Fay-Herriot Small Area Estimation in the Survey of Business Owners

Abstract

This paper will study the use of the Fay-Herriot small area estimation model on the Survey of Business Owners and Self-Employed Persons (SBO) 2002 data. Small area estimation continues to be an important topic as the demand for reliable small area statistics continues to grow. Because direct estimates can yield large standard errors due to small sample sizes, the need for small areas to borrow strength from related areas is present. The 2002 SBO has a state level design, which may contain several counties per state with small sample sizes. This paper investigates the plausibility and usability of the Fay-Herriot estimators at the county level for predicting Black ownership of businesses. These mixed-effect model predictors will be compared to linear fixed-effect models. This research will also investigate the parsimony of the two types of estimators.