Abstract:
|
The American Community Survey (ACS) publishes county-level estimates for hundreds of demographic and economic variables, with the statistical uncertainty only being quantified from the standpoint of sampling error. We propose a nonstationary spatial model (the lattice-kriging model) for 'flow' estimates (i.e., representing aggregate activity over a region) that accounts for the sampling error while also utilizing the latent structure of the population process. Our goals are two-fold: first, to offer estimates of all counties (including 'missing' counties, for which the ACS omits to publish estimates due to the sampling variability being too high) along with a variance estimate that combines both the population dynamics and the sampling mechanism; second, to offer estimates of custom regions that correspond to the needs of users at the state and local level. These custom regions can overlap county boundaries, and can have nontrivial homotopy (e.g., be discus-shaped).
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.