Relay-Version: ANU News - V6.2.0 06/23/97 OpenVMS AXP V6.2; site chasm
Path: chasm!tribune.meitca.com!bloom-beacon.mit.edu!panix!howland.erols.net!newsxfer3.itd.umich.edu!newbabylon.rs.itd.umich.edu!not-for-mail
Newsgroups: bionet.molbio.evolution
Subject: Re: rate variation in ML models
Message-ID: <343DC678.4C13 at umich.edu>
From: Mark Siddall <msiddall at umich.edu>
Date: Fri, 10 Oct 1997 02:08:56 -0400
References: <610d6n$kcp at net.bio.net>
Organization: Laboratory of Phylohirudinology
Lines: 18
NNTP-Posting-Host: pm121-02.dialip.mich.net
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
X-Mailer: Mozilla 3.01 (Win95; I)
Joe Felsenstein wrote:
> If there is no autocorrelation of rates among adjacent sites, the
> model is still i.i.d. (independent and identically distributed). But
> if, as is allowed in my DNAML and Yang's PAML, there is some autocorrelation
> among sites, then the model isn't i.i.d. This affects, for example,
> the validity of bootstrapping.
And, for that matter, the entire validity of the analysis (Felsentein,
1973). But, in any case, so long as DNA data are sequenced, and not
selected at random, the requirements of i.i.d. would seem to me to not
be met. There always is autocorrelation among sites in protein coding
genes and among some sites in rDNA data.
That is to say, we have ample evidence that all is not stochastic, so I
wonder why there is such widespread acceptance of models that are
predicated on stochasticity int he first place...
Mark
