Detecting the unseen poor in multidimensional poverty assessment using Bayesian approaches

Abstract This study introduces a hierarchical Bayesian framework designed to enhance multidimensional poverty measurement by explicitly accounting for classification errors often ignored in conventional methodologies. Traditional deterministic approaches, such as the Alkire-Foster method, assume zero measurement errors (i.e., data collection and, more importantly, unfulfiled assumptions), leading to systematic distortions in poverty estimates and vulnerability assessments.To address this methodological limitation, we develop a probabilistic model that integrates false positive and false negative rates into the measurement process. Applying this framework to household data from Medellín, Colombia, we demonstrate that ignoring classification uncertainty results in substantial underestimation of social deprivation. Our findings indicate that even modest error rates (1–2%) can inflate true poverty prevalence by approximately 16%, while vulnerability assessments are even more sensitive, potentially underestimating future risk by up to 290%. By reconceptualizing poverty measurement as an inherently probabilistic process, this paper offers a rigorous statistical tool for correcting these biases. The proposed model provides policymakers with more robust estimates for resource allocation, ensuring that statistically “invisible” households are accurately identified. This contribution aligns with the need for methodological precision in social indicators, offering a flexible solution to improve the quality of data used for equitable social policy design.