I would like to simply point that PCA knows nothing about groups. It merely provides a lower dimensional representation of higher dimensional data that maximizes representation of total variance. The appearance of groups separations along, say, PC1, often occurs, but is in no way designed into the method.
If you have clusters, then MANOVA, preferably of the nonparametric variety, will tell you if there are significant differences between the predefined groups, and other methods like CVA will give an optimal low-dimensional representation of those group differences. Get thee to any decent multivariate text book, e.g., Krazanowski's Principles of Multivariate Analysis or many others, and now might be a good time to start learning R (http://www.r-project.org). Are you using tps programs? If so, you can do the MANOVA with tpsREGRE - regress shape on g-1 dummy variables encoding groups. The excellent help file should have the instructions. Best, dslice morphmet wrote: > Dear morphometricians, > > I have the following problem: > > I have performed a PCA of shape (relative warps analysis) on a set of > mouse mandibles from animals of different geographic origins. Now in a > plot of PC1 vs PC2, I can "see" that PC1 sorts specimens into broadly > overlapping clusters corresponding to the respective origins of the > mice, while PC2 (and the other PCs) do not. The problem is now that the > overlaps of the "population" clusters are rather broad so the question > is how different they actually are. Also, I have several "populations", > so it looks like a continuum of overlapping clouds. > Could you recommend a means to quantify and/or somehow test the actual > differences between "populations" along PC1? > > Louis Boell > > ------------------------------------------------------------------------ > Machen Sie Ihre E-Mail-Kontakte zu Messenger-Freunden! Einfacher > Adressimport! > <http://redirect.gimas.net/?cat=hmtl&n=M1007AI&d=http://messenger.live.de/ersteschritte_adressimport.html> > -- Dennis E. Slice Department of Anthropology University of Vienna ======================================================== -- Replies will be sent to the list. For more information visit http://www.morphometrics.org
