-------- Original Message --------
Subject: Re: allometry: size correction of multiple groups
Date: Fri, 6 Mar 2009 06:30:19 -0800 (PST)
From: Chris Klingenberg <[email protected]>
Reply-To: [email protected]
Organization: University of Manchester
To: [email protected]
References: <[email protected]>
Dear Rebeca
Yes, size correction by pooled within-group regression is assuming that
the groups share the same allometry (including the implementation in
MorphoJ).
If the data at hand violate that assumption, there are tree choices:
(1) do nothing (analyze your groups without size correction),
(2) do some size correction for each group separately, or
(3) do a pooled within-group regression anyway.
Option (1) is the best way if the regressions within groups differ
because size ranges are small (i.e. poorly defined regressions). In that
case, a size correction wouldn't do much good anyway. Also, if the
regressions in the groups truly differ dramatically, this may be the
only honest option.
Option (2) may seem appealing at first, but you end up with different
groups that have undergone different size corrections and therefore
aren't really comparable. Also, if you are interested in identifying
group membership, this method of size correction would require you to
know group membership ahead of the analysis in which you want to find
out which group your specimen belongs to...
In practice, option (3) is still often the best. If the allometric
regressions in the different groups are not drastically different from
each other, the pooled within-group regression can provide a
'compromise' estimate of allometry that can be used for size correction.
In this case, the residuals from the regression on size will not be
perfectly uncorrelated with size within groups, but often the variation
within groups is still reduced quite substantially, so that group
discrimination is enhanced. There are no hard-and-fast rules when this
is satisfactory (after all, the statistical model of parallel
allometries is wrong), but the allometries in different groups are quite
often similar and then you might go for the compromise aproach.
In most cases that I have seen, size correction with pooled within-group
regression improves the separation of groups, sometimes quite
dramatically -- even when allometries aren't quite parallel.
I hope this helps.
Best wishes,
Chris
--
***************************************************************
Christian Peter Klingenberg
Faculty of Life Sciences
The University of Manchester
Michael Smith Building
Oxford Road
Manchester M13 9PT
United Kingdom
Telephone: +44 161 275 3899
Fax: +44 161 275 5082
E-mail: [email protected]
Web: http://www.flywings.org.uk
Skype: chris_klingenberg
***************************************************************
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
