Dear Dr. Ibrahim, As you note, standard LDA does not usually permit nested groupings. In this scenario, you would either need to have one LDA for sex and one LDA for ancestry OR you could have one LDA that classifies into compound sex-ancestry groups (i.e., male-Malay, female-Malay, male-Chinese, female-Chinese, male-Indian, and female-Indian). Obviously, the latter option doubles the number of groups, which could have set backs depending on the smallest sample size of any sex-ancestry combination.
If you are feeling adventurous, my advisor, a collaborator, and I have had some success in classification problems with GM data that use two sequential discriminate function analyses (see this paper: https://onlinelibrary.wiley.com/doi/full/10.1002/jmor.20626, which uses a comparable procedure to this paper: http://digitallibrary.amnh.org/bitstream/handle/2246/4942/N3109.pdf?sequence=1). We did this to decide whether a specimen had enough information content to classify it to species, but a similar procedure could apply here. You would need three LDAs: one that classifies the specimen to sex, one that classifies male specimens to ancestry, and one that classifies female specimens to ancestry. To classify a test specimen, you would classify the individual to sex using the first LDA, then, based on this classification, you would classify to ancestry using the appropriate sex-specific LDA. Hope something in my ramblings helps you out, James On Mon, Apr 2, 2018 at 12:41 PM, Dr.Abdelnasser Ibrahim < dr.nasse...@gmail.com> wrote: > Dear Dr. Murat Maga, > > Thank you very much for your cooperation. It was very helpful. > > You already have 400 cases that you know the ancestry and sex, and you > want to use this database to infer ancestry of two skulls, for which you > already know the sex. All 400 cases, as well as your two unknown, specimens > have associated landmark data that were landmarked in using identical > protocols. Does that sum it up accurately? yes, this is right. > > I will derive LDA for sexual dimorphism but how can I derive it > for ancestry? as there are three ancestries; Malay, Chinese and Indian. > > Regards, > Abdelnasser > > On Mon, Apr 2, 2018 at 1:21 PM, Murat Maga <m...@uw.edu> wrote: > >> We need to clarify a few things: >> >> >> >> You already have 400 cases that you know the ancestry and sex, and you >> want to use this database to infer ancestry of two skulls, for which you >> already know the sex. All 400 cases, as well as your two unknown, specimens >> have associated landmark data that were landmarked in using identical >> protocols. Does that sum it up accurately? >> >> >> >> If that’s the case, one solution would be to use derive a linear >> discriminant function, and see how well it accurately identifies your >> ‘known’ data. You can use all your landmark coordinates, your full (or >> partial) PCA results to derive the LDA. At this point your goal would be to >> validate the model. You have a large sample size and you can split half to >> derive the LD function, and the remaining to test the LD (i.e., treat them >> as unknowns). Once you a find a model and a parameter set that give you >> good predictions, than you can plug in your two samples, and get your >> results. >> >> >> >> In my opinion this kind of analysis is simpler in R than any of the >> programs you mentioned below, since it requires frequent revising of the >> GPA fit (scale or not scale), use allometric regression (or not), use all >> derived coordinates (or not), etc… to get the best ‘tuned’ model for your >> existing data. >> >> >> >> There are other ways than LD to do this, but the structure of the problem >> is more or less the same: Derive a model, test it, if it results in >> sensible predictions, then apply to your unknowns. >> >> >> >> Hope this helps some… >> >> M >> >> >> >> >> >> >> >> >> >> >> >> *From:* Dr.Abdelnasser Ibrahim <dr.nasse...@gmail.com> >> *Sent:* Monday, April 2, 2018 4:11 AM >> *To:* firstname.lastname@example.org >> *Subject:* [MORPHMET] How can I test the project for identification of >> unknown crania? >> >> >> >> Dear all, >> >> >> >> I'm a Ph.D. student and I'm working with 3D CT scan geometric >> morphometric analysis for classification of known Malaysian crania to >> different sexes and ancestries. >> >> >> >> I collected the landmarks using Stratovan software and did Principal >> component, canonical variate, procrustes ANOVA, and DF analysis using >> MorphoJ and SPSS. and did visualization using IDAV Landmark Editor software >> and did cluster analysis using PAST software and used SPSS for reliability >> tests. >> >> >> >> I need to classify unknown crania. How can I test the project for >> identification of unknown crania? >> >> I did CT scan for known 2 crania one male and another female and known >> ancestries. Then I collected the landmarks on the tested crania using the >> same protocol. Then I added the collected coordinates to the whole >> previously collected coordinates of 400 cases using Notepad ++ then run PCA >> and CVA on MorphoJ. then the tested crania were classified to the groups >> which already known from the archive before. >> >> Is this method true for testing the unknown case? or I need to use >> another method? >> >> >> >> Regards, >> >> Abdelnasser Ibrahim >> >> Ph.D. student in Pathology department (Forensic Anthropology) >> >> Hospital UKM, Cheras, Malaysia. >> >> -- >> MORPHMET may be accessed via its webpage at http://www.morphometrics.org >> --- >> You received this message because you are subscribed to the Google Groups >> "MORPHMET" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to morphmet+unsubscr...@morphometrics.org. >> > > -- > MORPHMET may be accessed via its webpage at http://www.morphometrics.org > --- > You received this message because you are subscribed to the Google Groups > "MORPHMET" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to morphmet+unsubscr...@morphometrics.org. > -- MORPHMET may be accessed via its webpage at http://www.morphometrics.org --- You received this message because you are subscribed to the Google Groups "MORPHMET" group. To unsubscribe from this group and stop receiving emails from it, send an email to morphmet+unsubscr...@morphometrics.org.