Empirical best prediction under a nested error model with log transformation (Q104740)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Empirical best prediction under a nested error model with log transformation |
scientific article |
Statements
46
0 references
5
0 references
1 October 2018
0 references
24 October 2018
0 references
Empirical best prediction under a nested error model with log transformation (English)
0 references
A finite population of size \(N\) is divided into \(D\) small domains (areas). Let \(W_{di}\) be the value of the study variate for a unit \(i\) within the area. Consider the problem of predicting the area mean \(\sum^{N_d}_{i=1}W_{di}/N_d\), where \(N_d\) is the size of the small area, or predicting the specific value \(W_{di}\) for an out-of-sample unit \(i\) within the small area. This is studied by assuming a nested error regression model with \(\log W_{di}\) as the dependent variable, since for many economic variables log transformation results in normality and variance stabilization. However, the predictors obtained by transforming back the individual predicted values may be biased. In this paper, the authors derive optimum predictors under the model assumed. The best predictors depend on the vectors of regression parameter \(\beta\) and variance components \(\Theta\) of the model which are estimated in two stages and thus Empirical Best (EB) predictors are obtained. Next, the exact Mean Squared Errors (MSE) of the best predictors and second-order approximations to the MSE as well as their estimates are derived. A simulation experiment is carried out to compare the efficiencies of the estimators studied. Furthermore, the results are illustrated by predicting the incomes of individuals and of average income in smaller domains of municipalities from Mexico. In an appendix the mathematical lemmas needed and their proofs are given. A reference to supplementary material relating to further details and additional results is mentioned at the end of the paper.
0 references
empirical best estimator
0 references
mean squared error
0 references
parametric bootstrap
0 references
best prediction
0 references
nested error
0 references
regression model
0 references