Abstract:
|
Most small area models can broadly be classified as either a unit-level model or an area-level model. Both models have seen a number of practical applications. In a 2016 paper, Hidiroglou and You examined the performance of some unit- and area-level SAE procedures for samples from simulated populations, finding that unit-level estimators had a distinct advantage over area-level ones. A subsequent paper (Fay, 2018) showed that the Hidiroglou-You result was less clear over a wider choice of simulated hypothetical populations, particularly when a survey regression estimator was combined with the area-level model. This paper sharpens these results using a combination of simulation studies and asymptotic approximations to the mean-square errors to provide guidance on when one approach is more advantageous than the other.
|