ABSTRACT
A recent development within the R world is the arrival of nimble. Nimble is a wide ranging package which can parse BUGS or JAGS code and to write its own Markov chain Monte Carlo samplers. At a basic level nimble can convert BUGS code almost directly into nimble models and then run the code via purpose written C++. The log normal model is straightforward to program in nimble and uses the same data as the GP model. The BYM or convolution model can also be fitted easily on nimble. Nimble provides both for an Improper Conditional autoregressive (ICAR) prior distribution and a proper Proper-CAR distribution. This chapter focuses on the ICAR. The BYM model has two component random effects : an uncorrelated heterogeneity term and a Correlated Heterogeneity term. A simple mixture model can also be fitted within nimble. This model assumes a non-informative prior distribution for the mixing parameter.
