Computer-Intensive Statistics

‘Computer-intensive statistics’ is statistics that could only be done with ‘modern‘ computing resources, typically either • Statistical inference on small problems which needs a lot of computation to do at all, or to do well. Quite small datasets can need complex models to explain, and even simple models can need a lot of computation for a realistic analysis (especially where dependence is involved). • Statistical inference on ‘huge ’ problems. All of these terms are relative and change quite rapidly—according to the most commonly quoted version of Moore’s Law (see section 6 and Ripley (2005)) computing power will quadruple during your doctoral studies. One very important idea for doing statistical inference ‘well ’ on analytically intractable statistical models (that is, most real-world ones) is to make use of simulation. So most of this module could be subtitled simulation-based inference, as in Geyer (1999)’s comments about MCMC for spatial point processes: If you can write down a model, I can do likelihood inference for it, not only maximu