[R-sig-phylo] Felsentsein's contrast LRT

Joe Felsenstein joe at gs.washington.edu
Tue Apr 26 15:44:11 CEST 2011


Scott Chamberlain wrote:

> In Felsenstein's 2008 AmNat paper he states "Likelihood ratio rest  
> (LRT) are available of the hypotheis that a set of q characters  
> have on phylogenetic covariation with the remaining p - q  
> characters." And his software contrast gives only one LRT result in  
> the output even if you have say 3 traits. If you are interested in  
> contrasts and a LRT for all pairwise relationships of the 3 traits,  
> do you just run analyses 3 different times, each with the pair of  
> traits you are interested in?

You can do all three tests, but they are not
independent tests.  In Contrast I allow the user to
specify two sets of characters and test whether
they have evolved independently.

If you want to do all three possible tests:
1 versus (2,3),  2 versus (1,3),  and 3 versus (1,2)
keep in mind that those tests are not
independent tests (for example, a strong
correlation between 2 and 3 could cause
two of the tests to show nonindependence.
You would need to run Contrast three times
to do those tests.

Perhaps what is needed is a test that
finds the two sets to divide the characters
into, sets that are as much independent as
possible.   I have no immediate ideas how.

J.F.
----
Joe Felsenstein, joe at gs.washington.edu
  Dept. of Genome Sciences, Univ. of Washington
  Box 355065, Seattle, WA 98195-5065 USA



More information about the R-sig-phylo mailing list