Effective sample sizes less consistent with increasing MCMC generations

[Brenen Wynd]

Hi all,

I am generating a tip-dated phylogeny in BEAST 2.0 for an entirely extinct group of organisms using morphological characters. I have been able to generate the tree just fine and run all of the analyses, but I am looking for parameters to have a minimum ESS of 200 in tracer. I have found that I can get an ESS close to or above 200 for the likelihood with 10,000,000 generations (sampling every 2k trees), but if I change the generation time to 100,000,000 (again sampling every 2k generations), I get much worse estimates of the likelihood (both models run with a 25% burn in). Is this partially due to the fact that I'm sampling too many trees or are there parameters that I can adjust to get better results? (I have adjusted the scalefactors based on what was recommended following each run of BEAST).

Thank you in advance for any and all help with this!

Best,

Brenen

--

Brenen M. Wynd

PhD Candidate in Paleobiology

Department of Geosciences and Global Change Center

Virginia Tech

4044 Derring Hall (3051)

Blacksburg, VA 24061

[Natalia Chousou-Polydouri]

Hello Brenen,

What do your traces look like? This sounds like your chain has not reached stationarity, so it looks ok with the short run, but it "discovers" another area of the posterior distribution later in the long run and so the ESS is worse. Also, how many independent chains are you running? Do they converge?

best,

Natalia

[Brenen Wynd]

Hi Natalia, 

Sorry, this got buried in my junk mail. The traces seem to be mostly fine, and the tree topology of all of the independent runs have been the same. Foolishly, I have not been storing the independent runs as separate outputs and have been overwriting the output files to look at tree topology and traces. I will look to do a number of independent runs and look to combine the outputs in Tracer and see if they are finding the same optimum.

Thanks!

Brenen

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/ac32438a-a70a-4e5d-aede-4fb3d3dd0f8bn%40googlegroups.com.