Abstract:

With the global expansion of shadow education, its development exhibits significant spatiotemporal disparities. Based on POI data of shadow education institutions in the region, this study integrates spatial analysis, regression modeling, geographical detectors, and automated machine learning to investigate the evolutionary patterns and driving factors of these institutions from both macro- and micro-level perspectives. The findings reveal that: ① From 2000 to 2020, shadow education institutions experienced rapid growth; however, following the implementation of the "Double Reduction" policy in 2021, growth rates declined sharply and institutional exit rates increased steadily. ② Spatially, the distribution pattern evolved from a "core-periphery" structure to a "point-axis" belt-like configuration, forming four major development corridors. While institutions remain concentrated in core cities, they are steadily diffusing into surrounding smaller cities, indicating a clear trend of spatial spillover. ③ Different types of institutions demonstrate distinct spatial heterogeneity and divergent evolutionary trajectories. ④ At the macro scale, regional economic development is the primary determinant of distributional differences, whereas at the micro scale, the interaction between residential density and school resources serves as the dominant driver. The multi-method approach employed in this study enhances the robustness of mechanism identification. Overall, the research uncovers the dynamic spatial restructuring of shadow education under policy intervention, offering theoretical support and practical guidance for evaluating the impacts of the "Double Reduction" policy and for advancing regional education governance.

Key words: shadow education, Yangtze River Delta Urban Agglomeration, spatio-temporal evolution, influencing factors, machine learning, geographic detectors

摘要：

随着影子教育在全球范围的扩展，长三角城市群各地影子教育的发展呈现出显著的时空差异。基于长三角影子教育机构的POI数据，结合空间分析、回归分析、地理探测器和自动机器学习方法，本文旨在从宏观-微观视角解析城市群内影子教育机构的演化特征与影响因素。研究发现：①2000年至2020年，影子教育机构经历了快速扩张；自2021年“双减”政策实施后，增长速度显著放缓，机构消亡率持续上升。②在空间结构上，影子教育机构分布由“核心-边缘”向“点-轴”带状结构转变，形成四条主要发展带。尽管机构数量仍集中于中心城市，但正向周边中小城市持续扩散，呈现出明显的空间溢出趋势。③各类型机构在空间上呈现显著异质性与分化演进路径。④宏观尺度下，区域经济水平主导机构分布差异，微观尺度下则表现为住宅密度与学校资源的交互作用驱动。多方法的联合应用增强了机制识别的稳健性。研究揭示了政策干预下影子教育空间重构的动态过程，为“双减”政策效果评估及区域教育治理提供了理论基础与实践参考。

关键词: 影子教育, 长三角城市群, 时空演变, 影响因素, 机器学习, 地理探测器