Anti-poverty Policy: Screening for Eligibility Using Village-level Evidence

In the context of targeting of state transfers based on income poverty lines, this study is concerned with the identification of households that may have been wrongly included in the target group. To this end, we investigate the relationship between self-declared private income and some 478 household variables obtained in a village level survey. We use class probability tree analysis which is a non-parametric multivariate method. Relationships are expressed as easily interpretable rules that give combinations of the important features that characterise the 'poor' households (income declared below the income poverty line) and the 'non-poor' (income declared above the income poverty line), rather than as mathematical equations as in previous regression based analyses. Approximately 20% of the households that declared income so as to be classified 'poor' were found to have feature combinations which were similar to those characterising 'non-poor' households. These cases would thus be worthy of further investigation for distortion of income, before being considered eligible for any transfers.