Targeting the poor using ROC curves

Abstract

This paper compares the performance of targeting indicators to identify the poor. If the ROC curve of one indicator lies above that of another, the first indicator dominates the second for all social welfare functions based on the two types of errors involved in targeting. The method is applied to Bangladesh. Fifteen indicators are used, including location, land ownership, education, occupation, demographics, age, family structure, and housing. The analysis is applied at the national, urban, and rural levels with two poverty lines. Education dominates land ownership in urban areas. The ranking is reversed in most cases in rural areas. (C) 1997 Elsevier Science Ltd. All rights reserved.