Hello, Murray:
I am doing the modifications you suggested yesterday, starting with the (instotal
+
weatherpc1|site) term. I was wondering if everytime I want to write an
interaction between instotal or weatherpc1 and any other variable, I
have to include the "|site" term. I mean, for example
(instotal|site):tlength. If I do that, I've seen that the software does
not interpret "instotal" and "weatherpc1" as fixed but random effects,
and that's wrong. Even when I fit a model without removing any
interaction, this way:
fledgecoltailmodel1<-lmer(nfledge~sex*briventral*(instotal + weatherpc1|site)*tlength+(site|pair),family=poisson)
I get this:
> summary(fledgecoltailmodel1)
Generalized linear mixed model fit by the Laplace approximation
Formula: nfledge ~ sex * briventral * (instotal + weatherpc1 | site) * tlength + (site | pair)
AIC BIC logLik deviance
87.91 119.7 -26.96 53.91
Random effects:
Groups Name Variance Std.Dev. Corr
site (Intercept) 0.55555555 0.745356
instotal 0.00016816 0.012968 0.000
weatherpc1 0.33505869 0.578843 0.000 0.000
pair (Intercept) 0.33333333 0.577350
site 0.00524246 0.072405 0.000
Number of obs: 48, groups: site, 10; pair, 6
Fixed effects:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -9.088e-01 7.907e+00 -0.115 0.908
sexM 3.312e-02 1.260e+01 0.003 0.998
briventral 1.653e-02 7.631e-02 0.217 0.828
tlength 2.726e-02 8.832e-02 0.309 0.758
sexM:briventral 1.400e-02 1.328e-01 0.105 0.916
sexM:tlength -5.034e-03 1.296e-01 -0.039 0.969
briventral:tlength -1.990e-04 8.690e-04 -0.229 0.819
sexM:briventral:tlength -9.434e-05 1.359e-03 -0.069 0.945
Correlation of Fixed Effects:
(Intr) sexM brvntr tlngth sxM:br sxM:tl brvnt:
sexM -0.670
briventral -0.965 0.662
tlength -0.995 0.675 0.973
sxM:brvntrl 0.574 -0.974 -0.610 -0.585
sexM:tlngth 0.715 -0.997 -0.710 -0.723 0.967
brvntrl:tln 0.955 -0.661 -0.997 -0.968 0.615 0.711
sxM:brvntr: -0.623 0.973 0.665 0.636 -0.996 -0.973 -0.672
I mean, "instotal" and "weatherpc1" are not included in the fixed effect interactions.
Is that correct?
Thank you very much.
Iker
--- El dom, 16/1/11, Murray Jorgensen escribió:
De: Murray Jorgensen
Asunto: Re: [R-sig-ME] Alternatives to lmer
Para: "Iker Vaquero Alba"
CC: r-sig-mixed-models@r-project.org
Fecha: domingo, 16 de enero, 2011 23:34
On 17/01/2011 11:22 a.m., Iker Vaquero Alba wrote:
>
>
> Let me check that I am interpreting your data correctly. It seems that
> you have 12 sites at each of which
you have a number of mated pairs. You
> have physical measurements on the male and the female of each pair and
> some environmental measurements on the sites.
>
> Yes, that's absolutely correct.
>
> The pairs are numbered within sites but some numbers are missing. Might
> they be the numbers corresponding to pairs that failed to fledge? I
> think maybe they should be there as well with nfledge = 0.
>
> Actually, the missing numbers correspond to the rows I have removed from
> the original data set because they contained "nfledge" missing values.
>
> 'instotal' and 'weatherpc1' are measured at the 'site' level, so the
> natural way to incorporate them in a model would be through a
>
> (instotal + weatherpc1|site)
>
> term.
>
> Would it be (instotal+weatherpc1|site) or
> (instotal|site+weatherpc1|site)? According to the way you've
written it,
> only "weatherpc1" seems to be in relation to "site".
In a model formula term ( .... |site) everything in the .... part is
at the site level.
>
> Surely also 'nfledge' should be at the 'pair' level ?
>
> I think it would then be better to re-shape the data so that each row is
> a pair and, for example, 'tlength' splits into two covariates 'tlengthM'
> and 'tlengthF'.
>
> I see the point, but what would be the difference compared to having one
> individual per row and sex as a factor, as it is now?
Well 'nfledge' depends only on the pair. The units at which the response
variable is measured must be the underlying units of the analysis.
>
> It looks doubtful to me that a Poisson model will fit this data well,
> with or without the addition of the nfledge = 0 pairs.
>
> Yes, the problem clearly seems
to be in teh fact that it's a Poisson
> model, as assuming Gaussian errors it doesn't return any error message.
But that does not mean that the analysis is correct, even in an
approximate way. I do not think your model formula reflects the
structure of the data, even if it runs without computing errors.
Murray
>
> I think that you should be talking more to some local statisticians
> before you attempt to fit models to this data.
>
>
> Regards, Murray
>
> Thank you so, so much for all your suggestions. I will work on it and
> tell you if it worked.
>
> Best wishes, Iker
>
> On 17/01/2011 9:15 a.m., Iker Vaquero Alba wrote:
> >
> > Hello:
> >
> > As I posted several days ago, I was trying to implement this model:
> >
> >
>
fledgecoltailmodel1<-lmer(nfledge~sex*briventral*inslarge*weatherpc1*tlength-sex:briventral:inslarge:weatherpc1:tlength-sex:briventral:inslarge:weatherpc1-sex:briventral:inslarge:tlength-sex:briventral:weatherpc1:tlength-sex:inslarge:weatherpc1:tlength-briventral:inslarge:weatherpc1:tlength-sex:briventral:inslarge-sex:briventral:weatherpc1-sex:briventral:tlength-sex:inslarge:weatherpc1-sex:inslarge:tlength-sex:weatherpc1:tlength-briventral:inslarge:weatherpc1-briventral:inslarge:tlength-briventral:weatherpc1:tlength-inslarge:weatherpc1:tlength+(site|pair),family=poisson)
> >
> > but I got the error message:
> >
> > Error in asMethod(object) : matrix is not symmetric [1,2]
> >
> > as no one seems to know what could be the reason for that or how to
> find a solution, I was thinking that maybe I could try using another
> function. Starting with the one
which seems more similar to "lmer", I
> tried with "GLMM", which I read in some post it could be taken from
> "lme4" package. However, when I try to use it (either "glmm" or "GLMM"),
> R tells me such function doesn't exist.
> >
> > Do you know why this could be happening? Do you know of any other
> functions I could use to fit my model? I was thinking as well of
> "MCMCglmm", but I'm not sure I can apply it to my model and I don't
> think I'm expert enough as to deal with its syntax or the overdispersion
> problems.
> >
> > I am using R 2.12.0
> >
> > Thank you very much for your help!
> >
> >
> >
> > Iker Vaquero-Alba
> >
> > Centre for Ecology
> > and Conservation
> >
> > Daphne du
Maurier
> > Building
> >
> > University of Exeter,
> > Cornwall Campus
> >
> > Treliever Road
> >
> > TR10 9EZ Penryn
> >
> > U.K.
> >
> >
> >
> >
> >
> >
> >
> > [[alternative HTML version deleted]]
> >
> >
> >
> >
> > _______________________________________________
> > R-sig-mixed-models@r-project.org
> mailing
list
> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>
> --
> Dr Murray Jorgensen http://www.stats.waikato.ac.nz/Staff/maj.html
> Department of Statistics, University of Waikato, Hamilton, New Zealand
> Email: maj@waikato.ac.nz
> majorgensen@ihug.co.nz Fax 7 838 4155
> Phone +64 7 838 4773 wk Home
+64 7 825 0441 Mobile 021 0200 8
>
>
--
Dr Murray Jorgensen http://www.stats.waikato.ac.nz/Staff/maj.html
Department of Statistics, University of Waikato, Hamilton, New Zealand
Email: maj@waikato.ac.nz majorgensen@ihug.co.nz Fax 7 838 4155
Phone +64 7 838 4773 wk Home +64 7 825 0441 Mobile 021 0200 8350
[[alternative HTML version deleted]]