The authors use an empirical Bayes (EB) approach to small area estimation under area-level unmatched sampling and linking models. Model parameters are estimated by a unified expectation and maximization (EM) algorithm and used to obtain EB estimators of area parameters. Results are extended to a nonparametric linking model based on a spline approximation. Approximate EB estimators that are computationally simpler are also obtained. Different bootstrap approaches to estimating the mean squared error (MSE) of the EB estimators are proposed. Results of a simulation study on the performance of the proposed methods are presented. Proposed methods are applied to data from a survey of family income and expenditure in Japan and poverty rates in Spanish provinces.

Additional Metadata
Keywords Bootstrap, Empirical Bayes, Expectation–maximization algorithm, Fay–Herriot model, Mean squared error, Penalized spline
Persistent URL dx.doi.org/10.1007/s11749-017-0551-5
Journal Test
Citation
Sugasawa, S. (Shonosuke), Kubokawa, T. (Tatsuya), & Rao, J.N.K. (2017). Small area estimation via unmatched sampling and linking models. Test, 1–21. doi:10.1007/s11749-017-0551-5