Re: [R-sig-phylo] trait correlations with PICs
Exactly. That's why I wrote: "A problem is that then the statistical model is unclear/confusing." But for nuisance variables in a multuple-regression type of model, I see no problem putting them on a star phylogeny if they are not phylogenetically distrubuted (e.g., assuming that a particular technique for measuring, say, home range size, does not tend to be applied in a clade-specific [biased] fashion). So far as I am aware, Nobody knows exactly jhow to do this, nor has tried to implement this, with PGLS or phylogenetic regression models! We mention this in Garland and Ives (2000). Cheers, Ted From: Alejandro Gonzalez [alejandro.gonza...@ebd.csic.es] Sent: Wednesday, April 17, 2013 9:38 AM To: Theodore Garland Jr Cc: Anne Kempel; r-sig-phylo@r-project.org Subject: Re: [R-sig-phylo] trait correlations with PICs Can I risk a question here, maybe a result of ignorance... "With IC, it is easier to employ different sets of branch lengths for different traits (e.g. Garland et al., 1992; Lovegrove, 2003; Rezende et al., 2004), which may be particularly useful when one trait does not actually show phylogenetic signal (e.g. Tieleman et al., 2003; Rheindt et al., 2004) and/or for traits that are only nuisance variables, such as details of measurement or calculation methods that differ among studies (e.g. Wolf et al., 1998; Perry and Garland, 2002; Rezende et al., 2004)." >From an evolutionary point of view, what does it mean when using different >sets of branch lengths for different traits? If we're testing co-evolution of >traits, I would tend to think - and I may very well be wrong - that the >interest would be in estimating a single rate describing this co-evolution and >by employing different branch lengths would we not loose the opportunity to >estimtate that rate? Cheers Alejandro On 17, Apr 2013, at 6:24 PM, Theodore Garland Jr wrote: For correlations per se, please check this paper: Ives, A. R., P. E. Midford, and T. Garland, Jr. 2007. Within-species variation and measurement error in phylogenetic comparative methods. Systematic Biology 56:252-270. And here comes my question: is it allowed to calculate only PICs based on my tree for those traits having a signal, and to calculate "pseudo-PICs" based on a star phylogeny (independence between species) for those traits having no signal? This would allow me to correct the "signal-traits" but leave the other ones more or less as they are. The benefit: I then can correlate all traits with each other. Obviously, it is possible to do this. A problem is that then the statistical model is unclear/confusing. I could not find any reference doing this, and I wonder whether this is legal? We have discussed this here: Garland, T., Jr., A. F. Bennett, and E. L. Rezende. 2005. Phylogenetic approaches in comparative physiology. Journal of Experimental Biology 208:3015-3035. Page 3032: "With IC, it is easier to employ different sets of branch lengths for different traits (e.g. Garland et al., 1992; Lovegrove, 2003; Rezende et al., 2004), which may be particularly useful when one trait does not actually show phylogenetic signal (e.g. Tieleman et al., 2003; Rheindt et al., 2004) and/or for traits that are only nuisance variables, such as details of measurement or calculation methods that differ among studies (e.g. Wolf et al., 1998; Perry and Garland, 2002; Rezende et al., 2004)." Cheers, Ted Theodore Garland, Jr., Professor Department of Biology University of California, Riverside Riverside, CA 92521 Office Phone: (951) 827-3524 Skype: theodoregarland Facsimile: (951) 827-4286 = Dept. office (not confidential) Email: tgarl...@ucr.edu http://www.biology.ucr.edu/people/faculty/Garland.html http://scholar.google.com/citations?hl=en&user=iSSbrhwJ Inquiry-based Middle School Lesson Plan: "Born to Run: Artificial Selection Lab" http://www.indiana.edu/~ensiweb/lessons/BornToRun.html From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Anne Kempel [kem...@ips.unibe.ch] Sent: Wednesday, April 17, 2013 5:37 AM To: r-sig-phylo@r-project.org Subject: [R-sig-phylo] trait correlations with PICs Dear all, I am dealing with plant traits of different plant species (28 plant species), and aim to correlate all traits with each other. I have a tree of my species, and some of those traits do have a phylogenetic signal, but most don't. Since I do not know, which trait depends on whom, I favour to do correlations instead of regressions - that's why I would like to use Phylogenetic independent contrasts (PIC) instead of PGLS. I have the impression that it does not make sense to calculate PICs for traits that do not have a signal. And here comes my question: is it allowed to calculate only PICs based on my tree for those traits having a signal, and to calculate "pseudo-PICs" based on a star
Re: [R-sig-phylo] trait correlations with PICs
Can I risk a question here, maybe a result of ignorance... > "With IC, it is easier to employ different sets of branch lengths > for different traits (e.g. Garland et al., 1992; Lovegrove, 2003; > Rezende et al., 2004), which may be particularly useful when > one trait does not actually show phylogenetic signal (e.g. > Tieleman et al., 2003; Rheindt et al., 2004) and/or for traits that > are only nuisance variables, such as details of measurement or > calculation methods that differ among studies (e.g. Wolf et al., > 1998; Perry and Garland, 2002; Rezende et al., 2004)." >From an evolutionary point of view, what does it mean when using different >sets of branch lengths for different traits? If we're testing co-evolution of >traits, I would tend to think - and I may very well be wrong - that the >interest would be in estimating a single rate describing this co-evolution and >by employing different branch lengths would we not loose the opportunity to >estimtate that rate? Cheers Alejandro On 17, Apr 2013, at 6:24 PM, Theodore Garland Jr wrote: > For correlations per se, please check this paper: > > Ives, A. R., P. E. Midford, and T. Garland, Jr. 2007. Within-species > variation and measurement error in phylogenetic comparative methods. > Systematic Biology 56:252-270. > >> And here comes my question: is it >> allowed to calculate only PICs based on my tree for those traits having >> a signal, and to calculate "pseudo-PICs" based on a star phylogeny >> (independence between species) for those traits having no signal? This >> would allow me to correct the "signal-traits" but leave the other ones >> more or less as they are. The benefit: I then can correlate all traits >> with each other. > > Obviously, it is possible to do this. > A problem is that then the statistical model is unclear/confusing. > >> I could not find any reference doing this, and I wonder >> whether this is legal? > > We have discussed this here: > > Garland, T., Jr., A. F. Bennett, and E. L. Rezende. 2005. Phylogenetic > approaches in comparative physiology. Journal of Experimental Biology > 208:3015-3035. > Page 3032: > "With IC, it is easier to employ different sets of branch lengths > for different traits (e.g. Garland et al., 1992; Lovegrove, 2003; > Rezende et al., 2004), which may be particularly useful when > one trait does not actually show phylogenetic signal (e.g. > Tieleman et al., 2003; Rheindt et al., 2004) and/or for traits that > are only nuisance variables, such as details of measurement or > calculation methods that differ among studies (e.g. Wolf et al., > 1998; Perry and Garland, 2002; Rezende et al., 2004)." > > Cheers, > Ted > > Theodore Garland, Jr., Professor > Department of Biology > University of California, Riverside > Riverside, CA 92521 > Office Phone: (951) 827-3524 > Skype: theodoregarland > Facsimile: (951) 827-4286 = Dept. office (not confidential) > Email: tgarl...@ucr.edu > http://www.biology.ucr.edu/people/faculty/Garland.html > http://scholar.google.com/citations?hl=en&user=iSSbrhwJ > > Inquiry-based Middle School Lesson Plan: > "Born to Run: Artificial Selection Lab" > http://www.indiana.edu/~ensiweb/lessons/BornToRun.html > > > > From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] > on behalf of Anne Kempel [kem...@ips.unibe.ch] > Sent: Wednesday, April 17, 2013 5:37 AM > To: r-sig-phylo@r-project.org > Subject: [R-sig-phylo] trait correlations with PICs > > Dear all, > > I am dealing with plant traits of different plant species (28 plant > species), and aim to correlate all traits with each other. I have a tree > of my species, and some of those traits do have a phylogenetic signal, > but most don't. Since I do not know, which trait depends on whom, I > favour to do correlations instead of regressions - that's why I would > like to use Phylogenetic independent contrasts (PIC) instead of PGLS. > > I have the impression that it does not make sense to calculate PICs for > traits that do not have a signal. And here comes my question: is it > allowed to calculate only PICs based on my tree for those traits having > a signal, and to calculate "pseudo-PICs" based on a star phylogeny > (independence between species) for those traits having no signal? This > would allow me to correct the "signal-traits" but leave the other ones > more or less as they are. The benefit: I then can correlate all traits > with each other. > > I could not find any reference doing this, and I wonder whether this is > legal? > > I am grateful for any comment and suggestion! > Thanks a lot in advance and best wishes from Switzerland, > Anne > > > -- > Dr. Anne Kempel > Institute of Plant Sciences > Altenbergrain 21 > 3013 Bern > Switzerland > > ___ > R-sig-phylo mailing list - R-sig-phylo@r-project.org > https://stat.ethz.ch/mailman/listinfo/r-sig-phylo > Searchable archive at http://ww
Re: [R-sig-phylo] trait correlations with PICs
For correlations per se, please check this paper: Ives, A. R., P. E. Midford, and T. Garland, Jr. 2007. Within-species variation and measurement error in phylogenetic comparative methods. Systematic Biology 56:252-270. >And here comes my question: is it >allowed to calculate only PICs based on my tree for those traits having >a signal, and to calculate "pseudo-PICs" based on a star phylogeny >(independence between species) for those traits having no signal? This >would allow me to correct the "signal-traits" but leave the other ones >more or less as they are. The benefit: I then can correlate all traits >with each other. Obviously, it is possible to do this. A problem is that then the statistical model is unclear/confusing. >I could not find any reference doing this, and I wonder >whether this is legal? We have discussed this here: Garland, T., Jr., A. F. Bennett, and E. L. Rezende. 2005. Phylogenetic approaches in comparative physiology. Journal of Experimental Biology 208:3015-3035. Page 3032: "With IC, it is easier to employ different sets of branch lengths for different traits (e.g. Garland et al., 1992; Lovegrove, 2003; Rezende et al., 2004), which may be particularly useful when one trait does not actually show phylogenetic signal (e.g. Tieleman et al., 2003; Rheindt et al., 2004) and/or for traits that are only nuisance variables, such as details of measurement or calculation methods that differ among studies (e.g. Wolf et al., 1998; Perry and Garland, 2002; Rezende et al., 2004)." Cheers, Ted Theodore Garland, Jr., Professor Department of Biology University of California, Riverside Riverside, CA 92521 Office Phone: (951) 827-3524 Skype: theodoregarland Facsimile: (951) 827-4286 = Dept. office (not confidential) Email: tgarl...@ucr.edu http://www.biology.ucr.edu/people/faculty/Garland.html http://scholar.google.com/citations?hl=en&user=iSSbrhwJ Inquiry-based Middle School Lesson Plan: "Born to Run: Artificial Selection Lab" http://www.indiana.edu/~ensiweb/lessons/BornToRun.html From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Anne Kempel [kem...@ips.unibe.ch] Sent: Wednesday, April 17, 2013 5:37 AM To: r-sig-phylo@r-project.org Subject: [R-sig-phylo] trait correlations with PICs Dear all, I am dealing with plant traits of different plant species (28 plant species), and aim to correlate all traits with each other. I have a tree of my species, and some of those traits do have a phylogenetic signal, but most don't. Since I do not know, which trait depends on whom, I favour to do correlations instead of regressions - that's why I would like to use Phylogenetic independent contrasts (PIC) instead of PGLS. I have the impression that it does not make sense to calculate PICs for traits that do not have a signal. And here comes my question: is it allowed to calculate only PICs based on my tree for those traits having a signal, and to calculate "pseudo-PICs" based on a star phylogeny (independence between species) for those traits having no signal? This would allow me to correct the "signal-traits" but leave the other ones more or less as they are. The benefit: I then can correlate all traits with each other. I could not find any reference doing this, and I wonder whether this is legal? I am grateful for any comment and suggestion! Thanks a lot in advance and best wishes from Switzerland, Anne -- Dr. Anne Kempel Institute of Plant Sciences Altenbergrain 21 3013 Bern Switzerland ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] trait correlations with PICs
Hi all, I will attempt some clarification. In the context of phylogenetic regression models, the three most common types of models are: 1. Ordinary Least Squares (OLS) = conventional statistics = assumes a star phylogeny with contemporaneous tips. 2. Phylogenetic Generalized Least Squares (PGLS) = GLS where you use a phylogenetic tree to compute the expected variance-covariance matrix of residuals. First developed by Grafen (1989) (using GLIM, an uncommon stats package). Later Martins and Hansen (1997), Garland and Ives (2000), others, Lavin et al. (2008): Grafen, A. 1989. The phylogenetic regression. Phil. Trans. Royal. Soc. Lond. B 326:119-157. Garland, T., Jr., and A. R. Ives. 2000. Using the past to predict the present: Confidence intervals for regression equations in phylogenetic comparative methods. American Naturalist 155:346-364. Lavin, S. R., W. H. Karasov, A. R. Ives, K. M. Middleton, and T. Garland, Jr. 2008. Morphometrics of the avian small intestine, compared with non-flying mammals: A phylogenetic perspective. Physiological and Biochemical Zoology 81:526-550. [provides Matlab Regressionv2.m, released as part of the PHYSIG package] Grafen called this the "standard regresison," which confused people (self included!). What he called the "phylogenetic regression" used a "transformation parameter" that he named rho and was purely statistical, not tied to a model of evolution. This method is not simply PGLS (see below), although people keep calling it that. We have a discussion of this in the Appendix to Lavin et al. (2008). It also gives the history of all these methods. As we say in the paper: "Because the foregoing analyses assume either no (OLS) or relatively strong (PGLS) phylogenetic signal, we also performed an analysis in which the strength of phylogenetic signal in the residual variation was estimated simultaneously with the regression coefficients (e.g., see Grafen 1989; Freckleton et al. 2002; Chown et al. 2007; Duncan et al. 2007)." The additional parameter that is estimated is Grafen's rho, Pagel's lambda, or what we call "d" when it is tied to an OU model (first suggested by Felsenstein 1988 and used in simulations by Garland et al. 1993) or "g" when it is tied to an ACDC model of residual character evoluton (see Blomberg et al. 2003). 3. Phylogenetic Regression with a ??? Transformation/Model assumed. This all started with Grafen (1989). He also dealt with soft polytomies (lack of knowledge about branching order), and this made his paper even more sophisticated but complicated and a bit hard to understand at the time (it was ahead of its time!). In Lavin et al. (2008) we write (page 544): "Deriving a sensible terminology to discuss estimation under these different transform models requires making a distinction between statistical models and estimation techniques used to fit models to data. All of these transform models are statistical models in that they describe a statistical distribution and its parameters. In contrast, OLS and GLS are estimation techniques. Several authors refer to one or more of the transform models as GLS models (or PGLS models, for phylogenetic GLS models), although GLS is an estimation procedure. This is particularly confusing because GLS actually cannot be used to estimate the parameters of these transform models because they all contain parameters in the variance-covariance matrix that must be estimated. To reduce confusion, our preference is to refer to transform models as distinct from the estimation approaches that can be applied to them. We will break this convention only when using the established monikers “OLS model” and “GLS model” (i.e., the Brownian motion model) because in these cases there is a one-to-one match between the structure of the model and the appropriate estimation technique." The Matlab Regressionv2.m program we provide uses REML to estimate parameters in the four transform models: Grafen’s rho, Pagel’s lambda, OU, and ACDC (OU and ACDC are the same if you have contempoiraneous tips). Note that the version of the OU model implemented in various programs may differ a bit! See also Butler and King work, O'Meara, and probably others! Enough rambling! Hope that helps!!! Cheers, Ted Theodore Garland, Jr., Professor Department of Biology University of California, Riverside Riverside, CA 92521 Office Phone: (951) 827-3524 Skype: theodoregarland Facsimile: (951) 827-4286 = Dept. office (not confidential) Email: tgarl...@ucr.edu http://www.biology.ucr.edu/people/faculty/Garland.html http://scholar.google.com/citations?hl=en&user=iSSbrhwJ Inquiry-based Middle School Lesson Plan: "Born to Run: Artificial Selection Lab" http://www.indiana.edu/~ensiweb/lessons/BornToRun.html From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Alejandro Gonzalez [alejandro.gonza...@ebd.csic.es] Sent: Wednesday, April 17, 2013 6:00 AM To: Anne Kempel Cc: r-sig-p
Re: [R-sig-phylo] trait correlations with PICs
Hello Anne, The problem of non-independence, from a statistical point of view, has to do with the residuals of the relationship, hence testing phylogenetic signal of the traits - and not the residuals - can be misleading. Liam Revell has a paper in Methods in Ecology and Evolution (I think) that deals with this issue. Labra et al 2009 also discuss this briefly in the introduction of their paper. Although GLS is based on regression the results can only be interpreted as correlations because we have no way of determining causality, and an advantage of PGLS is that it includes the lambda parameter which estimates the required "phylogenetic correction" (emphasis on the quotations) for the data (actually for the residuals of the model), also see Freckleton et al 2002 and Revell 2010. Cheers Alejandro On 17, Apr 2013, at 2:37 PM, Anne Kempel wrote: > Dear all, > > I am dealing with plant traits of different plant species (28 plant species), > and aim to correlate all traits with each other. I have a tree of my species, > and some of those traits do have a phylogenetic signal, but most don't. Since > I do not know, which trait depends on whom, I favour to do correlations > instead of regressions - that's why I would like to use Phylogenetic > independent contrasts (PIC) instead of PGLS. > > I have the impression that it does not make sense to calculate PICs for > traits that do not have a signal. And here comes my question: is it allowed > to calculate only PICs based on my tree for those traits having a signal, and > to calculate "pseudo-PICs" based on a star phylogeny (independence between > species) for those traits having no signal? This would allow me to correct > the "signal-traits" but leave the other ones more or less as they are. The > benefit: I then can correlate all traits with each other. > > I could not find any reference doing this, and I wonder whether this is legal? > > I am grateful for any comment and suggestion! > Thanks a lot in advance and best wishes from Switzerland, > Anne > > > -- > Dr. Anne Kempel > Institute of Plant Sciences > Altenbergrain 21 > 3013 Bern > Switzerland > > ___ > R-sig-phylo mailing list - R-sig-phylo@r-project.org > https://stat.ethz.ch/mailman/listinfo/r-sig-phylo > Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ __ Alejandro Gonzalez Voyer Post-doc Estación Biológica de Doñana Consejo Superior de Investigaciones Científicas (CSIC) Av Américo Vespucio s/n 41092 Sevilla Spain Tel: + 34 - 954 466700, ext 1749 E-mail: alejandro.gonza...@ebd.csic.es Web site (Under construction): Personal page: http://consevol.org/members/alejandro_combo.html Group page: http://consevol.org/people.html For PDF copies of papers see: http://csic.academia.edu/AlejandroGonzalezVoyer [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
[R-sig-phylo] Pathological case for the function 'brunch' of CAPER?
Dear R-enthusiats, I guess my question suits (hopefully) more to this mailing list. While using the regression fonction 'brunch' of CAPER (with R v2.15.4), in a simple case (binary variable Yes/No vs. a continuous variable) I ended with an unexplained error: Error in if (any(stRes > robust)) { : missing value where TRUE/FALSE needed I simplified my code so that you can run it, just copy everything in a directory and run source("analysis.R") > Code: http://iktp.tu-dresden.de/~prudent/Divers/R/analysis.R > Tree: http://iktp.tu-dresden.de/~prudent/Divers/R/vertebrates.tree > Data: http://iktp.tu-dresden.de/~prudent/Divers/R/data.txt The source of the error is the pruning, (particularly for these tips: "cavPor3", "myoLuc1") but after searching around I still have no clue of what is happening. Any hint is welcome! Thanks in advance, Xavier [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
[R-sig-phylo] trait correlations with PICs
Dear all, I am dealing with plant traits of different plant species (28 plant species), and aim to correlate all traits with each other. I have a tree of my species, and some of those traits do have a phylogenetic signal, but most don't. Since I do not know, which trait depends on whom, I favour to do correlations instead of regressions - that's why I would like to use Phylogenetic independent contrasts (PIC) instead of PGLS. I have the impression that it does not make sense to calculate PICs for traits that do not have a signal. And here comes my question: is it allowed to calculate only PICs based on my tree for those traits having a signal, and to calculate "pseudo-PICs" based on a star phylogeny (independence between species) for those traits having no signal? This would allow me to correct the "signal-traits" but leave the other ones more or less as they are. The benefit: I then can correlate all traits with each other. I could not find any reference doing this, and I wonder whether this is legal? I am grateful for any comment and suggestion! Thanks a lot in advance and best wishes from Switzerland, Anne -- Dr. Anne Kempel Institute of Plant Sciences Altenbergrain 21 3013 Bern Switzerland ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/