This paper suggests a low cost methodology for identifying and assessing the accuracy of poverty monitoring and targeting indicators. It proposes that the non-parametric technique of Receiver Operating Characteristic (ROC) curves be used to assess the accuracy of poverty indicators.
Furthermore, it suggests that the best individual indicators can be combined into a composite poverty indicator using a stepwise Probit approach. The composite poverty monitoring indicators so developed are intuitive, parsimonious (involving just six to nine indicators in the examples from Vietnam), and relatively easy to collect data on. A further advantage of the method is that it allows the trade-off between coverage of the poor and exclusion of the non-poor to be quantified in terms that are readily understandable by policy-makers.
This methodology (suitably adapted and expanded) could help bridge the gap between the aggregate poverty statistics generated by periodic household surveys and the need for more disaggregated and regular poverty statistics by welfare agencies.
