Poverty Assessment Tool Accuracy Submission for Nigeria

The full sample of 19,158 households is nationally representative, with 14,369 households used for tool construction and 4,789 households reserved for out-of-sample accuracy testing. The data were used to construct a poverty assessment tool that can predict household poverty status based on a set of selected indicators. The MAXR procedure in SAS was used to select the best poverty indicators from a pool of over 42 potential indicators. The procedure seeks to maximize explained variance by adding one variable at a time to the regression model and considering all combinations among pairs of regressors. The final set of indicators and their weights depended on selecting one of four statistical models: OLS, Quantile, Linear Probability, or Probit. The selection of the best model was based on the Balanced Poverty Accuracy Criterion (BPAC) and the Poverty Incidence Error (PIE), along with practicality considerations. The most accurate method for predicting household poverty in Nigeria, based on BPAC, is the 2-step Quantile regression. However, the 1-step Quantile regression is only slightly less accurate and requires only 15 indicators. The 1-step Quantile regression was selected as the best model, taking into consideration both accuracy and practicality. The selected model uses a set of 15 indicators to estimate household expenditures and predict poverty status. The decision rule for classifying households as very poor and not very-poor in Nigeria is based on the international poverty line of $1.25/day. Households whose predicted per capita daily expenditures fall below this line are classified as very poor, while those whose expenditures exceed this line are classified as not very-poor. The selected tool is based on a Quantile model, which estimates per capita consumption expenditures for each household. Households with estimated expenditures less than or equal to the $1.25/day poverty line are identified as very poor, while those with estimated expenditures exceeding this line are identified as not very-poor. The accuracy of the selected model was evaluated using in-sample accuracy results, which show that the model correctly identifies 58.2% of households as very poor and 64.2% of the population as very poor. The model also correctly identifies 45.1% of households as living below the national poverty line of 2,511 Nigerian Naira per adult equivalent per month. The twin errors possible in poverty assessment, misclassifying very poor households as not very-poor and misclassifying not very-poor households as very poor, are also evaluated in the accuracy results.