Tomorrow night is the May meeting for the NSW Branch of the Statistical Society. This month we are in the funky new UTS building designed by Frank Gehry! Details are below. All are welcome. Cheers, Michael =========================== Date: Tuesday, 26 May 2015 Time: 6:00pm - 6:30pm: Refreshments 6:30pm - 7:30pm: Lecture 7:45pm - onwards: Dinner (at a nearby restaurant) Venue: Room 002, Level 5, University of Technology, Sydney - Building 8 (Dr Chau Chak Wing Building), 14 Ultimo Rd, Haymarket NSW 2000 Dr. Craig Anderson University of Technology, Sydney Identifying Boundaries in Spatial Modelling Disease mapping is the field of spatial epidemiology interested in characterising how disease risk across different geographical regions. A key aim is to identify regions which exhibit significantly elevated disease risk levels, thus allowing public health interventions to be focused on these areas. Bayesian models utilising a so-called Conditional Auto-Regressive (CAR) structure are typically used in these settings. These models usually assume a spatially smooth risk surface across the entire region, but this is not necessarily realistic in practice. Using a case study of respiratory hospital admissions in Glasgow, Scotland, a city with many localised inequalities, I will present two alternative approaches which use clustering techniques to allow for discontinuities in the spatial structure. One of these approaches utilised Integrated Nested Laplace Approximation (INLA), and I will touch on its use as a computationally efficient tool for approximate Bayesian inference. Biography of Dr. Craig Anderson Craig Anderson graduated with an Honours degree in Statistics from the University of Glasgow, and then achieved his PhD in Statistics within the same department. After completing his PhD, he moved to Australia to take up a position as a Postdoctoral Research Fellow at the University of Technology, Sydney. He is interested in statistics for health data, with a particular focus on spatial and spatio-temporal modelling of disease risk.