Analyzing Small Geographic Area Datasets Containing Values Having High Levels of Uncertainty

Daniel A. Griffith1 and Robert P. Haining2
1. School of EPPS University of Texas at Dallas Richardson, TX, USA
2. Department of Geography University of Cambridge Cambridge, UK

1.; 2.

Abstract: Data collected and then post-stratified by small geo- graphic areas frequently result in small, or even zero, sample sizes for some areas. Government agencies faced with this out- come commonly suppress many of the sample-based attribute measures for confidentiality reasons. Meanwhile, modeling concerns of spatial scientists faced with this situation include the accompanying large standard errors for parameter esti- mates obtained with such data, as well as how to deal directly with any missing values. This paper addresses these two issues in the context of spatial statistical modeling that accounts for high levels of uncertainty for some data values in two specific datasets. Its purpose is to demonstrate ways of handling high levels of uncertainty in georeferenced data. In doing so, empiri- cally-based findings summarized in this paper illustrate se- lected approaches that can be employed to account for high levels of uncertainty for some data values in a dataset. Its impli- cations should be of interest to users in government, the private sector, and the academic community who engage in the model- ing of georeferenced data.

Keywords: Bayesian; error; imputation; missing data; Poisson regression; sample size; uncertainty

GriffithAccuracy2010.pdf507.46 KB