Scholarlink

<p>The global multidimensional poverty index (MPI) represents a significant advance beyond income-based poverty measures but faces critiques for its standardized, one-size-fits-all structure, which may obscure contextually salient drivers of deprivation. Drawing on a Khaldunian perspective emphasizing the role of the socio-material environment (<i>al-umran</i>) in shaping human capabilities, this study argues for a data-driven, context-sensitive poverty metric. Using microdata from the 2014 Moroccan census, we focus on the rural province of Taounate to empirically evaluate and adapt the standard MPI. We employ a machine learning framework, benchmarking random forest and XGBoost classifiers under various class imbalance treatments, to identify the most salient local predictors of poverty. Our results reveal a stark divergence from the global MPI: indicators of infrastructure, housing conditions, and spatial mobility, such as distance to a paved road, water access time, and tractor ownership emerge as dominant, while standard health and education indicators are absent. Leveraging these findings, we construct an adaptive MPI (A-MPI) by assigning equal weights to the top empirically selected indicators, allowing dimension weights to emerge <i>ex post</i>. The resulting A-MPI not only demonstrates a more severe and entrenched profile of rural poverty compared to the standard measure but also functions as a diagnostic tool, pinpointing specific, actionable constraints on human development. This research demonstrates that marrying the theoretical richness of the capability approach with the empirical power of machine learning yields a poverty metric that is both philosophically grounded and pragmatically vital for targeted, effective policy intervention in heterogeneous contexts.</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

Context-sensitive poverty measurement: an adaptive MPI using machine learning in rural Morocco

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Tariq Hadrachi,
  • Mohammed El Kamli

摘要

The global multidimensional poverty index (MPI) represents a significant advance beyond income-based poverty measures but faces critiques for its standardized, one-size-fits-all structure, which may obscure contextually salient drivers of deprivation. Drawing on a Khaldunian perspective emphasizing the role of the socio-material environment (al-umran) in shaping human capabilities, this study argues for a data-driven, context-sensitive poverty metric. Using microdata from the 2014 Moroccan census, we focus on the rural province of Taounate to empirically evaluate and adapt the standard MPI. We employ a machine learning framework, benchmarking random forest and XGBoost classifiers under various class imbalance treatments, to identify the most salient local predictors of poverty. Our results reveal a stark divergence from the global MPI: indicators of infrastructure, housing conditions, and spatial mobility, such as distance to a paved road, water access time, and tractor ownership emerge as dominant, while standard health and education indicators are absent. Leveraging these findings, we construct an adaptive MPI (A-MPI) by assigning equal weights to the top empirically selected indicators, allowing dimension weights to emerge ex post. The resulting A-MPI not only demonstrates a more severe and entrenched profile of rural poverty compared to the standard measure but also functions as a diagnostic tool, pinpointing specific, actionable constraints on human development. This research demonstrates that marrying the theoretical richness of the capability approach with the empirical power of machine learning yields a poverty metric that is both philosophically grounded and pragmatically vital for targeted, effective policy intervention in heterogeneous contexts.