Endogenous Weights and Multidimensional Poverty: A Cautionary Tale

Composite measures such as multidimensional poverty indices depend crucially on the weights assigned to the different dimensions and their indicators. A recent strand of the literature uses endogenous weights, determined by the data at hand, to compute poverty scores. Notwithstanding their merits, we demonstrate both analytically and empirically how a broad class of endogenous weights violates key properties of multidimensional poverty indices such as monotonicity and subgroup consistency. Without these properties, anti-poverty policy targeting and assessments are bound to be seriously compromised. Using real-life data from Ecuador and Uganda, we show that these violations are widespread. Hence, one should be extremely careful when using endogenous weights in measuring poverty. Our results naturally extend to other welfare measures based on binary indicators, such as the widely studied asset indices.