Professor Barbara Buttenfield in the Department of Geography at the University of Colorado Boulder and Nicholas Nagle in the Department of Geography at the University of Tennessee will develop methods that situate public use census microdata in space at finer resolutions than are currently available in the public domain. The methods they will use impute a set of sample weights (probabilities) that allocate microdata records to census tracts with the addition of ancillary data, even to sub-tract levels. These sample weights respect publicly available census tract statistics, as well as correlations between census data and ancillary landuse and land cover. Importantly, the methodology is derived using statistical methods of empirical likelihood, and allows for quantification of uncertainty and simulations of small area populations, thus offering a significant advance over existing methodologies. While the method depends entirely on publicly available data, validation will utilize confidential census records held at the Census Research Data Centers.
Deliverables from this research include methods, tools and usage guidelines that allow efficient construction of sample weights. Workshops will make these tools available to urban geographers, landuse planners, transportation specialists, demographers and educators. A website will disseminate the methods as Matlab, Python and R scripts, as a contribution to cyberinfrastructure, so that others may utilize tools. This website will be evaluated through a three-week short-course at the University of Colorado, by a planning workshop at Tennessee (an EPSCOR state) as well as by workshops at geography, urban planning and statistics national conventions. The project will provide specific recommendations to the US Census Bureau so that they may improve the relevance of public data for small area research, estimation and policy analysis.
