Small Area Estimation (SAE) models commonly assume Normal distribution or, more generally, exponential family. We propose a SAE unit-level model based on Generalized Additive Models for Location, Scale and Shape (GAMLSS). GAMLSS completely release the exponential family distributional assumption and allow each parameter to depend on covariates. Besides, a bootstrap approach to estimate MSE is proposed. The performance of the estimators is evaluated with model- and design-based simulations. Results show that the proposed predictor works better than the well-known EBLUP. The SAE model based on GAMLSS is used to estimate the per-capita expenditure in small areas, based on the Italian data.
翻译:小型地区估计模型通常采用正常分布模式,或更一般地采用指数式家庭模式。我们根据位置、规模和形状通用Additive模型(GAMLSS)提出SAE单位级模型。GAMLSS完全释放指数式家庭分布假设,并允许每个参数取决于共变。此外,还提议了估算MSE的陷阱式方法。根据模型和设计模拟对估测器的性能进行评估。结果显示,拟议的预测器比众所周知的EBLUP运行得更好。基于GAMLSS的SAE模型用于根据意大利数据估算小地区的人均支出。