The Information Basis of Multivariate Poverty Assessments

Measures of multivariate well-being, such as poverty or inequality, are scalar functions of matrices of several attributes, m, associated with a number of individual or households, N. This entails inevitable "aggregation" and summarization over individuals as well as attributes. There is no escape from this. Such aggregation, in turn, implies a set of weights attached to each individual, and some normative decision on how they relate. The aggregation over the attributes also forces decisions about the weight to be given to each attribute and the relation between the attributes as, perhaps, substitutes or complements. We argue in favor of information theory aggregation methods which are explicit about such normative choices, and help place other methods in this realistic context. According to axiomatically well developed measures of divergence in information theory, our measures are "ideal" and other methods are therefore sub-optimal. The advocacy of the latter must be accompanied by well argued positions in support of special properties and other considerations which may be compelling in a given context or application.