D. Morales González, M. D. Esteban Lefler

We consider models for normal random variables at the area level, by considering different types of temporal correlations.
The temporal correlation is introduced by means of time-specific random effects whose distributions might be AR(1), MA(1) or ARMA(1,1). We propose algorithms to fit the models and estimate its parameters; for example, by using the Fisher-scoring algorithm to obtain residual maximum likelihood (REML) estimates. We obtain EBLUPs of linear parameters. We obtain explicit estimators of mean squared error studying the order of efficiency. We introduce bootstrap resampling methods to estimate mean squared errors of EBLUPs. We study the consistency of the proposed methods. Finally, we carry out simulation experiment to study the behavior of the bootstrap procedures under controlled situations.

Palabras clave: small area estimation, linear mixed models, bootstrap

XC1b Pósters (Estadística Pública)
18 de abril de 2012  12:00
Salón Madrid

T. Iglesias Cabo, P. Díaz Díaz, E. Torres Manzanera

J. A. Marmolejo Martín, M. D. Huete Morales, M. Á. Pérez Castro

M. I. Sánchez-Rodríguez, Á. Ruiz Gándara, M. Tocón, A. Sanchís Vidal