Author: Bac Tran (US Census Bureau)
The presence of outliers in the small area estimation (SAE) raises concerns in the design based part of population parameters prediction due to the violation of model-based assumptions. Various techniques have been introduced to mitigate the effect of outliers in the unit-level and area-level models in the small area estimation in the literature. This paper applied an idea of a robust technique based on a scale mixture of two normal distributions to deal with those outliers and estimate the number of employees in the Annual Survey of Public Employment & Payroll (ASPEP) data using statistical methods from Gershunskaya & Lahiri (2011) but using the hierarchical Bayes approach. We then compared the research method to one existing methods. The two Employments census data 2007 and 2012 were used for the evaluation of this research.