Statistics for Spatial Data: What Lies Ahead?

• Activity Number: 331
• Type: Luncheons
• Date/Time: Wednesday, August 14, 2002 : 12:30 PM to 1:50 PM
• Sponsor: Section on Statistics & the Environment*
• Author(s): Oliver Schabenberger*+
• Affiliation(s): Virginia Polytechnic Institute and State University

Abstract

The core methods of statistical analysis for the three important spatial data types--statistical, lattice, and point data--are now well established. Their limitations have also become woefully clear: focus on continuous data, Gaussian random fields, stationarity, and isotropy assumptions. The flurry of research activities in the past years bears evidence to the inadequacy of the basic toolbox. Spectral methods, non-stationary random fields based on convolutions, hierarchichal models, and generalized spatial prediction based on generalized linear mixed model extensions tackle some of the shortcomings of the standard random field analysis. On the other hand, these extensions are computationally more intensive, often requiring MCMC or other simulation-based methods. Participants of this roundtable luncheon will discuss the merits and demerits of the standard and modern methods of spatial data analysis.