The data and the problem you describe are not unusual but they do not correspond to the type of data normally discussed on the morphmet discussion list (analyses of the relative positions of landmarks or the shapes of outline curves). Method of multivariate data analysis (phenetic numerical taxonomic methods) would be more appropriate.
You will first need to standardize the variables in some way so they are in comparable units. I would then suggest you try clustering and ordination methods to look for evidence of structure in your data. Since you have just 8 variables, the data will not be high-dimensional so an ordination using PCA should be a useful first step to provide an overall look at the data. Showing the results as a biplot will let you visually associate variation among specimens with the different variables. If not enough of the variation can be captured in 2 or 3 dimensions, then a non-metric MDS may be useful but you will give up the ability to make a biplot (although one can make a crude approximation). You could also try cluster analysis (UPGMA is a good place to start). You mention using CVA. That requires a priori grouping of specimens (species?) and you wish to test their distinctness rather than looking for clusters etc. as described above. It also assumes that the variation within groups is more or less normally distributed and with homogeneous covariance matrices. Neither are likely to be true for this type of data. It does not require equal sample sizes. ----------------------- F. James Rohlf, Distinguished Professor & Graduate Program Director State University of New York, Stony Brook, NY 11794-5245 www: http://life.bio.sunysb.edu/ee/rohlf > -----Original Message----- > From: morphmet [mailto:[EMAIL PROTECTED] > Sent: Tuesday, October 04, 2005 3:29 AM > To: morphmet > Subject: MDS, PCA, PCoA or CVA? > > Dear List, > > I followed the discussion of the prior task "PSW and MDS" and > it was very interesting what Fred Bookstein replied to the > outgoing posting regarding the use of MDS: > > "My rationale for using MDS instead of PCA is (1) uneven > sample sizes / missing data, (2) lack of replicate samples, > (3) a more general interest in overall patterns/distributions > in the data." > > Fred Bookstein replied: > "Your "rationale for using MDS" is actually the list of > reasons for using PCA, which in this specific context does > PRECISELY what you are asking MDS to do!" > > Well, this brings up a question that I could not answer > myself yet. Actually I'm not a professional biologist, but I > am very much interested in morphometrics as part of systematic work. > > I have a data set of a total of 70 specimens with eight > morphological characters recorded each. The number of > specimens in each assumed OTU is not equal though! > I have recorded ventrals, subcaudals, labials entering the > eye, suboculars, loreals, prefrontals, parietals, and the > presence/absence of a specific colour pattern (0= absent; 1=present). > Data were standard normalized (mean=0; st.deriv.=1). I > initially used NMDS to verify assumed OTUs and continued using PCoA. > > The question now is, can I use these methods for these data > to identify morphological distinctions (identify new > species/subspecies)? > > Thanks in advance, > Wulf > > --- > The data set looks something like this: > > [...] > 267 65 3 0 1 1 1 1 > 259 67 2 1 1,5 2 1 0 > 258 69 3 0 1 1 1 0 > 269 70 3 0 2 2 1 0 > 263 62 3 0 2 2 1 0 > 271 68 3 0 2,5 2 1 0 > 274 67 3 0 3 2 1 0 > 265 63 3 0 2 2 1 0 > 278 72 3 0 1 1 2 1 > 278 71 3 0 1 1 2 1 > 277 70 3 0 1 1 2 1 > 273 70 3 0 1 1 2 0 > 270 71 3 0 1 1 2 0 > 267 65 3 0 1 1 1 0 > 269 64 3 0 1 1 1 0 > [...] > > > -- > 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
