Dave,

Thanks for the advice.  I'll take a look at MANOVABOARD

Craig

>>> [EMAIL PROTECTED] 04/19/05 7:31 AM >>>

Dear Craig-

It sounds to me like you need to do a MANCOVA, and analyze the
dependence of shape on both a grouping variable and a size
variable.  This will let you know if there is a dependence of
size on shape, of size on group membership and if there is an
interaction of group x size (meaning different dependencies of
shape on size in the two groups).

Fred Bookstein outlined a resampling approach to this issue in
a morphmet posting a while back.  You may get some other
responses on morphmet related to that approach.

I wrote a piece of software (called Manovaboard, I'm afraid)
that carries out resampling based MANOVA and MANCOVA of shape
data, using one covariate and one or two group variables. 
This piece of software attempts to use the approach outlined
by Bookstein.  I haven't published any results based on it yet
, but if you are interested, go ahead and use it.   To the
best of my knowledge it is working correctly.

It is part of my IMP series software which may be found at

http://www2.canisius.edu/~sheets/morphsoft.html 

and/or

http://www2.canisius.edu/~sheets/moremorph.html 

You may need several pieces of IMP software in addition to
Manovaboard, including CoordGen, which allows you to change
file formats.

You can also do the Mancova based on partial warp scores using
Jim Rohlf's tpsRegress program.  tpsRegress is a very nice
piece of software, we just found it a bit daunting to do
complex Mancova analyses with it, your question is relatively
straightforward.

Best wishes-

Dave Sheets


---- Original message ----
>Date: Mon, 18 Apr 2005 14:58:15 -0700
>From: "Craig Busack" <[EMAIL PROTECTED]>  
>Subject: Thin Plate Splines and Size  
>To: <[email protected]>
>
>Morphometricians,
>
>I'm looking for shape differences between two groups of fish.
  My plan was to do a MANOVA of partial warp scores, and then
canonical variate analysis to explore the shape differences. 
I have just learned, however, that my groups of fish differ in
size (about 0.5 SD), so now am wondering how to evaluate how
much of the shape difference I see (and based on preliminary
analysis, it is enough to classify fish by CVA with about 95%
accuracy) is actually due to size differences.   I realize
this may be an elementary question, but what's the best way to
get at this problem?  The guidance I have so far says that
multivariate regression of partial warps on centroid is the
best way to begin.
>
>Thanks
>
>Craig Busack
>Washington Department fo Fish and Wildlife
>
>
H. David Sheets, PhD
Dept of Physics, Canisius College
2001 Main St
Buffalo NY 14208



Reply via email to