Identification of Poor: Effectiveness of Proxy Mean Tests

Abstract

The success of social protection programs depends upon accurate targeting of poor. Identification of correct poor has gained a considerable importance in recent years because it ensures effectiveness of social safety net programs in terms of poverty eradication goals of sustainable development. A plethora of studies exists in the literature for the measurement of poverty while employing proxy mean testing methodology and identifying the deserving poor. However, the selection of poor households depends purely upon the proxy mean test adopted by the social protection program as there PMT formula. These proxy means tests include both regression and correlation based techniques for the selection of proxy indicators that are best correlates of household poverty status. Therefore, the accuracy of PMT is crucial for correct targeting of poor. This study is therefore an attempt to evaluate the performance of four widely used PMTs for identification of poor that are linear regression, logistic regression, principal component analysis (PCA) and fuzzy regression. The accuracy of these PMTs is computed at different deprivation cutoffs to compute accuracy ratios and targeting errors i.e. Total accuracy, balanced poverty accuracy criteria (BPAC), under-coverage and leakage. The accuracy ratios and targeting errors are computed by using expenditure based official poverty line of Rs.3030 per adult equivalent as benchmark criteria. The findings of this study reveal that fuzzy regression turns out to be the best targeting tool while considering OPL as a benchmark, followed by logistic regression and principal component analysis. The linear regression is the last good option for identification of poor.