My software is, of course, also based on the Kuhl and Giardina paper so that cannot be the difference. As others have commented, the prior alignment of the shapes is critical to any meaningful interpretation of the Fourier coefficients as shape variables. The outlines have to have the same orientation,size, and starting point. They also have to have the same "sense" (all clockwise or else all counter clockwise).
The various methods of alignment are somewhat primitive and there is a need for additional work. A key problem is in using PC1 to get the same orientation of the outlines. It is arbitrary which end of PC1 one calls the positive end. If some are flipped the wrong way then that will result is what looks like major shape differences. My software tries to keep the orientation consistent but it may not be perfect if some specimens are initially in very different orientations. ---------------------- F. James Rohlf, Distinguished Professor Dept. Ecology & Evolution, Stony Brook University www: http:\\life.bio.sunysb.edu/ee/rohlf > -----Original Message----- > From: morphmet [mailto:[EMAIL PROTECTED] > Sent: Friday, June 08, 2007 10:22 AM > To: morphmet > Subject: Re: Fourier > > Both programs are using the covariance matrix, so this is not > the problem. > > I've did my analysis in PAST with the box 'Invariant to > rotation and starting position' on and off, but when it was > on the difference is even higher. When I look at the figures > of my specimens reconstructed with the harmonics (I've 20 > harmonics in each program) they look the same in both > programs(same rotation and size). > > The EFA in SHAPE is based on Kuhl and Giardina - Elliptic > fourier features of a closed contour (1982). > Could it be that is causing some of the differences? > > I'll have a look at the other programs that were suggested. > > Greetings, > > Celine > > > > > ----- Original Message ----- > From: "morphmet" <[EMAIL PROTECTED]> > To: "morphmet" <[email protected]> > Sent: Thursday, June 07, 2007 8:17 PM > Subject: Re: Fourier > > > > Celine, > > as Dr. Hammer has pointed out, check the starting position and > > rotation of your images in Past. I cannot exactly remember what was > > the origin of the problem in terms of data recording, but > one of our > > students had a similar problem when using EFA (I think) and > SHAPE. At > > least the symptoms were the same: a huge difference in the variance > > explained, which was higher in the EFA PC1. When we looked at the > > graphic representation of the first principal component > from the EFA > > data, we realized that it was merely expressing the > orientation of the > > structure (i.e. if it was "pointing to the left or to the > right"). So > > orientation (determined by rotation and starting point) of course > > accounted for most of the variability (and the variance > explained was > > much higher, because we had entered an artifact resulting in extra > > variance). When the origin and alignment were properly specified in > > SHAPE (i.e. all the images were "pointing in the same > direction"), the > > variance explained was lower, but the principal components > were expressing meaningful shape differences. > > > > Hope this helps, > > Luis > > ________________________________ > > Luis Cabo > > Mercyhurst Archaeological Institute, > > Mercyhurst College, > > Erie, PA > > E-mail: [EMAIL PROTECTED] > > > > ----- Original Message ----- > > From: "morphmet" <[EMAIL PROTECTED]> > > To: "morphmet" <[email protected]> > > Sent: Thursday, June 07, 2007 11:11 AM > > Subject: Re: Fourier > > > > > >> Hi, I'm the author of PAST. The EFA algorithm in PAST is the one > >> described in this paper: > >> > >> Ferson, S.F., F.J. Rohlf & R.K. Koehn. 1985. > >> Measuring shape variation of two-dimensional outlines. > >> Systematic Zoology 34:59-68 > >> > >> I don't know Shape. However, one possible source for the > difference > >> is the standardization procedure - whether, and if so how, the > >> program standardizes for size, rotation and the starting > position for > >> digitization. > >> > >> Did you tick the "Invariant to rotation and starting > position" box in > >> PAST? > >> > >> And as Dennis Slice said - check whether the PCA standardizes for > >> variance ("PCA on the correlation matrix"). > >> > >> It would also be interesting if you could test Rohlf's EFA > program, > >> which should give similar results as PAST (?). > >> > >> > >> Regards, > >> > >> Oyvind Hammer > >> Natural History Museum > >> University of Oslo > >> > >> > >> On Thu, 7 Jun 2007, morphmet wrote: > >> > >>> Dear morphometricians, > >>> > >>> I'm doing some research on headshape dimorphism in European Eel > >>> using elliptic Fourier analysis. > >>> I've used the programs TPSUtil and TPSDig to get the > coordinates of > >>> the contours of the heads and then used the program PAST > to analyze > >>> these (with PCA). > >>> I've also used the program Shape on the same specimens, but the > >>> results were different. > >>> > >>> When I used Shape the first principal component explained > about 48% > >>> of the variance, but when I used PAST the first PC > explained 66% of > >>> the variance. > >>> > >>> Is there anyone who can explain this difference and maybe tell me > >>> which method is the best to do a Fourier-analysis? > >>> > >>> Thank you in advance > >>> Celine Ide > >>> > >>> > >>> Ghent University > >>> Evolutionary Morphology of Vertebrates KL Ledeganckstraat 35 > >>> B- 9000 Ghent > >>> Belgium > >>> +32 92645220 > >>> [EMAIL PROTECTED] > >>> > >>> > >>> > >>> -- > >>> Replies will be sent to the list. > >>> For more information visit http://www.morphometrics.org > >>> > >>> > >> > >> > >> > >> -- > >> Replies will be sent to the list. > >> For more information visit http://www.morphometrics.org > > > > > > > > -- > > Replies will be sent to the list. > > For more information visit http://www.morphometrics.org > > > > > > -- > Replies will be sent to the list. > For more information visit http://www.morphometrics.org > -- Replies will be sent to the list. For more information visit http://www.morphometrics.org
