Hi Daniel, Offline, I'd like to see your actual tree and tip data.
Cheers, Ted ---- Original message ---- Date: Tue, 22 Mar 2011 15:17:02 -0500 From: dwe...@life.illinois.edu Subject: Re: [R-sig-phylo] Help Interpreting Phylogenetic ANOVA Results To: "Liam J. Revell" <liam.rev...@umb.edu> Cc: tgarl...@ucr.edu, r-sig-phylo@r-project.org >Hi Liam, > I've played around with your code and I see what you mean... >"tiny.edge" definitely influences whether it is significant or not. And >I re-ran my data playing with the edge length and that, too, is very >strongly influenced by the value of the edge length. And that makes me >very unsettled (and, frankly, a little depressed). > The only reason I put in an edge length is because, as far as I know, >the GEIGER program cannot work with branch lengths of zero. So, I >thought I could get around this by just putting in a nominal value. (I >hope my thought process makes sense). > Now, I don't know what to do. Is this analysis completely invalid? >Does anyone have a suggestion? Here's my original research question >again: I want to know if differences in body shape (geometric >morphometrics, "condensed" into PC scores) in a family of fish is >influenced by habitat (water velocity- moving water vs. still water) >and/or phylogeny (i.e. more closely related species look similar in >shape). > Thanks. > > > > >-Daniel > > > >> Just to add a little to this discussion, Luke Harmon suggested to me >> off-list that adding arbitrarily short terminal branches to the tree >> might perhaps *guarantee* a significant result. Indeed this seems like >> it could be the case. For instance, try running the following code for >> smaller (or larger) values of "tiny.edge." [Comments give you an idea >> of what's being done.] - Liam >> >> require(geiger) >> # how long is your "tiny edge?" >> tiny.edge<-0.0001 >> # is there error in measuring tip species >> error.variance<-0.01 >> # create balanced tree >> tree<-stree(n=32,type="balanced") >> # create branch lengths >> tree$edge.length<-rep(1,62) >> # set all terminal branch lengths to zero >> tree$edge.length[tree$edge[,2]<=32]<-0 >> # simulate and add a little random error >> y<-sim.char(tree,as.matrix(1))[,,1]+sqrt(error.variance)*rnorm(32) >> # create a factor >> x<-as.factor(rep(c(0,1),16)); names(x)<-names(y) >> # add a tiny little bit to the terminal edges >> tree$edge.length[tree$edge[,2]<=32]<-tiny.edge >> # compute phylogenetic anova >> phy.anova(tree,y,x) >> >> >> -- >> Liam J. Revell >> University of Massachusetts Boston >> web: http://faculty.umb.edu/liam.revell/ >> email: liam.rev...@umb.edu >> blog: http://phytools.blogspot.com >> >> On 3/17/2011 5:32 PM, tgarl...@ucr.edu wrote: >>> Dear Daniel, >>> >>> OK, then I think your results make sense. You have, in fact, designed >>> a comparative study in which the power to detect a statistical effect >>> of Habitat (i.e., a group difference) will be higher if analyzed >>> phylogenetically than if analyzed in the conventional fashion assuming >>> a stat phylogeny. This point has been made in various places, >>> including: >>> >>> 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. >>> Vanhooydonck, B., and R. Van Damme. 1999. Evolutionary relationships >>> between body shape and habitat use in lacertid lizards. Evolutionary >>> Ecology Research 1:785-805. >>> Garland, T., Jr. 2001. Phylogenetic comparison and artificial selection: >>> two approaches in evolutionary physiology. Pages 107-132 in R. C. Roach, >>> P. D. Wagner, and P. H. Hackett, eds. Hypoxia: From Genes to the >>> Bedside. Advances in Experimental Biology and Medicine, volume 502. >>> Kluwer Academic/Plenum Publishers, New York. >>> >>> In addition, you have made the within-species (and hence between >>> habitat) branch lengths arbitrarily very short relative to the >>> among-species branches in the rest f your tree. Hence, if you do have >>> a Habitat effect, it would be very unlikely to have occurred via >>> Brownian motion character evolution. So, you phylogenetic P values >>> becomes (much) lower than your non-phylogenetic P value. >>> >>> As a reality check, I would run your analyses on a star phylogeny (a >>> giant hard polytomy) and see if you get P values that are the same >>> (actually, just very close to unless you do a lot of simulations) as >>> for a nonphylogenetic analysis. >>> >>> Cheers, >>> Ted >>> >>> >>> >>> Theodore Garland, Jr. >>> Professor >>> Department of Biology >>> University of California, Riverside >>> Riverside, CA 92521 >>> Office Phone: (951) 827-3524 >>> Wet Lab Phone: (951) 827-5724 >>> Dry Lab Phone: (951) 827-4026 >>> 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: Wed, 16 Mar 2011 15:38:54 -0500 >>> From: dwe...@life.illinois.edu >>> Subject: Re: [R-sig-phylo] Help Interpreting Phylogenetic ANOVA >>> Results >>> To: tgarl...@ucr.edu >>> Cc: r-sig-phylo@r-project.org >>> >>> >Hi Liam and Ted, >>> > My apologies for not writing sooner. I've been sick. >>> > First off, let me thank you both for helping to clarify what the >>> >phylogenetic ANOVA is and what it means. >>> > I thought I would explain a little more than I did in my initial >>> post >>> >since Ted seemed curious to know more. >>> > I am using geometric morphometrics to analyze body shape in a >>> family of >>> >freshwater fish (topminnows). I ran a PCA to "condense" the >>> >morphometric analyses into PC scores. I'm using the PC scores in >>> the >>> >phylogenetic ANOVA. Because I would like to know how phylogeny >>> >"effects" habitat-specific differences in body shape, I run the >>> >phylogenetic ANOVAs with the tips being a species-habitat >>> combination. >>> >Essentially, it looks like this: >>> > >>> > |---- Species A Habitat 1 >>> > | >>> > |------| >>> > | | >>> >---| |---- Species A Habitat 2 >>> > | >>> > | >>> > | >>> > | |---- Species B Habitat 1 >>> > | | >>> > |-----| >>> > | >>> > |---- Species B Habitat 2 >>> > >>> > The tree for these species was made in Mesquite and based off of >>> a >>> tree >>> >from a recent molecular phylogeny that came out (Whitehead 2010 >>> >Evolution paper). The tree for the phylogenetic ANOVA includes >>> branch >>> >lengths, obviously. Because I don't know if there is any genetic >>> >distance between the same species from the two different habitats, >>> I set >>> >the branch lengths to a really small value (0.00001) in R. >>> > All of the 21 species used to run the phylogenetic ANOVA have >>> >populations in both habitats, so in that sense it's "balanced". >>> Where >>> >the species fall on the tree is pretty scattered, throughout the >>> tree >>> >(the tree contains about 20 more species that I did not include in >>> this >>> >analysis, so I don't know if that matters). Just visually glancing >>> at >>> >it, there are two main clades in the tree and 14 come from one >>> clade and >>> >7 come from the other clade, so, if anything, I would think they'd >>> be >>> >more "clumped" than overdispersed. However, I could be wrong. >>> > I would be happy to share the real tree and my data with anyone >>> who is >>> >interested. >>> > >>> > >>> > >>> > >>> > >>> >-Daniel >>> > >>> > >>> > >>> >> Exactly as Liam wrote, except that it is usually uncommon to >>> find >>> real >>> >> data that are "overdispersed phylogenetically," so normally one >>> finds that >>> >> the P values when done phylogenetically are larger (less >>> significant) than >>> >> when done via conventional, non-phylogenetic ANOVA. I'd be >>> curious what >>> >> your data are and how they appear to be scattered across the >>> tips >>> of your >>> >> phylogeny. Are they really overdispersed? >>> >> >>> >> Cheers, >>> >> Ted >>> >> >>> >> Theodore Garland, Jr. >>> >> Professor >>> >> Department of Biology >>> >> University of California, Riverside >>> >> Riverside, CA 92521 >>> >> Office Phone: (951) 827-3524 >>> >> Wet Lab Phone: (951) 827-5724 >>> >> Dry Lab Phone: (951) 827-4026 >>> >> 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: Mon, 14 Mar 2011 16:22:58 -0400 >>> >> From: "Liam J. Revell"<liam.rev...@umb.edu> >>> >> Subject: Re: [R-sig-phylo] Help Interpreting Phylogenetic ANOVA >>> >> Results >>> >> To: dwe...@life.illinois.edu >>> >> Cc: r-sig-phylo@r-project.org >>> >> >>> >> >Hi Dylan. >>> >> > >>> >> >The way the phylogenetic ANOVA (sensu Garland et al. 1993; >>> Syst. >>> >> Biol.) >>> >> >works is by first computing a standard ANOVA, and then >>> comparing >>> >> the >>> >> >observed F to a distribution obtained by simulating on the >>> tree >>> >> under a >>> >> >scenario of no effect of x on y. This "accounts for" the tree >>> in >>> >> the >>> >> >sense that it attempts to account for the possibility that >>> species >>> >> may >>> >> >have similar y conditioned on x because x influences y; or >>> because >>> >> they >>> >> >share common history and are thus similar by virtue of this >>> history >>> >> (and >>> >> >not at all due to x) >>> >> > >>> >> >It is not particularly surprising that your P-value was lower >>> in >>> >> the >>> >> >phylogenetic ANOVA than in your regular ANOVA. In general, the >>> >> effect >>> >> >of the phylogenetic ANOVA on P depends on the distribution of >>> the >>> >> >factor, x. If x is clumped on the tree, than the P-value of a >>> >> >phylogenetic ANOVA will tend to be higher than a regular >>> ANOVA. >>> By >>> >> >contrast, if x is overdispersed phylogenetically, the P-value >>> of >>> >> the >>> >> >phylogenetic ANOVA will tend to be lower than the regular >>> ANOVA. >>> >> > >>> >> >I hope this is of some help. >>> >> > >>> >> >- Liam >>> >> > >>> >> >-- >>> >> >Liam J. Revell >>> >> >University of Massachusetts Boston >>> >> >web: http://faculty.umb.edu/liam.revell/ >>> >> >email: liam.rev...@umb.edu >>> >> >blog: http://phytools.blogspot.com >>> >> > >>> >> >On 3/13/2011 9:28 PM, dwe...@life.illinois.edu wrote: >>> >> >> Hi, >>> >> >> I am relatively new to phylogenetic methods. I'm hoping >>> someone >>> >> can >>> >> >> help me to understand my results. >>> >> >> I am working with a group of 21 species of fish. I want to >>> know >>> >> how >>> >> >> their habitat may influence body shape and whether >>> phylogenetic >>> >> >> relatedness may influence body shape as well. I performed a >>> >> >> phylogenetic ANOVA using the GEIGER package in R. My >>> "metric" >>> of >>> >> body >>> >> >> shape is a PC score. Here's the output: >>> >> >> >>> >> >> Standard ANOVA: >>> >> >> Analysis of Variance Table >>> >> >> >>> >> >> Response: td$data >>> >> >> Df Sum Sq Mean Sq F value Pr(>F) >>> >> >> group 1 4.01 4.0134 0.4595 0.5017 >>> >> >> Residuals 40 349.35 8.7337 >>> >> >> >>> >> >> >>> >> >> Phylogenetic p-value: 0.000999001 >>> >> >> >>> >> >> I'm a bit uncertain as to how to properly interpret the >>> result. I >>> >> think >>> >> >> my confusion is two fold: >>> >> >> >>> >> >> 1. I am not sure I am interpreting what the phylogenetic >>> p-value >>> >> means. >>> >> >> Am I correct in saying that the phylogenetic p-value >>> essentially >>> >> says >>> >> >> that, after "accounting for" phylogeny, the habitat has an >>> effect >>> >> on body >>> >> >> shape (PC1)? >>> >> >> >>> >> >> 2. I am confused as to why it goes from non-significant (in >>> the >>> >> standard >>> >> >> ANOVA) to significant (phylogenetic p-value)? Does it mean >>> that >>> >> the >>> >> >> habitat does not have an effect on body shape if you don't >>> >> consider >>> >> >> phylogenetic relatedness? >>> >> >> >>> >> >> I realize these might be very simple questions but I'd >>> appreciate >>> >> it if >>> >> >> someone can help. I'm not well versed in phylogenetics, so >>> I >>> feel >>> >> a bit >>> >> >> lost. >>> >> >> Please feel free to respond directly to me at >>> >> dwe...@life.illinois.edu. >>> >> >> Thank you! >>> >> >> >>> >> >> >>> >> >> >>> >> >> >>> >> >> -Daniel >>> >> >> >>> >> >> >>> >> >> ----><((((º> -----><((((º> ----><((((º> ----><((((º> >>> >> ----><((((º> ---- >>> >> >> >>> >> >> Daniel P Welsh >>> >> >> University of Illinois at Urbana-Champaign >>> >> >> Champaign, IL, USA >>> >> >> >>> >> >> _______________________________________________ >>> >> >> 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 >>> >> >>> > >>> > >>> >----><((((º> -----><((((º> ----><((((º> ----><((((º> ---- >>> ><((((º> ---- >>> > >>> >Daniel P Welsh >>> >PhD Student >>> >Teaching Assistant >>> >Department of Animal Biology >>> >University of Illinois at Urbana-Champaign >>> >202 Shelford Vivarium >>> >606 E. Healey Street >>> >Champaign, IL 61821 >>> >lab phone: (217) 333-5323 >>> > >> > > >---- ><((((º> ----- ><((((º> ---- ><((((º> ---- ><((((º> ---- ><((((º> ---- > >Daniel P Welsh >PhD Student >Teaching Assistant >Department of Animal Biology >University of Illinois at Urbana-Champaign >202 Shelford Vivarium >606 E. Healey Street >Champaign, IL 61821 >lab phone: (217) 333-5323 > _______________________________________________ R-sig-phylo mailing list R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo