In article <D3HDxB.HHE at zoo.toronto.edu> mes at zoo.toronto.edu (Mark Siddall) writes:
>>In article <D3F8CA.Izz at sci.kun.nl> gertjan at sci.kun.nl (Gert-Jan Caspers) writes:
>>>I wanted a maximum parsimony tree with bootstrap values from my
>>>protein data set,
...
>>>However, when I ran the resulting tree and some other trees I
>>>made up myself through Protpars again, using the User Tree option,
>>>I found that one of my own trees was one step shorter than the
>>>consense, bootstrapped, tree.
...
>If you simply want bootstrap values on your most parsimonious tree(s)
>I would be surprised if Joe did not include an option in Phylip to
>do this.
[I have edited out parts of both postings less relevant to my point here]
*blush* no there is none.
>I have thought about this a bit recently and have trouble with the
>use of parsimony to construct the pseudoreplictes only to
>abandon parsimony for the final best hypothesis.
>Neuman characterizes the issues as follows:
>Hypothesis Type I: Most parsimonious tree A is the best current estimate
>of phylogeny.
>Hypothesis Type II: Most parsimonious tree A is a good current estimate
>of phylogeny.
>Hypothesis Type III: Non-parsimonious tree B is a better current
>estimate of phylogeny than is most parsimonious tree A.
>>Randomization has nothing to offer Type I. Parsimony quarantees that it is.
>Randomization has much to offer Type II.
>Type III is false.
>>Arguably bootstrap trees are hypotheses of Type III.
No, they're just diagrams showing which groups are well supported. That is,
they give information about sets of trees intended as confidence intervals.
But one _could_ use them as point estimates. They may or may not be good
for that (in jackknife statistics the mean of jackknife values is often a
less biased value than the grand mean). Someone would have to investigate
this.
But I cannot see why we should _assume_ that parsimonious tree A must be
a better estimate than any possible non-parsimonious tree estimate. If we
do that then we are really in the realm of Hypothesis Type I, and have entered
a zone where the rules of statistics no longer are deemed of interest. In
which case one doesn't bootstrap anyway.
I don't see why, if the most supported groups found on the majority-rule
consensus tree are also found on the most parsimonious tree, (and that will
usually be the case) it should bother us if some of the others aren't.
-----
Joe Felsenstein, Dept. of Genetics, Univ. of Washington, Seattle, WA 98195
Internet: joe at genetics.washington.edu (IP No. 128.95.12.41)
