Is there anyone who can tell when the settings are optimal in Bayestraits when using the bayesian option? I've tried different settings and noticed crazy high transition rate parameters when I use a hyperprior of 0-100 (q01 and q10 with peak around 300) compared with 0-10 (q01 and q10 max 20).
In the manual, they mention that you can use ML to get an idea about the priors. So with the same dataset doing a ML analysis, my LH and Harmonic mean are on average -15 and -18 resp. What does it say for my prior settings? And when is it better to you the reversible-jump hyperprior setting? Only when your model is overparameterized?
Hi Dieter,
I think you can take a look at your ML estimates of q01 and q10 and then set uniform priors around these values for the MCMC run, e.g. if the ML values range between say 0.01 and 0.50 for different trees (if you have a set), using a uniform 0-1 prior is more reasonable than the default uniform 0-100. See http://bodegaphylo.wikispot.org/BayesTraits_with_Discrete_Traits_(Glor)
/Maria
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics