经济复杂性作为巴西区域人类发展的决定因素：跨聚合尺度的证据
Economic Complexity as a Determinant of Regional Human Development in Brazil: Evidence across Aggregation Scales
AI总结：
研究巴西城市人类发展指数，比较线性与非线性模型，通过5折交叉验证等评估，在市和直接地理区域两个层次分析。结果显示区域聚合模型优势明显，ICE是主要决定因素，纳入道路网络指标可提升性能，可解释提升机预测性能最佳。
AI中文摘要：
本研究调查了基于经济复杂性和复杂网络理论的模型应用于巴西城市人类发展指数（IDHM）时的预测能力。为此，经济复杂性指数被适配到巴西背景。主要目的是评估不同的区域发展分析方法，如地方生产复杂性和融入国家交通网络的结构，在多大程度上决定社会经济发展。方法比较了线性回归模型（岭回归、套索回归和弹性网络）和非线性模型（决策树和可解释提升机），通过5折交叉验证和网格搜索进行评估。分析在两个空间聚合层次上进行：市和直接地理区域。结果表明，区域聚合模型表现出更少的统计噪声和更大的稳定性，相对于其他变量，对ICE有更高的决定系数和更强的解释力。特别是在直接区域层次，单独的ICE成为IDHM的主要结构决定因素，表明区域生产复杂性本身就能解释该地域尺度上人类发展变化的很大一部分。此外，纳入道路基础设施网络的拓扑指标通常会提高性能。可解释提升机实现了最佳预测性能，在纳入网络指标时在直接区域层次达到R^2 = 0.8196。
英文摘要：
This study investigates the predictive capacity of models based on Economic Complexity and Complex Network Theory when applied to Brazil's Municipal Human Development Index (IDHM). For this purpose, the Economic Complexity Index was adapted to the Brazilian context. The main objective is to assess the extent to which different approaches to regional development analysis, such as local productive sophistication and structural integration into the national transportation network, determine socioeconomic development. The methodology compares linear regression models (Ridge, LASSO, and Elastic Net) and nonlinear models (Decision Trees and the Explainable Boosting Machine), evaluated through 5-fold cross-validation with grid search. The analyses were conducted at two levels of spatial aggregation: municipalities and Immediate Geographic Regions. The results show that regionally aggregated models exhibit less statistical noise and greater stability, achieving substantially higher coefficients of determination and greater explanatory power for ICE relative to the other variables. Notably, at the Immediate Region level, ICE alone emerges as the main structural determinant of IDHM, suggesting that regional productive sophistication by itself explains a large share of the variation in human development at this territorial scale. In addition, the inclusion of topological metrics from the road infrastructure network generally improved performance. The Explainable Boosting Machine achieved the best predictive performance, reaching R^2 = 0.8196 at the Immediate Region level when network metrics were included.
