Expedient Small Area Estimation via Proportional Odds Ratio Log-linear Models
Li, L., & Levy, P. S. (2007, August). Expedient Small Area Estimation via Proportional Odds Ratio Log-linear Models. Presented at ASA JSM 2007, Salt Lake City, UT.
The synthetic methods for small area estimation are appealing due to simplicity in implementation and are suitable for categorical outcomes in situations such as public health surveillance based on sample survey data where small area estimates must be produced in a timely manner. However, they do not readily accommodate the local area characteristics. The extension of the synthetic methods to the structure-preserving estimation (SPREE) methods by Purcell and Kish opened up opportunities for incorporating local area characteristics and making use of Bayesian inference. We develop estimation procedures for SPREE via proportional odds ratio log-linear models and illustrate with data from the Behavioral Risk Factor Surveillance System and Census 2000 on employment disability in North Carolina. Our estimates always agree with direct survey estimates when aggregated to a level of larger areas at which direct estimates are deemed reliable, which is often a desirable feature for small area estimates.