Missing data and multidimensional analysis with the Alkire-Foster method: A literature review

Abstract

This article addresses the problem of missing data in measuring multidimensional poverty, with particular attention to its impact on the axiomatic validity of the Alkire-Foster method, regularly used to measure this broad concept of poverty. It argues that the absence of data is not purely a technical problem prone to generating bias, but can also compromise the identification of deprivation and violate some of the fundamental axioms of the method: monotonicity, decomposition by subgroups, and dimensional decomposition. Based on a critical review of the literature, four imputation approaches are compared: complete case analysis, multiple imputation by chained equations (MICE), matrix decomposition techniques (such as Soft-Impute), and Bayesian neural networks. Their strengths and limitations in relation to the preservation of the aforementioned axiomatic properties are discussed. In addition, arguments are presented in favor of a hybrid methodology that combines imputation algorithms with normative validation criteria to ensure ethical consistency and inferential robustness. The article concludes that the choice of imputation methods in social contexts should consider not only statistical efficiency but also structural compatibility with the theoretical model underlying the poverty measurement exercise. Future lines of research are proposed for the development of integrative frameworks that articulate imputation methods, respect for axioms, and use in public policy.