Full Citation

Title: Improving TIGER, Hidden Costs: The Uncertain Correspondence of 1990 and 2010 U.S. Census Geography

Citation Type: Miscellaneous

Publication Year: 2016

ISBN:

ISSN:

DOI:

NSFID:

PMCID:

PMID:

Abstract: The U.S. Census Bureau supplies GIS-compatible definitions of census geographic units via its TIGER (Topologically Integrated Geographic Encoding and Referencing) data product series. Between the 2000 and 2010 censuses, the U.S. Census Bureau completed major improvements to their MAF/TIGER geographic database, from which all TIGER products are derived. The 2010 TIGER products, which supply boundaries for both 2000 and 2010 census units, are therefore significantly more accurate than 2000 TIGER data, which supply boundaries for both 1990 and 2000 census units. The accuracy improvements should be highly beneficial for spatial analyses of recent census data, but for spatio-temporal analyses that span the 19902010 period (or longer), the improvements impose a cost: in many cases, it is impossible to determine exactly which 1990 units correspond to which 2010 units. Boundaries that are in fact coincident may have representations that are not coincident in the separate TIGER versions, and the representational discrepancies are sometimes very large. The National Historical Geographic Information System (NHGIS https://nhgis.org) has recently begun releasing geographically standardized time series, which provide U.S. census data from multiple times for a single censuss geographic units. To allocate one censuss data to another censuss geographic units, NHGIS interpolates data from the smallest source units for which the data are available. The first release, in 2015, supplied 2000 data for 2010 census units by interpolating data from 2000 census blocks. The next release will supply 1990 data for 2010 census units, again by interpolating block data, but in this setting, because of the improvements in TIGER data, the interpolation is complicated by the uncertain correspondence between 1990 blocks and 2010 census units. Fortunately, TIGER data do make it possible to determine correspondences between 1990 and 2000 units, and between 2000 and 2010 units, and from these crosswalks, we can impose certain constraints on possible 1990-2010 unit relationships. Still, not all relationships can be exactly determined. In this paper, I posit three general alternatives for implementing areal interpolation in this setting: simply overlaying 1990 and 2010 boundaries without regard to representational discrepancies; using 2000 units as a bridge between 1990 and 2010 units; or a combined approach, overlaying 1990 and 2010 boundaries, but also using known topological relationships with 2000 units to constrain and refine the interpolation. In order to assess potential relative advantages of these approacheswithout yet implementing themI present here an assessment of how much uncertainty there is in block-based 1990 population estimates for 2010 units, identifying in particular how much uncertainty may be added by inexact correspondence information (i.e., the hidden costs of improved TIGER data).

Url: http://geo.gmu.edu/AutoCarto2016/Schroeder.pdf

User Submitted?: No

Authors: Schroeder, Jonathan P

Publisher: University of Minnesota

Data Collections: IPUMS NHGIS

Topics: Land Use/Urban Organization, Methodology and Data Collection, Other

Countries: