[R-sig-ME] heteroscedastic model in lme4
vito muggeo
vmuggeo at dssm.unipa.it
Thu Jan 15 13:21:29 CET 2009
dear Thierry,
I am adding a simple comment only on your second point.
If I am not wrong, I think that the two alternatives underlie different
models
1)glmer(.., family = poisson) assumes a real Poisson distribution for
your response y (conditioned to random effects), i.e. y=rpois(n,exp(mu)).
2) nlme(..) assumes a gaussian distribution for your response with a
nonlinear mean model, i.e. y=rnorm(n,exp(mu))
Another (different) approach would be lmer() with log-transformed data,
i.e. y=exp(rnorm(n,mu))
Probably, in a pure likelihood framework the first approach should be
preferred if you have real count data..
Hope this helps,
vito
ONKELINX, Thierry ha scritto:
> Dear all,
>
> I would like to analyse some spatial data with mixed model. As I'm
> dealing with presence/absence data or counts I should use the bionomial
> or poisson family. These families are implemented in lme4 but
> correlation structures are not. I'm wondering if the steps from section
> 5 in Pinheiro and Bates can be applied in case of a GLMM. If one can do
> that, should one apply the transformation on the response in the
> original scale or the transformed (logit / log) scale?
>
> Another, more approximate, solution might be to code the GLMM as a NLMM.
> E.g. glmer(Count ~ A + B + (1|Group), family = poisson) versus
> nlme(model = Count ~ exp(mu), fixed = mu ~ A + B, random = mu ~ Group)
> Any ideas on that?
>
> Thierry
>
> ------------------------------------------------------------------------
> ----
> ir. Thierry Onkelinx
> Instituut voor natuur- en bosonderzoek / Research Institute for Nature
> and Forest
> Cel biometrie, methodologie en kwaliteitszorg / Section biometrics,
> methodology and quality assurance
> Gaverstraat 4
> 9500 Geraardsbergen
> Belgium
> tel. + 32 54/436 185
> Thierry.Onkelinx at inbo.be
> www.inbo.be
>
> To call in the statistician after the experiment is done may be no more
> than asking him to perform a post-mortem examination: he may be able to
> say what the experiment died of.
> ~ Sir Ronald Aylmer Fisher
>
> The plural of anecdote is not data.
> ~ Roger Brinner
>
> 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
>
> -----Oorspronkelijk bericht-----
> Van: r-sig-mixed-models-bounces at r-project.org
> [mailto:r-sig-mixed-models-bounces at r-project.org] Namens Doran, Harold
> Verzonden: vrijdag 19 december 2008 20:52
> Aan: Alan Cobo-Lewis; r-sig-mixed-models at r-project.org
> Onderwerp: Re: [R-sig-ME] heteroscedastic model in lme4
>
> This isn't an entirely accurate statement. nlme has built-in functions
> that implement the methods for correlational and variance structures as
> described in section 5 of Pinhiero and Bates. lme4 doesn't have these
> functions built in as does nlme, but those same methods can be
> implemented by the user and then the data can be analyzed using
> functions in lme4. So, functions in lme4 can "handle" the same issues as
> nlme, it just requires the user to perform the steps described in PB
> section 5 et seq on their own.
>
>
>
>
> -----Original Message-----
> From: r-sig-mixed-models-bounces at r-project.org on behalf of Alan
> Cobo-Lewis
> Sent: Fri 12/19/2008 11:19 AM
> To: r-sig-mixed-models at r-project.org
> Subject: Re: [R-sig-ME] heteroscedastic model in lme4
>
>
> Anna,
>
> lme4 cannot handle certain kinds of heteroscedasticity, but I believe it
> can handle the kind you have in mind. Search the r-sig-mixed-models
> archive for a discussion involving me and David Afshartous, especially
> the summary message titled
> "[R-sig-ME] random effect variance per treatment group in lmer" that
> David posted 07/13/2007 04:18:08 PM
>
> I can't be certain that the suggestion below would work without knowing
> more about your design, but if width were a factor with three levels
> then you might try setting up indicator variables Wind1, Wind2, and
> Wind3 (that each take on the value 1
> when a site is at the indicator's target width and 0 otherwise) and then
> fit the model with something like
> mrem <- lmer( log(Nhat+1)~Group + GreenPerc + sess + crop + VegDensity +
> Group:sess + Group:VegDensity + (0+Wind1|site) + (0+Wind2|site) +
> (0+Wind3|site), data=all, method="REML" )
>
> alan
>
>
> r-sig-mixed-models at r-project.org on Friday, December 19, 2008 at 6:00 AM
> -0500 wrote:
>> Message: 1
>> Date: Thu, 18 Dec 2008 11:23:46 +0000
>> From: "Renwick, A. R." <a.renwick at abdn.ac.uk>
>> Subject: [R-sig-ME] heteroscedastic model in lme4
>> To: "'r-sig-mixed-models at r-project.org'"
>> <r-sig-mixed-models at r-project.org>
>> Message-ID:
>>
> <B9D1301370916C44B5874AF340C18B9B28AE890D50 at VMAILB.uoa.abdn.ac.uk>
>> Content-Type: text/plain; charset="us-ascii"
>>
>> I have been using the nlme package to run some LMM's, however I would
> like to try rerunning them using the lme4 package so that I can use mcmc
> sampling. The data I am using shows some heteroscesdasticity of the
> within error group and so I have
>> been using the 'weights' argument and the varIdent variance function
> structure to allow different variances for each level of my factor
> (patch width).
>> My problem is how to code for a heteroscedastic model in lme4 and any
> suggestion wouuld be much apprecaited.
>> The code I used in the nlme package:
>>
>> # model fit using "REML"
>> mrem<-lme(log(Nhat+1)~Group + GreenPerc + sess + crop + VegDensity +
> Group:sess + Group:VegDensity ,random=~1|Site, data=all,
>> method="REML",correlation=NULL,weights=varIdent(form=~1|width))
>>
>>
>> Many thanks,
>> Anna
>>
>> Anna Renwick
>> Institute of Biological & Environment Sciences
>> University of Aberdeen
>> Zoology Building
>> Tillydrone Avenue
>> Aberdeen
>> AB24 2TZ
>>
>>
>> The University of Aberdeen is a charity registered in Scotland, No
> SC013683.
>
>
> --
> Alan B. Cobo-Lewis, Ph.D. (207) 581-3840 tel
> Department of Psychology (207) 581-6128 fax
> University of Maine
> Orono, ME 04469-5742 alanc at maine.edu
>
> http://www.umaine.edu/visualperception
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
> Dit bericht en eventuele bijlagen geven enkel de visie van de schrijver weer
> en binden het INBO onder geen enkel beding, zolang dit bericht niet bevestigd is
> door een geldig ondertekend document. The views expressed in this message
> and any annex are purely those of the writer and may not be regarded as stating
> an official position of INBO, as long as the message is not confirmed by a duly
> signed document.
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>
--
====================================
Vito M.R. Muggeo
Dip.to Sc Statist e Matem `Vianelli'
Università di Palermo
viale delle Scienze, edificio 13
90128 Palermo - ITALY
tel: 091 6626240
fax: 091 485726/485612
http://dssm.unipa.it/vmuggeo
More information about the R-sig-mixed-models
mailing list