Cutoff point Based on Poverty likelihood Percentage for Nepal

Thanks Anil, good question! If you're looking to calculate the percentage of households living below your chosen poverty line, then no cutoff is needed. The right thing to do will be to take the average of the poverty likelihoods of each individual or household. For example, if you have three people with likelihoods of 0%, 10%, and 23%, then your best estimate would be 11%. If you *must* categorize *each individual* dichotomously as poor/non-poor, then you do need a cutoff. An intuitive choices might be to pick 50% (i.e. "probably" below the poverty line). A more empirically grounded approach could be to calculate the poverty rate as described above and then set the cutoff at that percentile of your distribution. E.g. if the average poverty likelihood for your sample is 23%, then the 23% of individuals with the highest likelihood get called "poor". That way your poverty rate estimates and your individual cutoff yield the same poverty rate. However, the choice of that cutoff is not really a technical decision; it is a programming one. The relevant question is what "targeting error" rate you are willing to accept. This presents a trade-off between inclusion errors and exclusion errors. With a high cutoff, you are less likely to erroneously call a non-poor person "poor". With a low cutoff, you are less likely to erroneously call a poor person "non-poor". How important each of these errors is to your program should determine the cutoff. A guide to targeting errors is included in the User Guide for the Nepalese PPI if you need more information on that approach. Happy to discuss further! -------------------- Elliott Collins PPI Director