Small area estimation with multiple covariates measured with errors: A nested error linear regression approach of combining multiple surveys

JOURNAL OF MULTIVARIATE ANALYSIS

卷:167

Article

Datta, Gauri S.^1,2  Torabi, Mahmoud^3,4  Rao, J. N. K.⁵  Liu, Benmei⁶ 

[1] Univ Georgia, Dept Stat, Athens, GA 30602 USA

[2] US Bur Census, Washington, DC 20233 USA

[3] Univ Manitoba, Dept Community Hlth Sci, Winnipeg, MB R3E 0W3, Canada

[4] Univ Manitoba, Dept Stat, Winnipeg, MB R3E 0W3, Canada

[5] Carleton Univ, Sch Math & Stat, Ottawa, ON K1S 5B6, Canada

[6] NCI, Bethesda, MD 20892 USA

关键词: Conditional distribution;    Jackknife;    Linear mixed model;    Mean squared prediction error;    Measurement error;   

DOI  :  10.1016/j.jmva.2018.04.001

来源: Elsevier

【 摘 要 】

Small area estimation has become a very active area of research in statistics. Many models studied in small area estimation focus on one or more variables of interest from a single survey without paying close attention to the nature of the covariates. It is useful to utilize the idea of borrowing strength from covariates to build a model which combines two (or multiple) surveys. In many real applications, there are also covariates measured with errors. In this paper, we study a nested error linear regression model which has multiple unit- or area-level error-free covariates, possibly coming from administrative records, and multiple area-level covariates subject to structural measurement error where the data on the latter covariates are obtained from multiple surveys. In particular, we derive empirical best predictors of small area means and estimators of mean squared prediction error of the empirical best predictors of small area means. Performance of the proposed approach is studied through a simulation study and also by a real application. (C) 2018 Elsevier Inc. All rights reserved.

【 授权许可】

Free