Scholarlink

<p>Understanding the statistical interdependencies among Sustainable Development Goal (SDG) indicators poses significant methodological challenges, particularly in the presence of multicollinearity and limited cross-country observations. This study provides a comparative assessment of regression techniques for analysing income-related inequality (SDG indicator 10.1.1) in EU Member States, using selected indicators related to health (Goal 3), education (Goal 4) and economic conditions (Goal 8) as explanatory variables. Given the strong correlation structure characterising SDG data, we implement and compare shrinkage-based regression approaches, including Ridge, LASSO, Elastic Net and Partial Least Squares (PLS). Model performance and coefficient stability are evaluated through a train-test validation framework combined with bootstrap resampling, allowing for a systematic assessment of robustness in a highly collinear setting. The empirical application highlights a subset of indicators that emerge as relatively stable determinants of income inequality, with health-related dimensions frequently emerging across specifications. No single method clearly dominates across all evaluation criteria, although Partial Least Squares frequently ranks among the more stable and reliable approaches. The study contributes primarily from a methodological perspective by proposing a replicable strategy for modelling SDG interdependencies under multicollinearity, offering guidance for future empirical research on socio-economic inequalities within the EU context.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparing regression methods under multicollinearity in SDG inequality analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月24日

查看全文
  • Mario Musella,
  • Ida Camminatiello,
  • Rosaria Lombardo

摘要

Understanding the statistical interdependencies among Sustainable Development Goal (SDG) indicators poses significant methodological challenges, particularly in the presence of multicollinearity and limited cross-country observations. This study provides a comparative assessment of regression techniques for analysing income-related inequality (SDG indicator 10.1.1) in EU Member States, using selected indicators related to health (Goal 3), education (Goal 4) and economic conditions (Goal 8) as explanatory variables. Given the strong correlation structure characterising SDG data, we implement and compare shrinkage-based regression approaches, including Ridge, LASSO, Elastic Net and Partial Least Squares (PLS). Model performance and coefficient stability are evaluated through a train-test validation framework combined with bootstrap resampling, allowing for a systematic assessment of robustness in a highly collinear setting. The empirical application highlights a subset of indicators that emerge as relatively stable determinants of income inequality, with health-related dimensions frequently emerging across specifications. No single method clearly dominates across all evaluation criteria, although Partial Least Squares frequently ranks among the more stable and reliable approaches. The study contributes primarily from a methodological perspective by proposing a replicable strategy for modelling SDG interdependencies under multicollinearity, offering guidance for future empirical research on socio-economic inequalities within the EU context.