Choosing among rival poverty rates : some tests for Latin America

Poverty rates are now widely available, but are they reliable? Wide variations in estimated poverty rates for the same poverty line, year and country reflect an underlying reality: there is no widely accepted procedure for estimating national poverty rates. This paper proposes a simple, ex post procedure for selecting poverty rates that have certain desirable properties. Absolute poverty measures, estimated uniformly across countries, should be correlated with nonmonetary indicators that reflect the consequences of physical deprivation (e.g., malnutrition, birth rates, school attendance). A series of non-nested hypotheses tests are used to choose among competing poverty and income measures. This method is applied to screen the 66 alternate poverty measures computed by Szkeley, Lustig et al. (2000) for 17 Latin countries. These tests identify 10-15 poverty measures that meet the standards set forth for useful poverty measures. This final group of poverty measures is then ranked using various performance criteria.