Observed best selective prediction in small area estimation

Shonosuke Sugasawa, Yuki Kawakubo, Gauri Sankar Datta

Research output: Contribution to journalArticlepeer-review


In small area estimation methodology, selection of the suitable covariates and estimation in the selected model are usually considered separately. In this paper, we consider variable selection and estimation simultaneously to minimize the total mean squared prediction errors (MSPE)for estimation of small area means. The derived method, which we call observed best selective prediction (OBSP), can be regarded as an extension of the observed best prediction (OBP)method by Jiang et al. (2011). When the true model is included in the largest model, the resulting OBSP estimator is consistent. Based on the asymptotic result, we derive an estimator of MSPE by applying the parametric bootstrap method. Through simulation experiments, we investigate the finite-sample performance of OBSP together with OBP in which the variable selection is carried out by using AIC and BIC, and OBP using all the covariates. As an example, we applied OBSP to Japanese survey data.

Original languageEnglish
Pages (from-to)383-392
Number of pages10
JournalJournal of Multivariate Analysis
Publication statusPublished - 2019 Sept
Externally publishedYes


  • Bootstrap
  • Fay–Herriot model
  • Mean squared prediction error
  • Model selection
  • Unbiased risk estimate

ASJC Scopus subject areas

  • Statistics and Probability
  • Numerical Analysis
  • Statistics, Probability and Uncertainty


Dive into the research topics of 'Observed best selective prediction in small area estimation'. Together they form a unique fingerprint.

Cite this