Online Program

A Primer on Model-Based Small Area Estimation with Applications To Establishment Surveys
*J.N.K. Rao, Distinguished Research Professor at Carelton University 


Most sample surveys are designed to provide reliable direct estimates of totals or means for the population as a whole and subpopulations or domains with large enough sample sizes. However, direct or domain-specific estimates for domains with small sample sizes (called small areas) do not lead to acceptable precision and yet demand for small area statistics has greatly increased. As a result, it is necessary to use indirect estimates that borrow strength across related small areas through implicit or explicit linking models based on auxiliary population information such as census and administrative data. In this talk, I will focus on explicit model-based methods for constructing small area estimates and associated mean squared error estimates. In particular, simple area level and unit level models with random small area effects will be used to explain estimation methods based on empirical best linear unbiased prediction, empirical Bayes and hierarchical Bayes methods. Applications to business surveys and other establishment surveys will also be presented.