|
Abstract:
|
The U.S. Census Bureau conducts count imputation to estimate the population of housing units that do not provide such information. In 2000, a hot-deck methodology was used to do count imputation. Since then, the Bureau has tested several alternatives to the traditional hot deck imputation method, including variations on the hot-deck, spatial models, direct assignment from administrative records, and a model that uses data from both administrative records and the census. All methods were tested by running them on a truth deck. Several statistics were computed from these imputation results. Some of these statistics measured accuracy at an aggregate level, while others measured accuracy for individual housing units. This paper examines the results and interpretation of these various statistics.
|
