Autocorrelation and covariance

[Joshua Clayton]

Dear BEAST users,

I am unclear as to how the covariance statistic relates to the degree
of autocorrelation between mother-daughter lineages. I assume the
higher the covariance, the more rates are autocorrelated, but at what
point can we say rates are autocorrelated or not, and to what degree
(given the bounds of the HPD, which seem very large to me)? From my
own dataset the mean covariance is 0.00837, with 95% HPD [-0.141,
0.165], so what does this tell me about my data? Any help or
explanation would be greatly appreciated.

Thanks,
Josh

--------------------
Joshua Clayton
PhD candidate
University of Florida
Botany & Florida Museum of Natural History
PO Box 117800
Gainesville, FL 32611, USA
Phone: 1-352-273-1962
Fax: 1-352-846-2154
Email: jos...@ufl.edu

[alexei....@gmail.com]

Hi Joshua,

Given your large HPD, there is not much you can say about positive
correlation between parent and daughter branches from this analysis. I
should warn you that the way that the uncorrelated model is currently
implemented in BEAST (so that every branch must have a different rate
category) lends itself to *anti*-correlation. This means that real
autocorrelation has to be very strong before it will be detected by
BEAST. We are currently working on some refinements to the
uncorrelated models implemented in BEAST to make them more neutral in
this regard. In the mean time, your results don't suggest any positive
correlation. If the HPD does not include zero then you can consider
that strong evidence for correlation.

Cheers
Alexei

[Joshua Clayton]

Thanks Alexei, that's good to know.