Brooke Fridley
Iowa State University
Friday, February 14, 2003
3:30 PM
Moos 2-620
Minneapolis Campus
Abstract:
The analysis of spatially correlated data involving observations falling below a detection level occurs occasionally in environmental studies. Replacing the censored observations with the level of detection (LOD) or some function of the level of detection, like LOD/2, is a common practice. The resulting parameter and standard error estimates found with the use of a single imputation method are biased and thus lead to inaccurate prediction of the level of contamination at a given location. A data augmentation procedure for the analysis of censored spatial data in the context of a Bayesian spatial model will be presented. Comparison of the data augmentation method to the LOD method and the LOD/2 method will be illustrated via a contaminated soil site.