[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]

Re: Sci Am - present.

>In a message dated 98-02-18 02:35:17 EST, cbrochu@fmppr.fmnh.org writes:
><< Based on these, parsimony is a reliable tool.
> Likelihood methods, from these analyses, may be more reliable, but these
> are model-driven and much more appropriate for molecular data. >>
>Not particularly reliable. Talk at the SVP this fall noted about 18% of
>cladograms failed to reconstruct an artificial phylogeny even after
>stratigraphic data was included in the analysis.

I disagree with their assessment of "fail to reconstruct phylogeny."  Did
they fail to recover *all* of the real phylogeny, or only parts of it?  And
if only parts, by how many taxa or nodes were they off?  IIRC, the
recovered trees were not significantly different from the real tree, and
there are statistical tests we can perform to distinguish significant
conflict from minor blips.

This is an important distinction - is a recovered tree "way wrong" if it
agrees with the true phylogeny over all taxa but one?  No one has ever
claimed that parsimony analyses are guaranteed to always recover the
completely correct tree, but nearly all simulation studies done to date,
along with the one simulation that actually had a real phylogeny (the virus
work done in the Hillis, Bull, and Molineux labs at UT Austin), find that
under most conditions faced by systematists, it is a robust tool that will
recover a reliable phylogeny.

We must also note that in most cases, different data sets recover extremely
similar trees.  Again, they may differ over some parts of the tree, but in
most cases the fabled "molecules versus morphology" debate turns out to be
over minor differences.  If such disparate data sets so frequently recover
such similar trees, what other explanation could there be other than the
presence of a common evolutionary signal among disparate sources of

There was a "simulation study" published in Systematic Botany a couple of
years ago by Lamboy attempting to assess a similar stat, and trying to show
that parsimony based on morphology was hopelessly flawed.  He was soundly
refuted on this basis - I listed the refs below.  Lamboy's paper is worth
reading for the amusement value, at least.

Moreover, stratigraphic data are not heritable and therefore cannot be
included with biotic information in a parsimony analysis; this also renders
the value of stratigraphic data as an independent test of phylogeny useless.

I was willing to give
>cladistics 90-95%, but I didn't expect it to be as bad as 82%.

Again - over a single tree, or over all trees?  I strongly recommend
reading these:

Hillis, D.M., J.P. Huelsenbeck, and C.W. Cunningham.  1994.  Application
and accuracy of molecular phylogenies.  Science, 264:671-677.

Huelsenbeck, J.P., and D.M. Hillis.  1993.  Success of phylogenetic methods
in the four-taxon case.  Systematic Biology, 42:247-264.

And here's the Lamboy vs. UT group exchange.  I cite it not because
anything meaningful came out of the initial simulations - indeed, the
results in the first of these papers are compromised by the extremely poor
design of the experiment - but because of the points made in the response
by Wiens and Hillis, which are relevant to the point I made above.

Lamboy, W.F.  1994.  The accuracy of the maximum parsimony method for
phylogeny reconstruction with morphological characters.  Systematic Botany,

Wiens, J.J., and D.M. Hillis.  1996.  Acuracy of parsimony analysis using
morphological data:  A reappraisal.  Systematic Botany, 21:237-243.

Lamboy, W.F.  1994.  Morphological characters, polytomies, and homoplacy
indices:  resopnse to Wiens and Hillis.  Systematic Botany, 21:243-253.


Christopher Brochu

Department of Geology
Field Museum of Natural History
Lake Shore Drive at Roosevelt Road
Chicago, IL  60605  USA