[R-sig-ME] GLMM R package

[Ben Bolker]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 14-02-21 07:31 AM, Alexis Cerezo wrote:
> Dear Dr. Bolker,
> 
> My name is Alexis Cerezo, I´m a quantitative ecologist working in 
> Argentina and Guatemala.  I write to you because, in your paper in
> TREE from 2009 on GLMM´s (Generalized linear mixed models: a
> practical guide for ecology and evolution), you mention an R
> package called GLMM (table 1), which has the capability of running
> GLMM´s with temporal or spatial autocorrelation structure.  I´ve
> looked for it but was unable to find it on the web, perhaps you
> could point me in the right direction, or send the package to me.

  [I'm Cc'ing this to the r-sig-mixed-models mailing list: I don't
have any very helpful answers, but maybe someone else will ...]

  That's really weird -- I haven't looked at that table in a long
time, and I really don't know what I meant -- there are occasionally
packages that pop up but are then not maintained on CRAN, but I don't
know what I would have meant at the time.

> I understand glmmPQL runs glmm´s with spatial or temporal 
> autocorrelation, but you mention in your paper that under PQL 
> estimation, parameter estimates (for fixed effects?) are biased if
> the variance of random effects is large, so I want to compare the
> results.

   What is fairly well established is that *random-effects variance
estimates* are downwardly biased under PQL.  It's not clear in general
what effect this has on fixed-effect parameter estimates.

  I don't know of a solution for spatial GLMMs that is entirely
satisfactory (i.e., both statistically reliable and easy to use). I
could (and perhaps should) write a lot more about this; to my
knowledge there's not really a better *published* reference than the
Dormann et al 2009 paper referenced below.  Some possible clues to
follow up, and for others to comment on:

  Bayesian methods or approximate Bayesian methods -- INLA,, geoRglm

  Marginal likelihood approximations (I know Doug Bates doesn't like
this term but I don't know a better rubric): use gls() or lme() with
spatial structure and 'weights' argument to mimic the mean variance
relationship;  AD Model Builder (see the 'spatial examples' demo page)

  I ran across some discussion of copula-based approaches somewhere
recently.

  Does anyone else have useful ideas?

> I also came across an R script that you developed which calculates
> AIC (QAICc, actually) values from models derived with glmmPQL, but
> I can´t seem to find it again (I was looking for something else,
> and stumbled into it, but unfortunately didn´t save it), would you
> please send it to me?

  I'm not sure I ever managed to create that script -- I may have
talked about it, but the destruction of the likelihood and AIC slots
in glmmPQL is pretty thorough, so it's hard to get the necessary
information back.

  Again, others on the list are encouraged to chime in ...
> 
> Thanks so much, I hope I don´t take up too much of your time.
> 
> Best regards,
> 
> Alexis
> 
> -- Alexis Cerezo
> 
> Departamento de Métodos Cuantitativos y Sistemas de Información 
> Facultad de Agronomía-Universidad de Buenos Aires



@article{dormann_methods_2007,
	title = {Methods to account for spatial autocorrelation in the
analysis of species distributional data: a review},
	volume = {30},
	url = {http://dx.doi.org/10.1111/j.2007.0906-7590.05171.x},
	doi = {10.1111/j.2007.0906-7590.05171.x},
	abstract = {Species distributional or trait data based on range map
(extent-of-occurrence) or atlas survey data often display spatial
autocorrelation, i.e. locations close to each other exhibit more
similar values than those further apart. If this pattern remains
present in the residuals of a statistical model based on such data,
one of the key assumptions of standard statistical analyses, that
residuals are independent and identically distributed (i.i.d), is
violated. The violation of the assumption of i.i.d. residuals may bias
parameter estimates and can increase type I error rates (falsely
rejecting the null hypothesis of no effect). While this is
increasingly recognised by researchers analysing species distribution
data, there is, to our knowledge, no comprehensive overview of the
many available spatial statistical methods to take spatial
autocorrelation into account in tests of statistical significance.
Here, we describe six different statistical approaches to infer
correlates of species' distributions, for both presence/absence
(binary response) and species abundance data (poisson or normally
distributed response), while accounting for spatial autocorrelation in
model residuals: autocovariate regression; spatial eigenvector
mapping; generalised least squares; (conditional and simultaneous)
autoregressive models and generalised estimating equations. A
comprehensive comparison of the relative merits of these methods is
beyond the scope of this paper. To demonstrate each method's
implementation, however, we undertook preliminary tests based on
simulated data. These preliminary tests verified that most of the
spatial modeling techniques we examined showed good type I error
control and precise parameter estimates, at least when confronted with
simplistic simulated data containing spatial autocorrelation in the
errors. However, we found that for presence/absence data the results
and conclusions were very variable between the different methods. This
is likely due to the low information content of binary maps. Also, in
contrast with previous studies, we found that autocovariate methods
consistently underestimated the effects of environmental controls of
species distributions. Given their widespread use, in particular for
the modelling of species presence/absence data (e.g. climate envelope
models), we argue that this warrants further study and caution in
their use. To aid other ecologists in making use of the methods
described, code to implement them in freely available software is
provided in an electronic appendix.},
	number = {5},
	journal = {Ecography},
	author = {Dormann, Carsten F. and {Miguel B. Araújo} and Roger Bivand
and Janine Bolliger and Gudrun Carl and Richard G. Davies and
Alexandre Hirzel and Walter Jetz and W. Daniel Kissling and Ingolf
Kühn and Ralf Ohlemüller and Pedro R. Peres-Neto and Björn Reineking
and Boris Schröder and Frank M. Schurr and Robert Wilson},
	year = {2007},
	pages = {609--628}
}

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJTClRwAAoJEOCV5YRblxUHL90H/A8AHG0J3IZMIRNME5CfvC+5
vakqE3Ebk3nmK9lDxAApIVjHz3Hrz1+deZMleTh3NJ2lFs9kI7ng53By1qn4MHG3
h5C+w+R3y7qD32nlKeoa0iYBrHvCKQk7UmmzAqT1j0gxexSGSLRpfpyIstc5maeW
yXr1mAXSRfxVAsDN+7SMHw981GdhaZTQ1/1xNiV19r/sQjBu0NBV7TFUNkZ7gIwq
DZL5yiUlTiagu8SiScrql9/ywIRH38I29m1F663Bldl5WIQSuX0WRmBCDPJc0bN4
246lK+EXRWAb/d9qsNHn1pDX06UdmIpYFO2nhow7fUiEDP1UppnrI6tkIl0NaS8=
=5bow
-----END PGP SIGNATURE-----