Guoyi Zhang

Title: A semi-parametric model for small area estimation using support vector machine

Abstract:

The Fay-Herriot model (Fay & Herriot, 1979) is popularly used in small area estimation (SAE). Fay-Herriot model assumes a linear relationship between the response and predictor variables, which in practice may not be appropriate. In this research, we extend the Fay-Herriot model to a more general semi-parametric model for small area estimation (SP-SAE), in which the nonparametric component is estimated by machine learning tools such as support vector machine. A back-fitting algorithm is proposed to solve the SP-SAE problem. An example using the American Community Survey (ACS) data shows that the proposed SP-SAE significantly improved the health insurance coverage estimators than the Fay-Herriot model.