The regression approach to the measurement and decomposition of the multidimensional Watts poverty index

In this paper, we exploit the simplification associated with the regression approach for purposes of computing and decomposing the multidimensional Watts poverty index (MWPI) when the dimensions are expressed in terms of measurable achievements. It turns out that MWPI can be conveniently estimated by specifying an underlying system of Seemingly Unrelated Regressions linking the various poverty dimensions. The resulting decompositions turn out to be consistent with the 3 i’s of multidimensional poverty (i.e., incidence, intensity and inequality), augmented with poor population shares (in the unweighted dimensions case) and the product of poor population shares and poverty dimension weights (in the weighted dimensions case). Additional modifications to reflect the Alkire-Foster dual cut-off poor-group membership identification strategy are also discussed. An illustrative example using Canadian 2016 census data is provided together with a new proposal for choosing the best weights for the poverty dimensions among some competing alternatives.