Area-specific Data-driven Selection of Area Effects in the Basic Unit-level Small Area Estimation Model

I. Molina Peralta, A. Bikauskaite, D. Morales

The nested-error linear regression model, which adds random area effects to a linear regression, is used in small area estimation to "borrow strength''. When the covariates explain all the existing between-area heterogeneity, the random area effects may be removed from the model and the resulting synthetic regression model provides efficient small-area estimators. We consider the case in which area effects may be needed for some areas but not for all of them by introducing a mixture of the nested-error and the synthetic regression models. Two different specifications are considered for this mixture model. To fit these models, expectation-maximization algorithms are developed. Two alternative small-area predictors are proposed for general indicators. A parametric bootstrap method is used to estimate the mean squared errors of the proposed predictors. The properties of the new predictors and the bootstrap procedure are analyzed empirically and an application illustrates the results.

Keywords: Empirical best predictor, expectation-maximisation algorithm, nested error model, normal mixture model, parametric bootstrap.

Mixed Models
September 2, 2026  11:20 AM
Aula 30

W. F. Acero Ruge, I. Molina Peralta, D. Morales

X. Espunya Soler, J. A. Sanchez Espigares, L. Acosta Argueta