524 – Recent Advances and Applications of Spatial and Spatio-Temporal Models for Official Statistics
Predicting Coverage Error on the Master Address File Using Spatial Modeling Methods at the Block Level
Krista Heim
U.S. Census Bureau
Andrew M. Raim
U.S. Census Bureau
This paper explores methods of spatial modeling to identify opportunities for reduced fieldwork in census Address Canvassing operations. The purpose of Address Canvassing is to improve the coverage and quality of the Census Bureau's address list, the Master Address File (MAF), prior to census enumeration. Various modeling techniques such as zero-inflated negative binomial regression have been explored in the past to predict areas with many coverage errors on the MAF and identify blocks which would likely contain change (and those which would not). Such information could inform a reduction to the in-field canvassing workload and reduced field costs. We use a recently develope spatial mixed effects model with dimension reduction, and take New York County as an example. It is seen that accounting for spatial dependence has a large effect on the estimated coefficients, including which predictors are significant. The impact to predicted values is more subtle, with th spatial model producing slightly more accurate predictions.