Ludvig Mortberg wrote:
> The information value of a
> single nucleotide substitution isn't very high.
That's why you're supposed to perform statistical tests on:
each possible pair of aligned sequences (to ensure that they're not
the total initial data set (to ensure that its variable sites aren't
randomized with respect to phylogeny); and on
your final proposed tree (to show how strongly the data actually
support that specific tree, as opposed to alternate trees).
I'm not yet skilled in most methods of molecular phylogeny. But, for
protein parsimony, with which I've worked: pairs of aligned sequences can
be tested by shuffling-and-aligning 200x (with RDF2 or RSS); data sets can
be tested by tree-length skew (with PAUP or MacClade); and final trees by
bootstrapping (with Phylip or PAUP).
None of these three tests are entirely uncontroversial, but they're
intellectually respectable given the current limits of our knowledge. Any
tree based on "a single nucleotide [or codon] substitution" will get
easily rejected by such tests.
--Erich Schwarz (resisting the urge to write
a much snappier reply)