-------- Original Message -------- Subject: RE: CVA and MANOVA Date: Wed, 27 Feb 2008 15:09:24 -0800 (PST) From: F. James Rohlf <[EMAIL PROTECTED]> Reply-To: <[EMAIL PROTECTED]> Organization: Stony Brook University To: <[email protected]> References: <[EMAIL PROTECTED]>
Statistically (and geometrically) a matrix of relative warp scores is just a matrix of principal component scores. If you keep all of the dimensions for which the eigenvalues are > 0 then you are just doing a rigid rotation of a multivariate space so ordination analyses, probabilities computed in a MANOVA, etc. must be unchanged. If you only use a few dimensions then you are projecting into a lower dimensional space and thus losing information so subsequent analyses can give different results. Difficult to predict in advance what the consequences will be (it could be that dimensions that account for relatively little of the overall variance just happen to be good at distinguishing groups - that would be lost if you happened to have discarded those dimensions. If you use relative warps as your variables then to be able to construct shapes you will have to back transform to coordinates. This can be done (this operation is built into tpsRelw). For other analyses, such as CVA based on relative warps you can use the trick of using the tpsRegr program to regress shapes (your original tps file) onto each vector of CVA scores. If you use all of the relative warp dimensions then the CVA scores should be identical to those based on the original matrix of partial warp scores or a matrix of GPA coordinates after the invariant dimensions have been removed. ========================= F. James Rohlf Distinguished Professor, Stony Brook University http://life.bio.sunysb.edu/ee/rohlf
-----Original Message----- From: morphmet [mailto:[EMAIL PROTECTED] Sent: Wednesday, February 27, 2008 5:39 PM To: morphmet Subject: Re: CVA and MANOVA -------- Original Message -------- Subject: Re: CVA and MANOVA Date: Wed, 27 Feb 2008 08:02:38 -0800 (PST) From: Cristian Correa <[EMAIL PROTECTED]> To: <[email protected]> References: <[EMAIL PROTECTED]> Hi Amelia, Comments from another beginner: I think you can use RW to reduce dimensionality before conducting MANOVA. I don't see the problem, unless you're severely violating assumptions (in which case you can do NPMANOVA). It may (or may not) be more difficult to visualize the results of your analyses (depending on the RW obtained and kept) because you will be moving in the more integrated and distorted space of RW compared to the more shape-comprehensive space of PW (using all PW). You could use a few RW (e.g. the first two, if they account for much of the variance) in a MANOVA for statistic test. Then do a CVA to visualize significant differences between groups and get centroids for each group. It is very likely that each CV axis will pick up a different RW more heavily. In this case it is easy to interpret morphological differences in each of the CV axes; you just have to move along the RW plots. I don't think you can directly predict shape from the CVA using RW. Can you? In summary, I would go for RW if I really need to reduce dimensionality, otherwise I'd choose using the full set of PWs. But I'm a novice as well and I'd love to see the answers of experts. Buena suerte! Cristian ----- Original Message ----- *From:* morphmet <mailto:[EMAIL PROTECTED]> *To:* morphmet <mailto:[email protected]> *Sent:* Wednesday, February 27, 2008 9:11 AM *Subject:* CVA and MANOVA -------- Original Message -------- Subject: CVA and MANOVA Date: Wed, 27 Feb 2008 05:15:56 -0800 (PST) From: M. Amelia Chemisquy <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>> To: [email protected] <mailto:[email protected]> Hi! I'm using geometric morphometry for analyzing orchid seeds, and since I am quite new at this (and at statistics two) I have a few questions. I want to perform a CVA and a MANOVA with my morphometric results, and I want to know if I can use the relative warps scores for doing this, or it can only be done with the partial warps. What if I don´t fulfill the number of samples (4 times the number of variables) needed, may I use the relative warps in this case? Besides, is there any way to obtain the deformation grids for the CVA using any of the the tps software? Thanks a lot! Amelia -- Lic. Amelia Chemisquy Instituto de Botánica Darwinion Casilla de Correo 22 B1642HYD, San Isidro Argentina Tel. (54 11) 4743 4800/4742 8534 [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]> www.darwin.edu.ar <http://www.darwin.edu.ar> -- 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
