No subject

[Magnus Liden]

Dear fellow taxacomers,

TWICE the last year I have had papers on review that have attempted
bootstrapping on a posteriori-weighted data. To me it is nonsensical, and
against the very idea of bootstrapping, that characters not included in a
particular replicate should affect the weights of those that are included
and that the fit to a "best" all-character-tree should dictate how a subset
of the characters are weighted on this or alternative trees in a bootstrap
replicate. The relevance of a posteriori-weighting per se is an issue under
debate. I do not feel confident to take side here, but it is intuitively
obvious that slight initial differences can haphazardly tip the balance and
reinforce the support for marginally "better" trees. That is OK. I suppose,
as a means to select a tree for presentation (if, for some reason, one does
not want to present a consensus tree), but using these a posteriori weights
in a confidence estimate seems completely circular, and will give vastly
blown up, and quite meaningless (?) bootstrap-values, at least if weighted
characters are sampled as if repeated.

What are your opinions? Has anyone else encountered this approach to
bootstrapping? Has the possibility that weighting could be (mis-)used in
this way not been realized by program-constructors, and hence not warned
against? Or, is it even OK, and I am ute och cyklar?

Magnus Lid=E9n

-----------------------------------------------------------------------
Magnus Lid=E9n
Systematisk Botanik, Carl Skottsbergs gata 22B
S-413 19 G=D6TEBORG (GOTHENBURG), SWEDEN

Tel.nr.: xx46 (0)31 7732669
=46ax.nr.: xx46 (0)31 7732677
e-mail:magnus.liden at systbot.gu.se
homepage: http://www.systbot.gu.se/staff/maglid/welcome.html