performance in the labor market.As such,"receiv-ing education"is an endogenous decision variable,influenced by factors such as individual ability,school resources,fam-ily background,distance to school,and compulsory education laws.While randomized controlled trials(RCTs)are re-garded as the"gold standard"for causal inference,practical challenges such as spillover effects,identity switching,sample attrition,and changes in participant behavior complicate the generalizability of RCT results.Consequently,educa-tion economists have widely adopted natural experimental techniques,including instrumental variables(IVs),difference-in-differences(DID),and regression discontinuity design,to provide empirical foundations for policymaking and the pro-motion of educational interventions.This paper systematically reviews the application and recent development of natural experimental methods in the field of education.Firstly,in studies of private returns to education,existing research has employed IVs such as distance,physiological factors,educational reforms,and random assignments to control endogeneity problems,using shocks such as extended schooling years,restoration of the college entrance exam,implementation of compulsory education laws,and higher education expansion to estimate returns to education and group disparities.Empirical studies have demonstrated sig-nificant gender and regional disparities in educational returns,indicating that individuals with higher income levels tend to have access to superior education.Furthermore,these studies suggest that proactive measures should be implemented to mitigate further exacerbations of educational inequality,as it is transmitted intergenerationally.Secondly,in studies of education's external effects,peer effects have been found to exist widely across and within groups defined by race,gen-der,class,family,community,and immigration status,with both positive and negative outcomes,such as crowding out and excessive competition.Empirical evidence suggests that moving to better peer groups and communities can result in positive peer effects.The implementation of the natural experimental methodology in this domain facilitates the mitigation of confounding variables and provides a significant benchmark for the formulation and refinement of educational policies.Thirdly,research on educational input and output has applied natural experimental methods to examine the impact of school investment,family background,shadow education,online education,and government funding on educational out-comes,yielding substantial findings.Additionally,this paper explores the empowering role of artificial intelligence(AI)technologies in advancing experi-mental methods in educational economics.Currently,educational artificial intelligence(EAI)is progressively blurring the boundaries between formal and informal learning,facilitating the transformation of educational environments into a syn-thesis of virtual and realities.Intelligent education aims to establish a personalized,appropriate,and dynamic educational ecosystem,with AI-assisted educational technologies,such as speech recognition,image recognition,and automated as-sessment,being widely implemented and advanced.As data increasingly becomes a crucial factor of production,the methodology for promoting causal inference in the education economy increasingly relies on the mining,application,and generation of educational big data.In the era of big data,the integration of machine learning with causal inference meth-ods has significantly enhanced the ability of researchers and policymakers to address complex issues.Deep learning can au-tomatically select and construct Ⅳs,identify breakpoints,and create control groups based on complex datasets,thereby improving the validity of causal inference within the Rubin causal model(RCM)framework.Moreover,counterfactual reasoning within the Structural causal model(SCM)framework can enhance the interpretability and stability of deep learning processes.With the rapid development of large AI-driven models,educational economics experiments are mov-ing toward more scientific predictions,leading to major transformations in empirical research methods.
Educational Experiment MethodsReturns to EducationEducational ExternalitiesArtificial Intelligence
万方数据
维普
