|
Abstract:
|
The Small Area Income and Poverty Estimates (SAIPE) program produces estimated poverty counts for every county in United States. Currently, SAIPE uses a Fay-Herriot model to combine American Community Survey (ACS) direct estimates with administrative records through a regression model. The Fay-Herriot model is an area-level model in that it models direct estimates of the quantity of interest using aggregated covariates. This paper examines an alternative approach of modeling observed survey responses using auxiliary information for individuals. We are interested in estimating poverty counts using a multilevel logistic mixed regression model with post-stratification (MRP). This approach requires independence among individuals and does not account for survey design. In some applications, these problems have been remedied by estimating area-level design effects to adjust sample sizes and rescale survey weights. However, for very small demographic groups, sample sizes can be too small to reliably estimate design effects. This paper explores alternative methods for incorporating survey design and approximately meeting model assumptions.
|
