Using Survey-to-Survey Imputation to Fill Poverty Data Gaps at a Low Cost: Evidence from a Randomized Survey Experiment

Abstract Household consumption surveys are often unavailable or incomparable over time in many developing countries. Analyzing a unique randomized survey design experiment from Tanzania, this article offers new and rigorous evidence that survey-to-survey imputation can fill data gaps and provide low-cost and reliable poverty estimates. The study finds that basic imputation models featuring a modest set of predictors on demographics and employment yield accurate predictions, as do models that further add household assets and housing and utility expenditures. Imputation accuracy is robust to survey questionnaire lengths, alternative base surveys, different poverty lines, and frequency of CPI deflators, and is better than alternative estimation methods. Yet, employing shortened or aggregated food and nonfood consumption predictor variables requires standardizing these predictors against those in the base survey. For best-performing models to reach acceptable accuracy levels, the minimum-required sample size should be around 1,000 for both base and target surveys. The findings from this study have relevance for future survey design.