Date/Time
Date(s) – 14/12/2016
3:00 pm – 4:00 pm
Location
C106
Categories
Bayesian hierarchical models: anthropometric and cardiometabolic risk factors
Abstract
We have developed complex Bayesian hierarchical models of adult body mass index, diabetes and attained height, using data for millions of people to make our estimates. These models provide information for every country in the world over several decades or more. To do so, they must borrow strength across geographical units and time, making estimates for countries and years with no data.
I will describe the various features of the models, including their hierarchy, a non-linear random walk over time, the age model and the covariates used to inform model fitting. I will also explain the coding of the MCMC sampler used to fit the models. Finally, I will present recently published estimates for some of the variables, and will show how we have presented these results using dynamic visualisations.
Dr. James Bentham biography
I am a statistician with experience of analysing big data for several projects. I carried out an MSc in Probability and Statistics at the University of Sheffield, before moving to Imperial College to start a PhD in the Department of Mathematics. In this project, I created a numerical model of several million free text descriptions of patient safety incidents, and searched for small groups of similar incident types. I then carried out a postdoc at King’s College, where I worked on a genome-wide association study of lupus. For the past three years I have been working as a postdoc back at Imperial, carrying out the work which I will describe in my talk.