Individual and combined ESS issue

[Mihajlo Milic]

Dear group members,

I ran several independent MCC runs in the BEAST to check for the convergence between the chains. I've analyzed the results in the Tracer, and I observed fairly high (>200) ESS values for all parameters from each run. However, when I looked at the combined trace, I saw that the combined ESS value for the likelihood parameter was significantly lower (<200) compared to the individual runs. What was even more interesting to me is that the log file with the individually highest ESS values significantly contributed to this decrease in the likelihood's combined ESS.

Does anyone know why is this happening? Should I try to fix this, and how? Am I safe proceeding with the analysis in other programs (LogCombiner and TreeAnnotator)? I would appreciate any advice and suggestions.

Best regards,

Mihajlo

1. Marginal density distributions of the likelihood parameter from each run.

2. Trace for the combined likelihood parameter.