MCMC method: analyzing the posterior sample

As I mentioned in my previous post, the posterior sample can be treated as would any posterior sample from a Bayesian analysis. For instance, we should expect to compute parameter estimates, effective sample sizes (which takes into account the degree of correlation among our samples), and 95% credible intervals mostly in the typical way. We can do many of these things within R using the MCMC diagnostics package {coda} (Plummer et al. 2010); however, for those users that prefer to use the Java program Tracer (by Rambaut & Drummond), this is fine to. We just need to first write our MCMC samples to a text file:

> write.table(x=res$mcmc,file="mcmc.sample.txt",sep="\t", row.names=FALSE,quote=FALSE)

The image below is a screenshot of the program Tracer, showing the trace of the log-likelihood of our analysis from last time:



There is no explicit manual for Tracer, but it is very easy to use. In additional posts, I will discuss how to do these analyses in {coda} as well as how to summarize the posterior sample for the position of the rate shift (something that cannot be done easily using either Tracer or {coda}).