Re: where have all the etc.

[David Marjanovic <david.marjanovic@gmx.at>]

I wrote...

> > Referees have brought this up, suggesting I use 20,000 taxa
> > and as many characters,
>
> For 20,000 characters you'd obviously need a lot more than 20,000
> characters!!!

Er... for 20,000 _taxa_ you'd need lots more characters... :-]

While I am at it:

----- Original Message -----
From: "david peters" <davidrpeters@earthlink.net>
Sent: Tuesday, July 05, 2005 7:16 PM

> > No, as David mentioned, adding taxa while holding character
> > number constant will only *decrease* resolution.
> >
> > Nick Pharris
> > Department of Linguistics
> > University of Michigan
>
> NOW we're getting to the meat of the matter. Where is the reference that 
> supports this? [...] What are the "ideal" ratios? And how were these 
> numbers validated?

The numbers were found by simulation. The principle, which was corroborated 
by the numbers, is mere common sense, as I tried to explain via the 
signal/noise ratio.

One more thing. If you had no convergence at all in your data set, the 
numbers of characters and taxa could in principle be identical and still 
give you a fully resolved tree (if you happened to make a fortunate choice 
of characters -- one autapomorphy for every node). So, getting a fully 
resolved tree with a characters/taxon ratio of 1.2 is not a priori 
impossible. But you have told us how the homoplasy index in the tree you get 
is way higher than usual, instead of way lower. The more noise you have, the 
more characters you need to cancel out the noise and increase the signal.