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

Brusatte et al. 2008, Response to David Peters

Posted for Steve Brusatte.

Thanks to Bill and  David M. for chiming in and responding to David Peters' 
questions.  A  couple of things to add:

First, on the subject of supertrees, Bill  summarized our analysis quite 
well.  I wouldn't say that we utilized a  supertree per se, since proper 
supertrees are summary hypotheses of numerous  separate phylogenies (for 
example, see 
the recent Lloyd et al. 2008 supertree of  dinosaurs).  We didn't combine 
source tree topologies, but did use a  character-based approach.  However, as 
outlined, we didn't just take  the 437 characters in our disparity analysis, 
put them into PAUP, and then used  what came out.  Instead, we first ran a 
higher-level phylogenetic analysis  of all crown group archosaurs, which is 
on a separate dataset that is now  in review.  But, since this study was a 
higher-level look at all archosaurs  we only used a few exemplars to represent 
speciose ingroup clades such as  Aetosauria, Phytosauria, etc.  The disparity 
analysis needed many more  representatives of these clades.  So, we took the 
first MPT from our  higher-level study and then "grafted on" the ingroup 
relationships of these  newly added taxa based on reference to published 
topologies of these  clades.  Hope that clears things up on this.

Second, to answer David  P.'s specific question about the difference between 
a cladistic character list  (which we did not present in the Science paper) 
and a database that quantifies  overall similarity (which we DID present in the 
paper.  David M answered  this quite ably, but I have a few things to add.  
First, if you glance at  the supplementary info for the paper it really looks 
like we have a cladistic  dataset, since we have a set of several hundred 
characters that are coded into  discrete states with 0,1,2,3, etc.  But, just 
because the characters follow  this format does not mean they are cladistic 
characters.  Our goal was not  to construct a phylogenetic hypothesis, but 
rather to 
represent the entire  archosaur skeleton and quantify overall body form.  Of 
course, many  characters in that list ARE cladistic characters, but not all of 
them.   Three main differences separate proper cladistic characters and overall 
 similarity characters:

1) Autapomorphies: As David M mentioned, you don't  normally want to include 
autapomorhies in a cladistic dataset, since they don't  give any information 
on which groups are related to each other.  Rather,  they only record 
characters unique to one terminal taxon in the analysis.   But, for overall 
approaches, we want to quantify morphospace.   One terminal taxon may have 
numerous unique characters, which we need to include  so we can place it in 

2) Correlated Characters: Also  mentioned by David M, and very well 
explained.  In a cladistic analysis you  want to be sure that your characters 
independent of each other.  If you  have 20 characters that are all 
then they will bias the  dataset.  Why?  Because those characters all mean the 
same  thing.  They are interrelated, so they are really one feature.  But,  for 
overall similarity and morphospace studies, this is less of a problem.   Of 
course, you don't want to have several characters saying the same thing, but  
we wanted a complete coverage of the entire skeleton, so we were less stringent 
 with throwing out possible correlated characters.

3) Inapplicable  Characters: In a cladistic dataset, if you have a character 
related to a certain  feature of the arm, then you must score this as 
inapplicable (?) in animals  without arms, such as snakes.  But this isn't good 
morphospace.   We want to minimize question marks, both because they don't give 
any information  and because, operationally, the multivariate statistical 
techniques do not like  them.  So, when there were inapplicable characters, we 
always created a new  state ("feature absent") so we could give a positive 
(1,2,etc.).  For  instance, there are several characters relating to the form 
of the body armor  that are only applicable to aetosaurs, one small part of 
archosaur  morphospace.  If we only coded them as present in aetosaurs and as 
"?"  inapplicable in everything else, then this would tell us nothing about  
morphospace.  It wouldn't even lead to aetosaurs coming out as their own  
Adding a new state was necessary.

I realize these concepts  may be a bit abstract, and this explanation a bit 
long.  But, the important  message is that not all discrete characters scored 
as 0,1,2, etc. are cladistic  characters.  Cladistic characters should be 
carefully selected, so that  correlated characters are avoided and inapplicable 
character states are scored  as "?".  We intentionally did not do this so we 
could better quantify  overall morphology, and thus it's no surprise that if 
run our dataset in  PAUP that a weird tree may come out.

Stephen  Brusatte, MSc
American Museum of Natural History
Columbia University, New  York

**************Looking for simple solutions to your real-life financial 
challenges?  Check out WalletPop for the latest news and information, tips and 
calculators.      (http://www.walletpop.com/?NCID=emlcntuswall00000001)