Spatial patterns of events that occur on a network of lines, such as traffic accidents recorded on a street network,
present many challenges to a statistician. How do we know whether a particular stretch of road is a ``black spot'', with a higherthanaverage risk of accidents? How do we know which aspects of road design affect accident risk?
These important questions cannot be answered satisfactorily using current techniques for spatial analysis.
The core problem is that we need to take account of the geometry of the road network. Standard methods for spatial analysis assume that `space' is homogeneous; they are inappropriate for point patterns on a linear network, and give fallacious results. To make progress, we must abandon some of the most cherished assumptions of spatial statistics, with farreaching implications for statistical methodology.
The talk will describe the first few steps towards a new methodology for analysing point patterns on a linear network. Ingredients include stochastic processes, discrete graph theory and classical partial differential equations as well as statistical methodology. Examples come from ecology, criminology and neuroscience.
Bio
Professor Adrian Baddeley is a statistician specialising in methodology for spatial and geometrical data (spatial statistics, stereology, image analysis). He was chair of statistics at the University of Western Australia for 16 years and has been jointly employed by UWA and CSIRO since 2006. He is currently a professorial research fellow in the Centre for Exploration Targeting at UWA. Adrian Baddeley is a winner of the Hannan and Pitman Medals and a Fellow of the Australian Academy of Science.
