Disaggregating Imputed Poverty Estimates by Population Groups: New Evidence from a Multi-country Analysis

Can imputed poverty estimates be reliably disaggregated by population groups, especially when the interest is monitoring poverty levels for smaller, vulnerable groups that may not be represented as well in large-scale household surveys? The study tackles this question through a comprehensive literature review and empirical analysis that leverages 18 household surveys across four different low- and middle-income countries. The results suggest that the imputation accuracy widely varies by population group, with differences being as high as 10 percentage points in pairwise comparisons of groups. The imputation accuracy for the population groups of interest increases, on average, by 1.3 percentage points in response to increasing the sample size by 1,000 observations for the target survey that is used for sourcing the predictors for the imputation model. The results are robust to extensive sensitivity analyses and also suggest that incorporating geospatial predictors into the imputation model can help increase imputation accuracy. The discussion provides useful inputs for future survey design.