next up previous contents
Next: Transforming the Variables Up: Algorithm Previous: Algorithm

Assessing Convergence

Convergence of the Markov chains is assessed by examination of the traces from the sampler. These are included for completeness for a number of runs in the results section.

The chains exhibited very high autocorrelations, which was a feature of the posterior distribution, and for this reason very many samples were taken. The autocorrelation within the chain was reduced by thinning the samples, and only using one in every 10. Consequently, the number of samples needed for the estimates in the results section was quite large indeed. After a number of exceedingly long run-times, it was decided to examine the sampling strategy more closely.



Cathal Walsh
Sat Jan 22 17:09:53 GMT 2000