Imputing poverty indicators without consumption data: an exploratory analysis

Accurate poverty measurement relies on household consumption data, but such data are often inadequate, outdated, or display inconsistencies over time in poorer countries. To address these data challenges, we employ survey-to-survey imputation to produce estimates for several poverty indicators including headcount poverty, extreme poverty, poverty gap, near-poverty rates, as well as mean consumption levels and the entire consumption distribution. Analysing 22 multi-topic household surveys conducted over the past decade in Bangladesh, Ethiopia, Malawi, Nigeria, Tanzania, and Vietnam, we find encouraging results. Adding either household utility expenditures or food expenditures to basic imputation models with household-level demographic, employment, and asset variables could improve the probability of imputation accuracy between 0.1 and 0.4. Adding predictors from geospatial data could further increase imputation accuracy. The analysis also shows that a larger time interval between surveys is associated with a lower probability of predicting some poverty indicators, and that a better imputation model goodness-of-fit (R2) does not necessarily help. The results offer cost-saving inputs into future survey design.