[R-sig-eco] [R-sig-phylo] GEE and AIC

tgarland at ucr.edu tgarland at ucr.edu
Fri May 30 02:18:37 CEST 2008

29 May 2008

Hello BriAnne,

	I thought I would pass on some comments.  I will assume that you have a continuous-valued (or at least quantitative) dependent variable.

> because I have a relatively small sample size

I don't understand why this would steer you away from contrasts (or the equivalent GLS-type approaches), as the power of them is identical to that of conventional statistics, assuming that you know the correct phylogeny in both cases (the former assumes the hierarchical tree you specify, whereas the latter assumes a star phylogeny [usually with contemporaneous tips, but not if you do weighted regression]).  In their simplest form, both phylogenetic and non-phylogenetic statistical methods assume, in effect, a Brownian-motion type model of character evolution.

> because I have ... a variety of continuous and multilevel
> discrete factors

This is not a problem for contrasts of GLS-type methods.  It can be a pain in the a** to code a whole bunch of categorical variables as dummy variables and then compute contrasts (depending on your software), but it is not a problem from the perspective of the math/stats.

	I'll now take this offline for a few additional comments and software suggestions related to Matlab and this paper:

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. In press.  [http://biology.ucr.edu/people/faculty/Garland/Lavin_et_al_2008.pdf]




Theodore Garland, Jr., Ph.D.
Department of Biology
University of California
Riverside, CA 92521
Phone:  (951) 827-3524 = Ted's office (with answering machine)
Phone:  (951) 827-5724 = Ted's lab
Phone:  (951) 827-5903 = Dept. office
Home Phone:  (951) 328-0820
FAX:  (951) 827-4286 = Dept. office
Email:  tgarland at ucr.edu
List of all publications with PDF files:
Associate Director
Network for Experimental Research on Evolution
(A University of California Multicampus Research Project)

  ---- Original message ----

    Date: Fri, 30 May 2008 10:00:27 +1000
    From: Simon Blomberg <s.blomberg1 at uq.edu.au>
    Subject: Re: [R-sig-phylo] [R-sig-eco] GEE and AIC
    To: BriAnne Addison <brianne.addison at gmail.com>
    Cc: phylo <r-sig-phylo at r-project.org>,
    r-sig-ecology at r-project.org

    >See Pan (2001). Akaike's information criterion in
    generalized estimating
    >equations. Biometrics 57: 120-125. See also the book by
    Hardin and Hilbe
    >2003 for examples and applications.
    >The only reason to jump to gee is if your dependent
    variable is not
    >Gaussian (but is an exponential family, e.g. binomial,
    Poisson, etc.),
    >and you have several covariates, some of which may be
    >You should be careful with your model of evolution if
    your dependent
    >variable is non-Gaussian. Brownian motion cannot apply to
    >variables. Instead, a discrete-state, continuous time
    Markov model might
    >be more appropriate, in which case you can't directly use
    the shared
    >branch-lengths of the phylogeny as estimates of
    phylogenetic covariance.
    >See Martins and Hansen 1997 Phylogenies and the
    comparative method: a
    >general approach to incorporating phylogenetic
    information into the
    >analysis of interspecific data. American Naturalist
    >Erratum 153:448. Tony Ives and Ted Garland have submitted
    a manuscript
    >on this topic, but you will have to contact them for
    further info. The
    >sample size has little to do with the issue, although
    fitting gee models
    >with small sample sizes can be extremely difficult. There
    are also
    >formidable problems with bias in the parameter estimates.
    >I've cc'ed this to r-sig-phylo, which is probably a
    better forum for
    >discussing comparative analyses.
    > Thu, 2008-05-29 at 15:50 -0700, BriAnne Addison wrote:
    >> This is tangential to the previous string.
    >> Has anyone calculated (pseudo)AIC values for
    generalized estimation
    >> equations in the GEE package (or related packages)? I
    wish to compare
    >> among several models where the data are not
    >> independent. I plan to use GEE to account for my
    phylogeny, rather
    >> than conventional contrast values, because I have a
    relatively small
    >> sample size and a variety of continuous and multilevel
    >> factors. The statistical problem is analogous to data
    which is
    >> spatially autocorrelated. Thoughts?
    >> Thanks,
    >> BriAnne
    >> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    >> BriAnne Addison
    >> Ecology, Evolution and Systematics
    >> Biology Department
    >> University of Missouri - St Louis
    >> brianne.addison at umsl.edu
    >> _______________________________________________
    >> R-sig-ecology mailing list
    >> R-sig-ecology at r-project.org
    >> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
    >Simon Blomberg, BSc (Hons), PhD, MAppStat.
    >Lecturer and Consultant Statistician
    >Faculty of Biological and Chemical Sciences
    >The University of Queensland
    >St. Lucia Queensland 4072
    >Room 320 Goddard Building (8)
    >T: +61 7 3365 2506
    >email: S.Blomberg1_at_uq.edu.au
    >1. I will NOT analyse your data for you.
    >2. Your deadline is your problem.
    >The combination of some data and an aching desire for
    >an answer does not ensure that a reasonable answer can
    >be extracted from a given body of data. - John Tukey.
    >R-sig-phylo mailing list
    >R-sig-phylo at r-project.org

More information about the R-sig-ecology mailing list