Re: [MORPHMET] PC Ranks in Evolutionary PCA
Thanks to everyone for the quick replies! My main question was about the phylogenetic PCA. Ryan: yes my data are landmark configurations after Procrustes superimposition. I understand allometry still has an effect after Procrustes, so I have regressed centroid size on shape and used the residuals. I have assumed all my specimens are adults so I suppose this is correction for static allometry? I didn't mention I'm using MorphoJ recommended by my supervisor. I will definitely try the morphometrics R packages if I have more analysis. MorphoJ is a little user-friendlier but cumbersome for some of what I'm doing. Many thanks! On Saturday, 28 March 2015 09:14:12 UTC, ryanfelice wrote: Hi Alex, I'm assuming that your shape data are landmark configurations that have been subjected to a Procrustes superimposition. is that correct? If so, I dont think the regression you described is really necessary- Procrustes analysis will remove the effects of size, orientation, and position. What software are you using for your analysis? you might find it easier/more streamlined to use the phyl.pca function in the phytools R package. If you use that function, make sure that you are using the original shape data and not the independent contrasts of shape data. Good luck! -Ryan Ryan N. Felice, PhD Ohio University Department of Biological Sciences 107 Irvine Hall Athens, OH 45701 www.rnfelice.com ryanf...@gmail.com javascript: (201)981-8642 On Fri, Mar 27, 2015 at 7:49 AM, Alex Marshall alpm...@googlemail.com javascript: wrote: Hello everyone, I'm a MSci student new to morphometrics and this group. I'm studying morphological integration in squamate crania and one of the things I'd like to do is an Evolutionary PCA of all my species, accounting for phylogeny and allometry. I think I do this correctly but in the PCA results the PCs are not ranked by proportion of variance e.g. PC5 has greater % variance that PCs 3 4. Is this normal? To conduct the evolutionary PCA I created independant contrasts of all my shape data, regressed centroid size on shape and conducted PCA on the residuals. After this I applied the resultant PC scores to another PCA of my original data. Would anyone kindly confirm if this is the right way to do it? Many thanks, Alex Marshall MSci Student University College London alexander@ucl.ac.uk javascript: -- MORPHMET may be accessed via its webpage at http://www.morphometrics.org To unsubscribe from this group and stop receiving emails from it, send an email to morphmet+u...@morphometrics.org javascript:. -- MORPHMET may be accessed via its webpage at http://www.morphometrics.org To unsubscribe from this group and stop receiving emails from it, send an email to morphmet+unsubscr...@morphometrics.org.
Re: [MORPHMET] PC Ranks in Evolutionary PCA
Hi Alex, Another type of allometry that is worth mentioned is the static allometry. So, if you want to remove this allometry effect (differences in shape in a same age group in adults, per example) you can use the same regression. Cheers Carlos On Sat, Mar 28, 2015 at 8:04 PM, oliverhoo...@prstatistics.co.uk wrote: Hi Alex, If you want to remove the effect of size on shape that is caused by ontogenetic allometry (differences in shape due to age/size per-say) you can regress the Procrustes coordinates of your samples on the centroid size and then use the residuals. Hope this helps, Oliver Hooker PR~Statistics On 28 March 2015, F. James Rohlf ro...@life.bio.sunysb.edu wrote: Procrustes superimposition only removes isometric size variation. Shape changes that are a function of size (allometry) are not removed so a regression of some sort is needed. F. James Rohlf, Distinguished Professor, Emeritus. Ecology Evolution Research Professor, Anthropology Stony Brook University -Original Message- From: Ryan Felice [mailto:ryanfel...@gmail.com] Sent: Friday, March 27, 2015 7:34 AM To: Alex Marshall Cc: morphmet@morphometrics.org Subject: Re: [MORPHMET] PC Ranks in Evolutionary PCA Hi Alex, I'm assuming that your shape data are landmark configurations that have been subjected to a Procrustes superimposition. is that correct? If so, I dont think the regression you described is really necessary- Procrustes analysis will remove the effects of size, orientation, and position. What software are you using for your analysis? you might find it easier/more streamlined to use the phyl.pca function in the phytools R package. If you use that function, make sure that you are using the original shape data and not the independent contrasts of shape data. Good luck! -Ryan Ryan N. Felice, PhD Ohio University Department of Biological Sciences 107 Irvine Hall Athens, OH 45701 http://www.rnfelice.com ryanfel...@gmail.com (201)981-8642 On Fri, Mar 27, 2015 at 7:49 AM, Alex Marshall alpmar...@googlemail.com wrote: Hello everyone, I'm a MSci student new to morphometrics and this group. I'm studying morphological integration in squamate crania and one of the things I'd like to do is an Evolutionary PCA of all my species, accounting for phylogeny and allometry. I think I do this correctly but in the PCA results the PCs are not ranked by proportion of variance e.g. PC5 has greater % variance that PCs 3 4. Is this normal? To conduct the evolutionary PCA I created independant contrasts of all my shape data, regressed centroid size on shape and conducted PCA on the residuals. After this I applied the resultant PC scores to another PCA of my original data. Would anyone kindly confirm if this is the right way to do it? Many thanks, Alex Marshall MSci Student University College London alexander.marshall...@ucl.ac.uk -- MORPHMET may be accessed via its webpage at http://www.morphometrics.org 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 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 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 To unsubscribe from this group and stop receiving emails from it, send an email to morphmet+unsubscr...@morphometrics.org. -- Carlos Eduardo R. D. Alencar - Biólogo, PhD. - Academia.edu Profile https://ufrn.academia.edu/CaducasRochaDuarte carlosce2...@gmail.com Grupo de Estudos em Ecologia e Fisiologia de Animais Aquáticos - GEEFAA [ www.geefaa.com] -- MORPHMET may be accessed via its webpage at http://www.morphometrics.org To unsubscribe from this group and stop receiving emails from it, send an email to morphmet+unsubscr...@morphometrics.org.
RE: [MORPHMET] PC Ranks in Evolutionary PCA
Hi Alex, If you want to remove the effect of size on shape that is caused by ontogenetic allometry (differences in shape due to age/size per-say) you can regress the Procrustes coordinates of your samples on the centroid size and then use the residuals. Hope this helps, Oliver Hooker PR~Statistics On 28 March 2015, F. James Rohlf ro...@life.bio.sunysb.edu wrote: Procrustes superimposition only removes isometric size variation. Shape changes that are a function of size (allometry) are not removed so a regression of some sort is needed. F. James Rohlf, Distinguished Professor, Emeritus. Ecology Evolution Research Professor, Anthropology Stony Brook University -Original Message- From: Ryan Felice [mailto:ryanfel...@gmail.com] Sent: Friday, March 27, 2015 7:34 AM To: Alex Marshall Cc: morphmet@morphometrics.org Subject: Re: [MORPHMET] PC Ranks in Evolutionary PCA Hi Alex, I'm assuming that your shape data are landmark configurations that have been subjected to a Procrustes superimposition. is that correct? If so, I dont think the regression you described is really necessary- Procrustes analysis will remove the effects of size, orientation, and position. What software are you using for your analysis? you might find it easier/more streamlined to use the phyl.pca function in the phytools R package. If you use that function, make sure that you are using the original shape data and not the independent contrasts of shape data. Good luck! -Ryan Ryan N. Felice, PhD Ohio University Department of Biological Sciences 107 Irvine Hall Athens, OH 45701 http://www.rnfelice.com ryanfel...@gmail.com (201)981-8642 On Fri, Mar 27, 2015 at 7:49 AM, Alex Marshall alpmar...@googlemail.com wrote: Hello everyone, I'm a MSci student new to morphometrics and this group. I'm studying morphological integration in squamate crania and one of the things I'd like to do is an Evolutionary PCA of all my species, accounting for phylogeny and allometry. I think I do this correctly but in the PCA results the PCs are not ranked by proportion of variance e.g. PC5 has greater % variance that PCs 3 4. Is this normal? To conduct the evolutionary PCA I created independant contrasts of all my shape data, regressed centroid size on shape and conducted PCA on the residuals. After this I applied the resultant PC scores to another PCA of my original data. Would anyone kindly confirm if this is the right way to do it? Many thanks, Alex Marshall MSci Student University College London alexander.marshall...@ucl.ac.uk -- MORPHMET may be accessed via its webpage at http://www.morphometrics.org 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 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 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 To unsubscribe from this group and stop receiving emails from it, send an email to morphmet+unsubscr...@morphometrics.org.
RE: [MORPHMET] PC Ranks in Evolutionary PCA
Procrustes superimposition only removes isometric size variation. Shape changes that are a function of size (allometry) are not removed so a regression of some sort is needed. F. James Rohlf, Distinguished Professor, Emeritus. Ecology Evolution Research Professor, Anthropology Stony Brook University -Original Message- From: Ryan Felice [mailto:ryanfel...@gmail.com] Sent: Friday, March 27, 2015 7:34 AM To: Alex Marshall Cc: morphmet@morphometrics.org Subject: Re: [MORPHMET] PC Ranks in Evolutionary PCA Hi Alex, I'm assuming that your shape data are landmark configurations that have been subjected to a Procrustes superimposition. is that correct? If so, I dont think the regression you described is really necessary- Procrustes analysis will remove the effects of size, orientation, and position. What software are you using for your analysis? you might find it easier/more streamlined to use the phyl.pca function in the phytools R package. If you use that function, make sure that you are using the original shape data and not the independent contrasts of shape data. Good luck! -Ryan Ryan N. Felice, PhD Ohio University Department of Biological Sciences 107 Irvine Hall Athens, OH 45701 www.rnfelice.com ryanfel...@gmail.com (201)981-8642 On Fri, Mar 27, 2015 at 7:49 AM, Alex Marshall alpmar...@googlemail.com wrote: Hello everyone, I'm a MSci student new to morphometrics and this group. I'm studying morphological integration in squamate crania and one of the things I'd like to do is an Evolutionary PCA of all my species, accounting for phylogeny and allometry. I think I do this correctly but in the PCA results the PCs are not ranked by proportion of variance e.g. PC5 has greater % variance that PCs 3 4. Is this normal? To conduct the evolutionary PCA I created independant contrasts of all my shape data, regressed centroid size on shape and conducted PCA on the residuals. After this I applied the resultant PC scores to another PCA of my original data. Would anyone kindly confirm if this is the right way to do it? Many thanks, Alex Marshall MSci Student University College London alexander.marshall...@ucl.ac.uk -- MORPHMET may be accessed via its webpage at http://www.morphometrics.org 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 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 To unsubscribe from this group and stop receiving emails from it, send an email to morphmet+unsubscr...@morphometrics.org.
Re: [MORPHMET] PC Ranks in Evolutionary PCA
Phylogenetic PCA has many unusual properties, and does not remove the effects of phylogeny. Yes, it is the case that the variance of the scores on the first phylogenetic principal component can be higher than the second PC. We recently had a paper in the Hystrix morphometrics volumes about the properties of pPCA and what it does and does not do. Polly, P. D., A. M. Lawing, A.-C. Fabre, and A. Goswami. 2013. Phylogenetic principal components analysis and geometric morphometrics. Hystrix, 24: 1-9 Paper is available open access from the publisher or from here: http://mypage.iu.edu/~pdpolly/Publications.html P. David Polly Geological Sciences Adjunct, Anthropology and Biology Indiana University 1001 E. 10th Street Bloomington, IN 47405-1405 pdpo...@indiana.edu +1 (812) 855-7994 http://mypage.iu.edu/~pdpolly/ On 27 Mar 2015, at 7:49 AM, Alex Marshall alpmar...@googlemail.com wrote: Hello everyone, I'm a MSci student new to morphometrics and this group. I'm studying morphological integration in squamate crania and one of the things I'd like to do is an Evolutionary PCA of all my species, accounting for phylogeny and allometry. I think I do this correctly but in the PCA results the PCs are not ranked by proportion of variance e.g. PC5 has greater % variance that PCs 3 4. Is this normal? To conduct the evolutionary PCA I created independant contrasts of all my shape data, regressed centroid size on shape and conducted PCA on the residuals. After this I applied the resultant PC scores to another PCA of my original data. Would anyone kindly confirm if this is the right way to do it? Many thanks, Alex Marshall MSci Student University College London alexander.marshall...@ucl.ac.uk -- MORPHMET may be accessed via its webpage at http://www.morphometrics.org 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 To unsubscribe from this group and stop receiving emails from it, send an email to morphmet+unsubscr...@morphometrics.org.