25 January 2021 – This paper authored by Indranil Dutta, Ricardo Nogales and Gaston Yalonetzky explores the dangers with endogenous (i.e. data-driven) weights in measuring poverty.
Multidimensional Poverty Indices (MPIs) depend on the weights assigned to the different dimensions and their indicators. A common approach is to use exogenous weights, which are independent of the dataset and reflect some reasoned value judgements of the society, the analyst or the policymaker. In contrast, this paper focuses on a growing body of literature which relies on the alternative of endogenous (data-driven) weights, which are determined entirely by the dataset, as a way to derive the importance of the different indicators in the composite measure of deprivation.
This paper demonstrates how a broad and popular class of endogenous weights violates key properties of MPIs. These properties are monotonicity, where the overall poverty experience of a society cannot improve if the poverty experience of any of its members worsens in any indicator. Endogenous weights can also violate subgroup consistency, which is where changes in the overall poverty of a society should coherently reflect changes in poverty in a group of its members. The paper provides analytical proofs for the existence of these violations, and explains why and when they can occur. Using real-life data from Ecuador and Uganda, it also shows that they can be ubiquitous in practice.
The OPHI Working Paper series covers research into innovations in multidimensional poverty measurement and makes new academic research by OPHI researchers publicly available prior to their publication.