Hi, I had a comparable problem as Marko in the past, and I do think that Marko has a point. Indeed, partial warps cannot be used as homologous characters that are of any direct biological-evolutionary significance (James and Dennis convinced me of that in the Rome workshop). However, in the case of a dataset, where the goal is to discriminate between possible biological groups using shape variables like partial warps, this is to my opinion something else. Whether or not partial warps are less homologous for groups being analysed within the same analysis (thus partial warp 'space' is the same for all groups), than length measurements would be isn't even the issue here (I think). If the reason for using a discriminant function analysis is to search for that part of biological variation that is largest between groups, than why not reduce a dataset to those partial warp scores that just do that? Isn't a discriminant analysis doing exactly that in a way, that is, looking for those variables that allow you to discriminate between two or more groups? Dominique
Prof. Dr. Dominique Adriaens Ghent University Evolutionary Morphology of Vertebrates & Zoology Museum K.L. Ledeganckstraat 35, B-9000 Gent BELGIUM tel: +32 9 264.52.19, fax: +32 9 264.53.44 E-mail: [EMAIL PROTECTED] URL: http://www.fun-morph.ugent.be/ http://www.zoologymuseum.ugent.be/ -----Original Message----- From: morphmet [mailto:[EMAIL PROTECTED] Sent: donderdag 27 januari 2005 15:19 To: morphmet Subject: Re: use of partial warps in DA/CVA Hi, Thanks for the two answers (and one personal correspondence). But why so? Why should I choose uninformative points of outline instead of informative only? My logic says that differences in partial warps tell that there are between-species shape differences in corresponding outline points. But is this logic wrong? If MANOVA tells that there are differences for example in partial warps 1, 3, 7, 11 and 15, doesn't this mean that the corresponding landmarks differ too? I am aware about Rohlf's warnings (Syst, Biol.) on the use of single partial warps as homologous characters, but I am not yet convinced this apply to my case. I know I could use relative warps, but then the classification would not be that reliable. Personally, I find it difficult to believe that using a wrong method could produce better identifications by chance. In my case also cross-validated specimens (test specimens) were regularly classified correctly. In constructing phylogenies one cannot say whether the obtained phylogeny is closer to actual phylogeny than some other, but in my case I can be sure that by selecting "informative" partial warps I can get safer identifications. Sorry, I am not yet totally satisfied, Yours, Marko Mutanen At 11:28 27/01/2005, you wrote: >As to my knowledge, partial warps are not orthogonal, so only the complete >set of them makes a sense to analyze. If you wish to reduce the morphospace >dimensionality you better use relative warps. They are principal components >of the partial warps and hence are orthogonal. You may select several first >RWs, say 1st to 5th, which may appear enough to represent similarities among >your objects. >Cheers >Igor > >----- Original Message ----- >From: morphmet <[EMAIL PROTECTED]> >To: morphmet <morphmet@morphometrics.org> >Sent: Wednesday, January 26, 2005 6:41 PM >Subject: use of partial warps in DA/CVA > > > > Dear all, > > > > I have a question concerning the use of partial warp scores in >discriminant > > analysis (DA)/CVA. I have an outline data, including a couple of true > > landmarks and many sliding semilandmarks, on moth genitalia. I use this > > data in order to test how well it applies in categorizing study specimens > > into correct species (closely resembling each other in genitalia but not >in > > wing patterns) and I repeat this with several species groups. There are, > > however, too many partial warps for the DA since I have only 20-40 > > specimens for creating discriminant functions (the rest one-third of > > specimens are used for cross-validation). For this reason, I selected only > > those partial warp scores showing strong statistical differences between > > the species in MANOVA. I got pretty nice results since also those >specimens > > that were not used in creating discriminant functions were classified > > correctly. But is this correct? Can I select those partial warp scores >that > > differ in MANOVA and use them alone in DA/CVA? I could also reduce the > > number of sliding semilandmarks, but since I use outline data with sliding > > landmarks, this could possibly reduce the power of DA/CVA? > > > > I would be grateful for all advices, > > > > Marko Mutanen > > University of Oulu > > Finland > > [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
