In article <6inick$54t at net.bio.net>, "James O. McInerney"
<j.mcinerney at nhm.ac.uk> wrote:
> There are two very separate issues here. Firstly, do you leave the
> parsimony-uninformative sites in the alignment when you are sampling and
> secondly, do you leave the parsimony-uninformative sites in the analysis when
> you are searching for a tree?
First I would like to point out that I have a problem with the traditional
description of the class of characters called "parsimony uninformative."
This includes variable characters for which all taxa but one share a
particular state. Such sites are in fact potentially informative. They
will be informative when the shared state is a true synapomorphy and the
single excluded taxon is basal to the rest. For this reason, I will
confine the rest of my comments to characters that are definitely
uninformative; i.e., the invariant ones.
> In my opinion the answer to the first [sampling] question is yes. You should
> leave the parsimony-uninformative sites in the alignment when you are
> generating the samples. The reasons are given in my answer.
I disagree on this point, because the number of invariant characters
originally measured is independent of the phylogenetic signal present or
absent in the potentially informative characters. Why should we have less
confidence in a node if we happened to measure 1000 invariant characters
instead of 10? It is the quality of the information that we are trying to
> To the second question, I think the answer is no. For precisely the reasons
> that were given by the submitter of answer 1.
I agree that invariant characters should also be omitted from this step,
simply because including them carries a computational cost without the
adding any information for a maximum parsimony analysis. This does not
hold in the context of a distance based analysis, or for some maximum
Guy Hoelzer e-mail: hoelzer at med.unr.edu
Department of Biology phone: 702-784-4860
University of Nevada Reno fax: 702-784-1302
Reno, NV 89557