Developing Grid enabled spatial regression models
Richard Harris, Min hua Jen, David Kilham, Edward Thomas, Chris Brunsdon, Claire Jarvis
Richard Harris, Min hua Jen, David Kilham, Edward Thomas
School of Geographical Sciences, University of Bristol, UK
Chris Brunsdon, Claire Jarvis
Department of Geography, University of Leicester, UK
Email address of corresponding author: M.Jen@bristol.ac.uk
Various methods of spatial analysis have been developed to detect and to explain geographical `hot spots' and clustering (in data), undertaking localized and non-parametric statistical testing to assess the significance of a spatial pattern. They do so by sequential repeat testing, creating computational demands as the algorithm cycles through spatial subsets of the data and then again, repeatedly, as the data randomly are redistributed across geographical locations within the study region. In this paper we introduce one such method known as Geographically Weighted Regression (GWR) and give the rationale for adapting it to the computational `grid' infrastructure of e-social science. The aim of the project is to run GWR's sequences of calibration, analysis and significance testing in parallel, applying the processes to 2001 Census data to produce a geographically calibrated index of deprivation. The research is funded by the ESRC's small grant scheme for e-social science.
