TIGER Positional Accuracy & Geocoding: 2 Articles

The following two recent articles focusing on positional accuracy and geocoding are very intriguing. Learned quite a bit about the science behind positional accuracy and geocoding. I usually advise our students to use our local parcel boundary feature class for geocoding (as I discuss here), but I'll focus first on these articles.

• A Snake-based Approach for TIGER Road Data Conflation, Song, Wenbo; Haithcoat, Timothy L.; Keller, James M. Cartography and Geographic Information Science, Volume 33, Number 4, October 2006, pp. 287-298(12) [abstract]


• Modeling the probability distribution of positional errors incurred by residential address geocoding, Dale L Zimmerman, Xiangming Fang, Soumya Mazumdar and Gerard Rushton, International Journal of Health Geographics 2007, 6:1 [full-text]

The first article describes a new conflation method for improving the positional accuracy of TIGER files. As described by this article, conflation is the combining of attribute-rich TIGER data with positionally-superior local data. Traditional conflation methods include (1) feature matching, (2) map alignment (rubber-sheeting), and (3) attribute transfers. The article describes a new conflation method, based on snakes. Snakes [background information] are active (dynamic) contour models based on image or 3D data. See Mark Schulze's nifty java tool to learn more about snakes. Traditional conflation are between TIGER and other vector files. This new method extends conflation to interact between TIGER and snakes based on raster orthophotos. This article was a fantastic insight into the creation of the improved vector street files we enjoy here at the university as a result of the hard work of local government analysts (and their contractors).

The second article presents the results of comparing three geocoding methods: (1) Automated Geocodes, (2) E911 Geocodes, (3) Use of an Orthophoto. The automated geocoding was accomplished using TIGER data and ArcGIS 9.1. The E911 geocoding used the local 911 listing. The orthophoto method overlaid parcel boundaries over an orthophoto to enhance the E911 geocode. As expected, the orthophoto method was the most accurate, and was termed the "gold standard". The real beauty of this article is the analysis of the error measurements and their attempts to model the errors.

So, how does this impact how I do my job? Reckon the best is to continue to advise students to geocode based on our local government-produced parcel shapefile or cadastral. If I was not fortunate to work in the GIS-rich DFW Metroplex, and so did not have access to these high quality files? Guess then I would try to work with local faculty and governments to work on conflation techniques to improve the accuracy of TIGER data.

FYI, here is a good resource to scan the recent table of contents of the major GIS-related journals.