|
Abstract:
|
The U.S. Census Bureau maintains the Master Address File (MAF) as a basis for the decennial census and household surveys it conducts throughout the decade. To prepare for the 2010 Census, the Census Bureau organized a full-scale address canvassing operation where field representatives walked most of the United States to find and correct errors on the MAF. The Census Bureau is now developing strategies to avoid the high cost of a full in-field canvassing for the 2020 Census and to reduce errors throughout the decade. One idea has been to use statistical models to study and predict coverage errors found on the MAF. Such models could potentially help to inform address listing fieldwork and other less costly alternatives being considered by the Census Bureau. Previous MAF modeling work has simply ordered point estimates to suggest regions which might contain significant coverage error and require further attention. In this work, we consider the problem in a decision theoretic framework. By choosing an appropriate loss function, we can study consequences of canvassing decisions under a selected model with data from past operations.
|
