Poverty evaluations differ from welfare evaluations in one significant aspect, the existence of a threshold or reference point, the poverty line. It is therefore possible to build up normative evaluation models in which comparisons are made taking distances from this reference point and not only from the origin to be ethically relevant. This is the case in our model of poverty comparisons over heterogeneous populations, which focuses upon poverty gaps and not incomes. When poverty lines differ for the different groups in the population we show that choosing poverty gaps instead of incomes as the relevant indicator brings in normatively appealing classes of poverty indices not previously accommodated. For these indices poverty comparisons over heterogeneous populations are implemented through sequential poverty gap curves (or poverty gap distributions) dominance. These novel conditions are logically related to those suggested in Atkinson and Bourguignon (1987) for welfare comparisons, and can also be grounded firmly upon those of Bourguignon (1989). The proportion of poor individuals in the society or their average poverty gap play a role in our comparisons that was neglected in the existing poverty dominance criteria for heterogeneous populations. Various intermediate poverty dominance conditions and a generalization of the poverty gap approach are also investigated.
