Maximum Entropy Dasymetric Modeling for Demographic Small Area Estimation

摘要

This article describes a framework for maximum entropy dasymetric modeling based on spatial allocations of public use microdata sample (PUMS) files provided by the U.S. Census Bureau. The spatial units of the PUMS (PUMAs; public use microdata areas) are too large for fine-scale geographic analysis of populations because the common expectation is high degrees of variation within one PUMA (containing about 100,000 people). Limited demographic attribution is available at finer spatial resolutions in census summary tables for tracts and block groups. The described method (i.e., the coupling of spatial allocation procedures with dasymetric modeling) extends the literature and implements related variable associations and limiting variable constraints for allocating microdata household records to census tracts, based on sampling weights imputed using maximum entropy models. We present techniques to quantify household-level uncertainty and to show how this information is useful for guiding the dasymetric modeling and for improving the choice of limiting and related ancillary variables. We demonstrate our methods with a PUMA in Davidson County, Tennessee. Census summary statistics are used as related variables, and land cover-derived residential areas are included as limiting variables to refine the solution spatially to a subtract level.