-------- Original Message --------
Subject: Trad mulitvariate allometry
Date: Wed, 12 Oct 2011 08:18:34 -0400
From: Greg Campbell <[email protected]>
To: MORPHMET mailing list <[email protected]>
I am trying to study multi-variate allometry in the common marine
bivalve, the blue mussel, to see if the pattern of allometry varies with
height on the shore.I have several samples from each of a number of
known shore heights.I am using simple dimensions (length, height, width,
shell thickness), so this is a ?trad morphometrics? problem, not GMM (I
am faced with thousands of shells). I cannot use anova or manova, since
the average size is quite arbitrary (depends on who harvested the
mussels more than anything else).
Would you test for significant differences by carrying out a factorial
mancova on the log-transformed dimensions (using shore height as a
factor), even though this employs ordinary-least-squares regression
(OLS) where a functional relationship (not a predictive one) suggests
reduced-major-axis regression (RMA)?
I see there are some software programs that carry out pair-wise
comparisons following RMA, by calculating the cosine of the angle
between the regressed lines (the dot-product of the vectors), often with
bootstrapping.Do these work with ?traditional? dimensions as well as
GMM?Since they are pair-wise comparisons, do you have to correct for
experiment-wise error rates, and if so, how (Dunn-Sidak? Bonferroni?)?Do
they test for significant differences in the intercepts if slopes are
not significantly different?
Any clarification will be gratefully received.
Greg Campbell
The Naive Chemist
[email protected] <mailto:[email protected]>
