Appropriate size correction technique?

[morphmet_moderator]

----- Forwarded message from Sophie E Webster -----

Date: Sun, 17 Jun 2012 13:02:24 -0400
From: Sophie E Webster
Reply-To: Sophie E Webster
Subject: Appropriate size correction technique?
To: morphmet@morphometrics.org

Dear Morphometricians,

I am attempting to perform PCA on size corrected groups, but I am new to morphometrics and am uncertain if my method is valid - I would be very grateful for any comments you may offer!

I have four populations with a juvenile group and an adult group from each. I wish to compare the mean change in shape from juvenile to adult stage (though not the allometric change - the juveniles are very phentoypically variable, the adults are less so. I am trying to detect the 'direction' of shape change driven by selection without confounding this with the allometric component) and then compare this juvenile to adult change between all four populations.

I used tpsDig to place landmarks on the shape. In terms of analysis, I began by following the guidance in Viscosi and Cardini (2011); Leaf Morphology, Taxonomy and Geometric Morphometrics: A Simplified Protocol for Beginners. Following the guidance for size correction, I compared slopes using tpsRegr, which indicated that some slopes were significantly different so each other. In all but one case, juvenile slopes were significantly different to the adult slopes from the same population. There are also significant differences between some juvenile-juvenile and adult-adult slope comparisons between populations, but not all. The paper states "When slopes are different, allometric trajectories are pointing to different directions and one cannot easily control for the effect of size on shape in tests of group differences", so it is at this point where I have become unsure whether my method is appropriate. I have found it difficult to find information in the geometric morphometric literature which clearly explains what to do if the allometric trajectories differ between the groups being compared.

After checking slopes in tpsRegr (and becoming thoroughly confused as to my next step!) I went ahead and performed a multivariate regression in MorphoJ on each group separately (Population 1 juveniles, population 1 adults, population 2 juveniles, population 2 adults etc.). This was regressing procrustes coordinates (dependent variable) against log centroid size. I then exported the residuals from each individual regression into a csv file, which I imported into R to perform a PCA on the regression residuals.

As I previously mentioned, I am unsure if this is a valid way of approaching this problem or whether I have made some major oversight. Any help is most gratefully received, as are any suggestions of papers relevant to my question I may have overlooked!

Many thanks,


Sophie Webster


PhD student
Department of Animal and Plant Sciences
University of Sheffield

----- End forwarded message -----