Abstract

Many National Statistics Offices (NSO) has been widely recognized the use of small area estimation (SAE) in producing official statistics including BPS-Statistics Indonesia. Multivariate Fay-Herriot (MFH) model in SAE is the extension of a Fay-Herriot model which leads to more efficient estimators of small area means and takes advantage of the correlations between the variables of interest, unlike the univariate model. For official statistics usage, a set of small area estimates is expected to be more more efficient than unbiased direct estimates and aggregately consistent to the aggregation of direct estimation in large area that cannot be guaranteed by the multivariate Fay-Herriot model. Difference benchmarking method for multivariate Fay-Herriot models to estimate small-area indicators are introduced by combining Multivariate Fay-Herriot Model with difference benchmarking. An approximation to the matrix of mean square error (MSE) is given by adjusting the estimation of difference benchmarking MSE in the univariate form proposed by Prasad and Rao. Simulation experiments are performed to assess behaviour of the difference benchmarking method for the multivariate Fay-Herriot model and for comparing the MSE. The result shows that multivariate benchmarking can produce a more efficient MSE than univariate benchmarking.