Small-Area Health Estimation

Borrowing strength means using information from outside an area's own small sample to improve its estimate. Each area's direct survey estimate is combined with a prediction from a regression on auxiliary variables that are known for all areas, so an area with too few respondents draws on the patterns seen across other areas and in covariates like age, poverty, and education. The weighting is reliability-driven: areas with precise direct estimates keep them, while areas with imprecise ones lean on the model. This is the core idea Fay and Herriot introduced and that Rao and Molina formalize.

Use the area-level Fay-Herriot model when you only have aggregated direct estimates per area along with their sampling variances and area-level covariates, which is common when microdata are confidential or unavailable. Use a unit-level model when you can link individual survey records to individual or area covariates, since modeling at the unit level can exploit more information and often yields more efficient estimates. Rao and Molina cover both; the choice is driven by the granularity of the data you can access rather than by a universal preference.

Both use area random effects, but the goals differ. Multilevel neighborhood-effects models aim to explain individual health and to separate contextual from compositional effects, focusing on coefficients and variance partitioning. Small-area estimation aims to predict an indicator for each specific area as accurately as possible, focusing on the predicted area values and their mean squared errors. The Fay-Herriot estimator is explicitly a predictor that shrinks unstable area estimates toward a model, whereas a neighborhood-effects analysis is primarily inferential about the determinants of individual outcomes.