In article <3q8oem$2us_001 at is.co.za>,
Neil Caithness <neil at gecko.biol.wits.ac.za> wrote:
>I think it's clear that one cannot treat RFLPs as caldistic character data
>due to the non-independence of fragment lengths (add one cut site between two
>others, and you loose one long fragment and gain two shorter ones).
One can use data in which some pairs of characters are correlated, and
where you don't know enough about the correlations to correct for them.
In effect, the fragments are incorrectly weighted. If the absence of
one and the presence of two others are, for example, the result of
a single event, you are weighting the collection as 3 when it should be 1.
That introduces a bit of noise. But it makes no particular bias in the
tree. With enough characters you should converge on the same tree as you
do with all the fragments correctly weighted.
So tossing the lot out seems to be an overreaction.
Joe Felsenstein joe at genetics.washington.edu (IP No. 188.8.131.52)
Dept. of Genetics, Univ. of Washington, Box 357360, Seattle, WA 98195-7360