Title: Small area estimation with partially linear mixed measurement error models
Authors: Mohammad Arashi - Ferdowsi University of Mashhad (Iran) [presenting]
Elahe Hosseini - Shahrood University of Technology (Iran)
Davood Shahsavani - Shahrood University of Technology (Iran)
Mohammad Reza Rabiei - Shahrood University of Technology (Iran)
Abstract: In small area estimation, the use of direct conventional methods will not lead to reliable estimates because the sample size is small compared to the population. Fay-Herriot model is commonly used in small area estimation in which borrowing strength from the related sites and other sources and uses auxiliary information to improve estimation. However, the assumption of normality is a limiting assumption for heavy-tailed data and outlying observations. Also, it is usually assumed that the predictors are measured without errors, which can be easily violated in small-area estimation. We provide a more flexible model beyond these limitations, which is more accurate than the existing models. Specifically, we study small area estimation in the partially linear mixed-effects model where measurement error is present for the predictors. We consider a large class of distributions for the error disturbances. Numerical studies are carried out to illustrate the superior performance of the proposed model in the prediction accuracy sense.