R2admb-package - R-SIG-mixed-models

Mon, Jul 14, 2014 7:28 PM #

[cc'ing to r-sig-mixed-models]

  Yes, sort of: see the caveats listed at
http://glmm.wikidot.com/faq#random-sig, in particular the last one (the
LRT is conservative for tests of null hypotheses where the parameter is
at the boundary). anova(m1,m2) should give you a likelihood ratio test
for the difference.

  glmmADMB uses the Laplace approximation to maximum likelihood estimation.

On 14-07-14 05:16 AM, Elena Guerrero wrote:

Dear Ben,

Is it possible to compare a glmmADMB model with a random term and and
the same model without that random term, with an ANOVA, in order to know
if the random term is significantly important? Does glmmADMB package use
ML method? I have seen (http://glmm.wikidot.com/pkg-comparison) that it
uses the Laplace estimation method, it is possible to make such
comparison with this method?

My model is of the following type:
Mugg_glmmADMB<-glmmadmb(Mugg ~ S10 + T10 + F10 + Prof+ offset (L.vol) +
(1|Est), link = "log",
                     data =tabla10m, zeroInflation=FALSE, family="nbinom")

Thank you very much in advance.
Best regards,
Elena

--
Elena Guerrero S?nchez-Guerrero
PhD student
Dept. Biologia Marina i Oceanografia
INSTITUTO DE CIENCIAS DEL MAR - CSIC
Pg. Mar?tim de la Barceloneta, 37-49
08003 BARCELONA
Spain
Phone:(+34) 93 230 95 00 (ext. 1209)
Fax: (+34) 93 230 95 55 
http://www.icm.csic.es/icmdivulga/es/mediterraneo-monograficos-08.htm
http://www.icm.csic.es/bio/

El 04/07/14 17:18, Elena Guerrero escribi?:

Great, thank you very much Ben. I've got it!

All the best,
Elena
--
Elena Guerrero S?nchez-Guerrero
PhD student
Dept. Biologia Marina i Oceanografia
INSTITUTO DE CIENCIAS DEL MAR - CSIC
Pg. Mar?tim de la Barceloneta, 37-49
08003 BARCELONA
Spain
Phone:(+34) 93 230 95 00 (ext. 1209)
Fax: (+34) 93 230 95 55 
http://www.icm.csic.es/icmdivulga/es/mediterraneo-monograficos-08.htm
http://www.icm.csic.es/bio/
El 03/07/14 00:42, Ben Bolker escribi?:

-2*logLik() is the deviance, so you should be able to fill that into
the formula.  To get the null deviance you will need something like
null.model <- update(full.model, [formula including only an intercept
term in the fixed effects]).  It is up to you to decide what a
'sensible' null model is -- i.e. whether it includes the random
effects or not ...


On Wed, Jul 2, 2014 at 9:23 AM, Elena Guerrero <eguerrero at icm.csic.es
<mailto:eguerrero at icm.csic.es>> wrote:

    Dear Ben,

    Thank you very much for your fast reply.

    I've tried: "-2*logLik(My glmmADMB)" and it gave me: 'log Lik.'
    1854.08 (df=7).

    How should I interpret this result? Is is possible to obtain a
    proportion % value of the explained deviance by the model?

    Thank you very much!
    All the best,
    Elena


    --
    Elena Guerrero S?nchez-Guerrero
    PhD student
    Dept. Biologia Marina i Oceanografia
    INSTITUTO DE CIENCIAS DEL MAR - CSIC
    Pg. Mar?tim de la Barceloneta, 37-49
    08003 BARCELONA
    Spain
    Phone:(+34) 93 230 95 00 <tel:%28%2B34%29%2093%20230%2095%2000>
    (ext. 1209)
    Fax: (+34) 93 230 95 55 <tel:%28%2B34%29%2093%20230%2095%2055>
    http://www.icm.csic.es/icmdivulga/es/mediterraneo-monograficos-08.htm
    http://www.icm.csic.es/bio/

    El 01/07/14 23:03, Ben Bolker escribi?:

          [cc'ing to R-sig-mixed-models]

           There appears to be a deviance() method for 'admb' objects
        (produced
        by R2ADMB), but not for 'glmmadmb" objects (produced by
        glmmADMB).
        However, you should be able to use -2*logLik(object) to
        obtain the
        deviance (to be precise, -2*logLik(object) is equal to the
        deviance in
        general only _up to an additive constant_, but I think this
        workaround
        will be OK for your purposes).

           Ben Bolker


        On 14-07-01 07:10 AM, Elena Guerrero wrote:

            Dear Ben Bolker,

            I have run a glmmADMB model to my count data with a
            family = "nbinom" .
            It worked very nice.

            Now I want to calculate the *deviance explained* by this
            model
            (100*(Null deviance-Residual deviance)/Null deviance). I
            installed the
            R2admb package to use the deviance() function, however
            when I write
            deviance(my model) R gives me "NULL" as a response. When
            I try with the
            "admbex" example it works perfectly.

            What is it happening? What does mean NULL in this case?
            Is there any
            other way to obtain the explained deviance of a glmmADMB
            model?

            Thank you very much in advance.
            Best regards,
            Elena

            -- 
            --
            Elena Guerrero S?nchez-Guerrero
            PhD student
            Dept. Biologia Marina i Oceanografia
            INSTITUTO DE CIENCIAS DEL MAR - CSIC
            Pg. Mar?tim de la Barceloneta, 37-49
            08003 BARCELONA
            Spain
            Phone:(+34) 93 230 95 00
            <tel:%28%2B34%29%2093%20230%2095%2000> (ext. 1209)
            Fax: (+34) 93 230 95 55
            <tel:%28%2B34%29%2093%20230%2095%2055>
            http://www.icm.csic.es/icmdivulga/es/mediterraneo-monograficos-08.htm
            http://www.icm.csic.es/bio/