Wednesday 15 Jan 2014: Small Area Estimation on a National Scale
Paul Hewson - Plymouth University
Plymouth University, Fitzroy 208 14:00-15:00
There is a vast literature on small area estimation. Our contribution has been to apply this work on a national scale, for example for the 2011 Skills for Life Survey. Currently, we are developing simple Hamiltonian MCMC algorithms on a GPU equipped machine in order to t models with large numbers of random effects in order that we can deal with household structure in the survey data as well as incorporating random effects that respect the survey design. The auxiliary variables (largely drawn from the census) are only available in limited aggregate form, typically in no more than three or four way tables. We describe methods for simulating census data from mixtures of graphical models fitted to the released census data which are combined with the survey models to yield small area predictions with a full accounting for uncertainty in the prediction process.