Department of Biostatistics Seminar/Workshop Series

On Borrowing Information over Time in Small Area Estimation with Application to Small Area Income and Poverty Estimates (SAIPE)

Dr. Carolina Franco

Research Mathematical Statistician, Center for Statistical Research and Methodology, US Census Bureau, Suitland, MD

Small area estimation typically seeks to improve direct survey estimates by borrowing information across areas or from covariate data. For repeated surveys, one can also consider borrowing information over time, i.e., from past survey estimates. Some questions arise: Under what circumstances does borrowing information over time yield significant benefits? How much past data should be incorporated into the model? Can past data be summarized for use in a model, say via an average of some number of previous survey estimates? We examine how the answers to these questions will depend on the underlying parameters of assumed models, and address them for application to the U.S Census Bureau’s Small Area Income and Poverty Estimates (SAIPE) Program. Our motivation is to provide estimates of rates of school-aged children in poverty for U.S. counties by modeling data from the Census Bureau’s American Community Survey (ACS). The primary data source for area-level SAIPE models is estimates based on yearly data from the ACS. As ACS also produces estimates based on five years of data-collection, it is natural to consider borrowing strength from past estimates by using several individual past year estimates or the most recent previous five year estimates. The former lends itself to treatment based on time series small area models, and the latter is a logical setting for application of bivariate models. These approaches are contrasted. Moreover, we characterize under what settings we may expect significant benefits from borrowing strength from past estimates through either of these approaches.