Zapraszamy do udziału w seminarium prof. Christophera Brunsdona

Seminarium „Model averaging and combination as a tools spatial data analysis” odbędzie się w środę 15.11 o godz. 15.00 w s. A203 na Wydziale Nauk Ekonomicznych UW.

Chris Brunsdon is Professor of Geocomputation and Director of the National Centre for Geocomputation at the National University of Ireland, Maynooth, having worked previously in the Universities of Newcastle, Glamorgan, Leicester and Liverpool, variously in departments focusing on both geography and computing. He has interests that span both of these disciplines, including spatial statistics, geographical information science, and exploratory spatial data analysis, and in particular the application of these ideas to crime pattern analysis, the modelling of house prices, medical and health geography and the analysis of land use data. He was one of the originators of the technique of geographically weighted regression (GWR). 
He has extensive experience of programming in R, going back to the late 1990s, and has developed a number of R packages which are currently available on CRAN, the Comprehensive R Archive Network. He is an advocate of free and open source software, and in particular the use of reproducible research methods, and has contributed to a large number of workshops on the use of R and of GWR in a number of countries, including the UK, Ireland, Japan, Canada, the USA, the Czech Republic and Australia.

Abstract: Modelling geographical data can lead to some interesting techniques, but also some uncertainty as to which technique is the most appropriate in a particular context. For example, one may wish to make use of models including spatial autocorrelation in predictions, but there are many ways of encapsulating spatial autocorrelation. In many cases there is no clear guideline as to which approach is appropriate. In this situation, a number of techniques can be used to ’score’ models on how well they may fit the data. However, often a number of models show similar performance - in such cases choosing a single model as ‘best’ may not be a good approach. Here tools and some theory related to the combination of models (mainly by various kinds of model averaging) will be considered. Examples applied to the United Kingdom’s Brexit vote, and early rates go Covid-19 in Ireland will be used to illustrate these ideas.