Hi Dave- > If I understand correctly, that's a separate issue than what Ted is > getting at; I think he's worried about the effect of data dredging by > running all possible models (as Burnham and Anderson warn against) and > not mis-parametrization of the simple model.
I don't think I'd call this data dredging. You are still comparing two models, but one has a shift parameter (the two-rate model) and one doesn't. Finding the maximum likelihood estimate of the shift parameter involves looking at the combined likelihood of the data partitions when the shift location is treated as a free parameter. However, assessing the *improvement in fit* of the 2 rate model over the simple 1 rate model is the trickier issue and may require simulation under the null hypothesis to evaluate correctly, as Joseph points out. ~Dan > Data-dredging will bias > us toward finding complex models that fit overly well. The common > solution in data mining is to subset your data and thus confirm a > preferred model for one subset by testing the preference in another > subset, but I don't think that can be applied to studies with a single > tree. > > -Dave > > On Tue, Mar 22, 2011 at 12:37 PM, Dan Rabosky <drabo...@berkeley.edu> wrote: >> >> Hi Ted- >> >> Others can weigh in here, but - at the very least - I would consider a model >> with 2 rates as having at least 1 additional parameter (the location of the >> rate shift). This seems to perform OK in practice. I admit that I have >> always been a bit uncomfortable with this, though - the null hypothesis >> model does not have a shift point. I think there are parallels here to >> finding breakpoints in segmented regression models. Here's a paper I've been >> meaning to read that might be relevant. >> >> Davies, R.B. (1987) Hypothesis testing when a nuisance parameter is present >> only under the alternative. Biometrika 74, 3343. >> >> ~Dan >> >> >> >> >> >> >> >> On Mar 22, 2011, at 10:19 AM, <tgarl...@ucr.edu> <tgarl...@ucr.edu> wrote: >> >>>> This can also be done using the package "diversitree", with the >>>> function make.bd.split and variants thereof. You will probably >>>> have to write a function to iteratively partition the tree >>>> into all possible splits. >>> >>> Hi Dan, >>> >>> How would this approach deal with the problem of multiple comparisons, >>> i.e., in effect testing many different hypotheses (or exploratory data >>> mining) on the same data set and overall tree? >>> >>> 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: Tue, 22 Mar 2011 09:44:26 -0700 >>> From: Dan Rabosky <drabo...@berkeley.edu> >>> Subject: Re: [R-sig-phylo] Q: Likelihood ratio test and rate shift >>> test on complete dated phylogeny >>> To: "R.J. den Tex" <rjden...@yahoo.com> >>> Cc: r-sig-phylo@r-project.org >>> >>>> >>>> Hi Robert- >>>> >>>>> (i) Can you use a Likelihood ratio test to test whether a birth >>> death model fits your data better than a pure birth model? >>> (likelihood scores are obtained from LASER). >>>> >>>> Yes, this would be valid, as the models are nested. >>>> >>>>> (ii) How can I test if a rate shift has occurred on a particular >>> node/branch in a complete taxon phylogeny? >>>> >>>> There are lots of possible methods for this. Or perhaps it is >>> better to say that there are lots of variants of a single general >>> approach. There are the 1-rate versus 2-rate model in LASER >>> (fitNDR_2rate, fitNDR_1rate) after Rabosky et al. (Proc. R. Soc. B, >>> 2007); there is the MEDUSA approach, which would allow more than 1 >>> rate shift across your tree (Alfaro et al., PNAS; this method is >>> implemented in GEIGER). >>>> >>>> This can also be done using the package "diversitree", with the >>> function make.bd.split and variants thereof. You will probably have >>> to write a function to iteratively partition the tree into all >>> possible splits. Actually, I think the diversitree calculations >>> *may* be the most accurate if you have different extinction rates >>> across the tree. >>>> >>>> There is also the method from Moore and Donoghue (2009, PNAS) - i >>> think tRate is the name of the package - although I don't think this >>> is distributed on the archive for R packages. >>>> >>>> ~Dan Rabosky >>>> >>>> >>>> >>>>> Thank you very much beforehand! >>>>> >>>>> Robert den Tex >>>>> PhD student >>>>> Dept. Evolutionary Biology >>>>> EBC, Uppsala University >>>>> Uppsala, >>>>> Sweden >>>>> >>>>> _______________________________________________ >>>>> R-sig-phylo mailing list >>>>> R-sig-phylo@r-project.org >>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-phylo >>>> >>>> >>>> >>>> >>>> >>>> [[alternative HTML version deleted]] >>>> >>>> _______________________________________________ >>>> R-sig-phylo mailing list >>>> R-sig-phylo@r-project.org >>>> https://stat.ethz.ch/mailman/listinfo/r-sig-phylo >> >> >> >> >> >> >> [[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/ [[alternative HTML version deleted]]
_______________________________________________ R-sig-phylo mailing list R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
