Hi David, I am not sure what you mean. If you add a tiny value to all of your data, then all you've done is shifted the mean ever so slightly. You then log-transform to put the variables on a scale unbounded by 0, and then you measure phylogenetic signal. The BM process models evolutionary changes as random draws from a normal distribution. So this would be the log-transformed values here. Adding a tiny value to all data doesn't change things very much. Log transformation changes things a lot more. So you're modeling a BM process on the log-transformed values. But there is nothing "illegal" about that. You do have to think about it, however.
This makes a lot of sense to me when I'm modeling body size evolution, because the model would imply that when you're small, evolutionary changes tend to be smaller than when you're big (evolutionary changes on larger animals would tend to have bigger "jumps" on the original scale). Ignoring any clade-specific biases, evolutionary changes in the body size of mice would tend to be a lot smaller than evolutionary changes in the size of horses, for example. We can imagine that some simple mechanistic reasons such as their subunits are smaller (they have smaller cells, smaller vertebrae, etc., so adding on or taking away would have bigger size effects). This seems biologically reasonable. I am not sure if it would apply to your spine size data, but maybe it does. Of course if your size variation is not that great, then it doesn't much matter (because log-trasformation will then not have much effect on compressing the scale of the values). Anyway, this may be a longer-winded comment than you were asking for. I can't think of any reason why adding tiny values would do anything to the evolutionary simulation. Marguerite On Mar 6, 2011, at 9:53 AM, David Bapst wrote: > Marguerite and others- > I had missed that Scales et al paper. Thank for pointing it out. I > have also had to deal with variables that were bounded by 0 and 1 > previously, and the use of the logit is something I'll have to try > when I go back to those analyses. > > Although I am interested in transformations of such traits for > comparative analyses in general (measuring correlated trait evolution, > model fitting of OU, measurements of rate, etc.), I am mainly > interested in measurements of phylogenetic signal. Is there any reason > to think the kludge (adding an arbitrarily small value to trait > values) would bias a test for phylogenetic signal (i.e. do the trait > show a pattern of inheritance)? I'm worried that measurements of > signal might be prone to odd results with the kludge, because the data > structure would still be quite different from BM expectation. Perhaps, > the threshold model might be of a lot of use here; I'm going to have > to look into that much more. > > Thanks for you advice, all! > -Dave > > > On Sun, Mar 6, 2011 at 12:50 AM, Marguerite Butler <mbut...@hawaii.edu> wrote: >> Hi David, Liam and everyone, >> >> Reflecting traits at boundaries or absorbing them is something that can be >> done, but I guess I'd like to encourage everyone to think carefully about >> the interpretation of such simulations. What are you trying to model and >> what does it mean at the end? Doing these boundary manipulations produces >> odd shaped-distributions of traits at the end of the simulated process. But >> probably more importantly, does this a good model for the biological process >> which might be occurring? If so, then great. But I would guess not with >> traits such as spine length. >> >> The zero issue is a hard problem. Is it a combination of a discrete trait >> (presence/absence) and a continuous trait (length)? Or perhaps a better >> model would be a threshold trait with underlying continuous variation but >> below some boundary it will not be expressed? Has anyone developed a model >> that can be used for such a scenario? In the absence of such a model, I've >> done things like what Enrico suggests, with a kludge solution like adding a >> very small value to all values to avoid the zero. This doesn't make much >> difference to the analysis, and avoids the singularity at log(0). If you >> have a trait that is a frequency, you can use the logit() function instead >> of log(), as we did in Scales et al 2009. >> >> Scales J.A., King A.A., and Butler M.A. (2009) Running for your life or >> running for your dinner: What drives fiber-type evolution in lizard >> locomotor muscles? Am. Nat.173:543-553. >> >> Marguerite >> >> On Mar 5, 2011, at 11:06 AM, Liam J. Revell wrote: >> >>> I'm not sure that this is what Dave has in mind, but if anyone is >>> interested in simulating bounded evolution in R, I just added it to my >>> "fastBM()" function (code here: >>> http://anolis.oeb.harvard.edu/~liam/R-phylogenetics/fastBM/v0.3/fastBM.R). >>> >>> In the process of evolving traits up the tree, I just bounce back any >>> phenotypes that exceed the lower or upper boundary conditions specified by >>> the user (by default they are -Inf and Inf). I think I did this properly. >>> Feedback welcome though. >>> >>> - Liam >>> >>> -- >>> Liam J. Revell >>> University of Massachusetts Boston >>> web: http://faculty.umb.edu/liam.revell/ >>> (new) email: liam.rev...@umb.edu >>> (new) blog: http://phytools.blogspot.com >>> >>> On 3/5/2011 12:55 PM, tgarl...@ucr.edu wrote: >>>> Hello David, Enrico, et al., >>>> >>>> I may have lost track of what Dave was originally trying to do, and I am >>>> not familiar with all of the options presently available in r for >>>> simulating continuously valued traits along a specified phylogenetic tree. >>>> However, I wanted to point out that MANY possibilities, including trends, >>>> the OU process, and actual limits to trait evolution implemented in >>>> several ways, are available in our original DOS program PDSIMUL.EXE that >>>> accompanies this paper: >>>> >>>> Garland, T., Jr., A. W. Dickerman, C. M. Janis, and J. A. Jones. 1993. >>>> Phylogenetic analysis of covariance by computer simulation. Systematic >>>> Biology 42:265-292. >>>> >>>> It has been used many times to look at trends, limits, etc., e.g., in >>>> these papers: >>>> >>>> Díaz-Uriarte, R., and T. Garland, Jr. 1996. Testing hypotheses of >>>> correlated evolution using phylogenetically independent contrasts: >>>> sensitivity to deviations from Brownian motion. Systematic Biology >>>> 45:27-47. >>>> Laurin, M. 2010. Assessment of the relative merits of a few methods to >>>> detect evolutionary trends. Syst. Biol. 59:689-704. >>>> >>>> Cheers, >>>> Ted >>>> >>>> >>>> >>>> Theodore Garland, Jr. >>>> Professor >>>> Department of Biology >>>> University of California, Riverside >>>> Riverside, CA 92521 >>>> Office Phone: (951) 827-3524 >>>> Lab Phone: (951) 827-5724 >>>> Home Phone: (951) 328-0820 >>>> Facsimile: (951) 827-4286 = Dept. office (not confidential) >>>> Email: tgarl...@ucr.edu >>>> >>>> Main Departmental page: >>>> http://www.biology.ucr.edu/people/faculty/Garland.html >>>> >>>> List of all Publications: >>>> http://www.biology.ucr.edu/people/faculty/Garland/GarlandPublications.html >>>> >>>> Garland and Rose, 2009 >>>> http://www.ucpress.edu/books/pages/10604.php >>>> >>>> >>>> ---- Original message ---- >>>> >>>> Date: Sat, 05 Mar 2011 15:36:13 +0100 >>>> From: Enrico Rezende<enrico.reze...@uab.cat> >>>> Subject: Re: [R-sig-phylo] Dealing with Bounded Trait Measures >>>> To: David Bapst<dwba...@uchicago.edu> >>>> Cc: R Sig Phylo Listserv<r-sig-phylo@r-project.org> >>>> >>>> >David, >>>> >on the top of my head, if no species measurement strictly >>>> corresponds to >>>> >zero, you may log-transform the data. You may then simulate >>>> Brownian >>>> >motion in log-transformed values, which will correspond to a >>>> boundary of >>>> >zero in a linear scale (i.e., the more negative the log number, the >>>> >closer the trait value is to zero - but never zero - in a linear >>>> scale). >>>> >This also explains why you can simulate the evolution of body mass >>>> >employing Brownian motion in log-transformed units and no species >>>> will >>>> >ever be assigned a body mass of zero. On more speculative grounds, >>>> this >>>> >may simply reflect the fact that many biological processes and >>>> their >>>> >regulation occur in a multiplicative, not additive, scale. >>>> > >>>> >The problem with regards to this approach is that you cannot really >>>> have >>>> >any species with a trait = 0 given that the log-transformation is >>>> >impossible in this case, so you might add some constant in case >>>> this >>>> >occurs (caution because the constant would be arbitrary and might >>>> have >>>> >an impact on the outcome of analyses). Did not think about this for >>>> too >>>> >long, though. >>>> > >>>> >Hope this helps, >>>> >Enrico >>>> > >>>> > >>>> > >>>> > >>>> > >>>> >El 4/3/11 9:14 p.m., David Bapst escribió: >>>> >> All- >>>> >> As far as I understand it, the vast majority of continuous >>>> character >>>> >> analyses assume that the trait is distributed normally and >>>> without >>>> >> bounds. Is there an appropriate transformation to for >>>> measurements of >>>> >> a trait that does have one or more bounds and where some taxa >>>> actually >>>> >> are at that bound? I have several traits where the bound is zero, >>>> and >>>> >> some taxa are actually at zero for this trait. (A practical >>>> example is >>>> >> 'spine length', where some taxa have virtually no spine.) And if >>>> there >>>> >> is no transformation applicable, is it analytically appropriate >>>> to >>>> >> remove taxa that have 'zero units' for that trait? Must we >>>> convert >>>> >> these traits to discrete categories to deal with them at all? >>>> >> >>>> >> As always, I appreciate your advice. >>>> >> -Dave Bapst, UChicago >>>> >> >>>> > >>>> > >>>> >-- >>>> >>>> >************************************************************************ >>>> >Enrico L. Rezende >>>> > >>>> >Departament de Genètica i de Microbiologia >>>> >Facultat de Biociències, Edifici Cn >>>> >Universitat Autònoma de Barcelona >>>> >08193 Bellaterra (Barcelona) >>>> >SPAIN >>>> > >>>> >Telephone: +34 93 581 4705 >>>> >Fax: +34 93 581 2387 >>>> >E-mail: enrico.reze...@uab.cat >>>> > >>>> >_______________________________________________ >>>> >R-sig-phylo mailing list >>>> >R-sig-phylo@r-project.org >>>> >https://stat.ethz.ch/mailman/listinfo/r-sig-phylo >>>> >>>> _______________________________________________ >>>> R-sig-phylo mailing list >>>> R-sig-phylo@r-project.org >>>> https://stat.ethz.ch/mailman/listinfo/r-sig-phylo >>> >>> _______________________________________________ >>> R-sig-phylo mailing list >>> R-sig-phylo@r-project.org >>> https://stat.ethz.ch/mailman/listinfo/r-sig-phylo >> >> ____________________________________________ >> Marguerite A. Butler >> Associate Professor >> Department of Zoology >> University of Hawaii >> 2538 McCarthy Mall, Edmondson 259 >> Honolulu, HI 96822 >> >> FAX: 808-956-9812 >> Dept: 808-956-8617 >> http://www2.hawaii.edu/~mbutler >> http://www.hawaii.edu/zoology/ >> >> >> [[alternative HTML version deleted]] >> >> >> _______________________________________________ >> R-sig-phylo mailing list >> R-sig-phylo@r-project.org >> https://stat.ethz.ch/mailman/listinfo/r-sig-phylo >> >> > > > > -- > David Bapst > Dept of Geophysical Sciences > University of Chicago > 5734 S. Ellis > Chicago, IL 60637 > http://home.uchicago.edu/~dwbapst/ ____________________________________________ Marguerite A. Butler Associate Professor Department of Zoology University of Hawaii 2538 McCarthy Mall, Edmondson 259 Honolulu, HI 96822 FAX: 808-956-9812 Dept: 808-956-8617 http://www2.hawaii.edu/~mbutler http://www.hawaii.edu/zoology/ [[alternative HTML version deleted]]
_______________________________________________ R-sig-phylo mailing list R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
