|
The American Community Survey (ACS) spends resources to determine if a nonresponding address is occupied or vacant. We develop a model for predicting ACS vacancy using survey paradata, information about the neighborhood, and administrative data. The goal is to use this vacancy information to make decisions regarding ACS operations. For example, field visits can be reduced for units modeled as vacant.
The quality of model predictions are analyzed retrospectively against outcomes from the field. We examine differences between predictions and field results. The purpose is to understand the characteristics of the occupied units misclassified as vacant. Another part of the analysis is to see how ACS estimates might change if occupied units are incorrectly classified as vacant.
|
