A Bayesian disaggregation model for mapping household food
insecurity in metropolitan areas of Northeast Brazil
Marcus L. Nascimento (FGV EMAp)
The growing demand for disaggregated statistics, driven by the Sustainable Development Goals,
has highlighted the need for methods that produce reliable small-area estimates. This challenge
is particularly relevant to food insecurity indicators in Brazil, where survey-based data are often
unavailable at the geographic resolution needed to identify vulnerable populations and support tar
geted public policies. This article proposes a Bayesian disaggregation model for ordinal survey data
that combines aggregated and disaggregated information through a hierarchical framework. Sur
vey weights are incorporated via a pseudo-likelihood approach, and a double data augmentation
strategy enables efficient posterior computation. The methodology is motivated by data from the
2023 Continuous National Household Sample Survey, in which the Brazilian Food Insecurity Scale
is observed at different geographic resolutions: detailed information is available only for selected
municipalities, while aggregated information is available for the remaining areas.
Joint work with Rebecca O. Souza (FCM– UERJ) and Alexandra M. Schmidt (EBOH– McGill Uni
versity).