ESS of multiple BEAST runs

[Mihajlo Milic]

Dear group members,

I have run several independent MCMC runs in the BEAST to check for the convergence of the runs. When I analyzed obtained log files in the Tracer, I observed fairly high ESS values (>200) for all parameters from each run. However, when I inspected the combined results in the Tracer, the ESS for the likelihood parameter appeared to be low (121). What makes things even more interesting to me is that the addition of the log file with the highest ESS values among all made the combined trace file's ESS decrease from 191 to 121. What could be the reason for this happening? How can I fix this, and am I "safe" to proceed with the analysis in other programs (LogCombiner and TreeAnnotator)?

I am attaching images for marginal density distribution of the likelihood parameter from each run. And also the results before and after the addition of the mentioned log file with the highest ESS values. I hope this can help you formulate your answer.

Best regards,

Mihajlo