treeModel.rootHeight

[Ben]

Hi,

I'm trying to estimate the TMRCA of the clades within each of my four
intraspecies datasets. I have four seperate datasets, varying in
number of terminals (from 12 - 76) with different levels of variation
within each dataset, each dataset comprises two mtDNA regions. For
each analysis I'm using a fossil calibration for the root height, with
a normal distribution as below:

<normalPrior mean="11.0" stdev="3.0">
<parameter idref="treeModel.rootHeight"/>
</normalPrior>

using a UPGMA tree with a starting root height equal to the mean of
the fossil calibration as below:

<upgmaTree id="startingTree" rootHeight="11.0">
<distanceMatrix correction="JC">
<patterns>

However no matter how many generations i run each analyses for (10 -
100 Million) i get small ESS (>100) values and bimodal distributions
for the posterior and prior estimates, coupled with estimates of the
treeModel.rootHeight which are nowhere near the prior with 99% of the
samples either 0 or very small!

Any ideas what might be wrong, or if i have missed a setting?

Thanks in advance!

Ben :)

[Simon Ho]

Hi Ben,

This problem sometimes occurs because the prior on the rootHeight is
not sufficiently informative. You could try repeating the analysis
using a rootHeight prior with a much smaller standard deviation (e.g.,
0.5 or 0.1). If you continue to encounter MCMC convergence
difficulties, then the source of the problem probably lies elsewhere.

One other thing to check is that the starting value of the clock.rate
parameter is approximately the 'correct' order of magnitude, given the
time units.

Cheers,
Simon

[Ben]

Hi Simon,

Thanks for the suggestions, reducing the standard deviation on the
rootHeight prior did help increase the ESS's, but the biggest increase
came from adjusting the starting value for the clock.rate. The
analyses are behaving much better now!

Cheers,

Ben :)