Closing Data Gaps in Fertilizer Subsidy Analysis in Bangladesh : A Survey-to-Survey Imputation Approach

Bangladesh's fertilizer subsidy costs $2.5 billion annually and accounts for nearly two-thirds of agricultural spending, yet its distributional impact remains unknown due to data limitations. This impedes reform of a policy that may favor larger farmers while crowding out investment in public goods, the real engines for long-term productivity growth. The study develops a survey-to-survey imputation method to address this gap: the Household Income and Expenditure Survey 2022 records total fertilizer expenditure but not subsidized types, preventing accurate incidence analysis. The method combines cross-validated least absolute shrinkage and selection operator regression with randomized hot-deck matching to transfer type-specific fertilizer patterns from the Bangladesh Integrated Household Survey 2018-19 to the Household Income and Expenditure Survey 2022. The procedure predicts household urea shares using 42 harmonized predictors, then assigns complete fertilizer compositions through nearest-neighbor matching within welfare-by-agro-ecological strata. The method achieves strong predictive accuracy (test root mean square error = 0.169) and preserves distributional properties. Imputed shares replicate donor patterns closely: mean urea share is 51.5 percent versus 51.1 percent in the Bangladesh Integrated Household Survey, with overlapping confidence intervals across fertilizer types and regions. The enriched dataset provides the foundation for assessing whether subsidy benefits are concentrated among larger, wealthier farmers or distributed more equitably across farm households—a question that was previously unanswerable with existing data. More broadly, the study demonstrates a scalable framework for integrating complementary surveys in data-constrained settings.