Spatial methods in areal administrative data analysis

Abstract

Administrative data often arise as electronic copies of paid bills generated from insurance companies includ-
ing the Medicare and Medicaid programs. Such data are widely seen and analyzed in the public health area,
as in investigations of cancer control, health service accessibility, and spatial epidemiology. In areas like
political science and education, administrative data are also important. Administrative data are sometimes
more readily available as summaries over each administrative unit (county, zip code, etc.) in a particular set
determined by geopolitical boundaries, or what statisticians refer to as areal data. However, the spatial
dependence often present in administrative data is often ignored by health services researchers. This can
lead to problems in estimating the true underlying spatial surface, including inefficient use of data and
biased conclusions. In this article, we review hierarchical statistical modeling and boundary analysis
(wombling) methods for areal-level spatial data that can be easily carried out using freely available statisti-
cal computing packages. We also propose a new edge-domain method designed to detect geographical
boundaries corresponding to abrupt changes in the areal-level surface. We illustrate our methods using
county-level breast cancer late detection data from the state of Minnesota.