With increasing demand for reliable estimates for small areas, small area estimation (SAE) methods are exposed to growing popularity in survey sampling. In this method, one can take the advantage of area-level models if area-level summary of ancillary variables exists. A fundamental area-level model was first suggested by Fay and Herriot in which we need to estimate the sampling variance (SV). Several methods have been suggested for smoothing of SV so far. Therefore, evaluation studies are required. This research examines four techniques for SV estimates including direct, probability distribution (PD), bootstrap and a new perspective of the Bayesian techniques. We modelled household food expenditure data for Iran in 2013 and taking socio-economic data as ancillary variables. We also conducted a simulation study in order to compare the effects of four variance estimation methods on precision of small area estimates. Although direct method had best fitting on real data, simulation study revealed that highest precision was achieved when PD method was used.

small area estimation, household food expenditure, sampling variance.