|
Abstract:
|
After the completion of all Census data collection operations a small number of households remain for which either the count of the number of persons resident, or the status as occupied, vacant or non-existent is not known. In order to produce an accurate count of persons in the U.S., it is necessary to impute this missing data. An investigation of a Spatial Modeling approach suggested by Thibaudeau (2002) for the imputation of missing household size, occupancy or housing unit status was undertaken using 2000 Census household data. The paper will describe aspects of the selection of model variables, the model selection process, and implementation of the models. In addition, the results from this approach for count imputation are compared to the results from a traditional Nearest Neighbor Hot Deck method relative to a Truth Deck of simulated missing household sizes.
|
