We compare the classes of empirical Bayes and composite estimators of the population means of the districts (small areas) of a country and show that the commonly adopted modelling strategy of searching for a well-fitting random coefficient (two-level) model and using it for estimation of the district-level means can be ineffective. In particular, we show that variables with small district-level variation are not useful as covariates even when they are strong predictors of the target variable.
Claeskens G and Hjort NL (2003) The focused information criterion. Journal of the American Statistical Association, 98, 900–16.
2.
Ghosh M and Rao JNK (1994) Small area estimation. An appraisal. Statistical Science, 9, 55–93.
3.
Longford NT (1996) Model-based estimation in surveys with stratified clustered design. Australian Journal of Statistics, 38, 333–52.
4.
Longford NT (1999) Multivariate shrinkage estimation of small area means and proportions. Journal of the Royal Statistical Society Series A, 162, 291–302.
5.
Longford NT (2005) Missing Data and Small-Area Estimation. Modern Analytical Equipment of the Survey Statistician. New York: Springer-Verlag.
6.
Longford NT (2007) On standard errors of small-area estimators. Survey Methodology, 33, 69–79.
7.
Pfeffermann D, Skinner CJ, Holmes D, Goldstein H, et al. (1998) Weighting for unequal selection probabilities in multilevel models. Journal of the Royal Statistical Society Series B, 60, 23–40.
8.
Potthoff RF, Woodbury MA and Manton KG (1992) “Equivalent sample size” and “equivalent degrees of freedom” refinements for inference using survey weights under superpopulation models. Journal of the American Statistical Association, 87, 383–96.
9.
R Development Core Team (2006) R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing.
10.
Rao JNK (2003) Small Area Estimation. New York: Wiley.
