[R-sig-ME] Fwd: Could the random effect at the level of each observation be a trap?

[Jarrod Hadfield]

Hi Billy,

As Drew mentioned, if the data are binary then over-dispersion does  
not exist. There may well be heterogeneity at the observation-level  
not accounted for by the fixed/random effects, but the data would look  
the same irrespective of the magnitude of this heterogeneity. Because  
this "residual" variation cannot be estimated it is usual to set it to  
zero (i.e omit 1|resid).

The sampling correlation between the x coefficient and the intercept  
is not a problem. For example you could center x,  and the correlation  
should be close to zero.

Cheers,

Jarrod




Quoting Billy <billy.requena at gmail.com>:

> ---------- Forwarded message ----------
> From: Billy <billy.requena at gmail.com>
> Date: Fri, Dec 17, 2010 at 3:57 PM
> Subject: Re: [R-sig-ME] Could the random effect at the level of each
> observation be a trap?
> To: Drew Tyre <atyre2 at unl.edu>
>
>
> Hi Drew!
>
> Well, below I'm sending the summary of two corresponding models (just
> for simplicity, because I have at least 5 models including onlye the
> 'MaleID' random effect and other 5 including 'MaleID' + 'resid'
> randiom effects):
>
> m1 <- glmer (y~ x + (1|MaleID), family=binomial)
> m1.1 <- glmer (y~ x + (1|MaleID) + (1|resid), family=binomial)
>
>> summary(m1)
> Generalized linear mixed model fit by the Laplace approximation
> Formula: y ~ x + (1 | MaleID)
>  AIC  BIC logLik deviance
>  1012 1027 -502.9     1006
> Random effects:
>  Groups Name        Variance   Std.Dev.
>  MaleID (Intercept) 6.9038e-11 8.309e-06
> Number of obs: 1045, groups: MaleID, 464
>
> Fixed effects:
>            Estimate Std. Error z value Pr(>|z|)
> (Intercept)   -8.273      2.036  -4.063 4.84e-05 ***
> x                   1.192      0.354   3.366 0.000762 ***
> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
> Correlation of Fixed Effects:
>   (Intr)
> x  -0.999
>
>> summary(m1.1)
> Generalized linear mixed model fit by the Laplace approximation
> Formula: y~ x + (1 | MaleID) + (1 | resid)
>   AIC   BIC logLik deviance
>  619.4 639.2 -305.7    611.4
> Random effects:
>  Groups Name        Variance Std.Dev.
>  resid  (Intercept) 2089.1   45.707
>  MaleID (Intercept)    0.0    0.000
> Number of obs: 1045, groups: resid, 1045; MaleID, 464
>
> Fixed effects:
>            Estimate Std. Error z value Pr(>|z|)
> (Intercept)  -19.049     85.552  -0.223    0.824
> x                    1.281     14.844   0.086    0.931
>
> Correlation of Fixed Effects:
>   (Intr)
> x  -0.999
>
>
> We can see that the variance associated to the 'resid' random variable
> is really huge, but I don't really know if that's a problem or it's
> just expected due the nature of this individual observation-based
> variable.
> Furthermore, the correlation between fixed effects seems pretty
> strong, but again I don't know if it's a problem.
> Does anyone have an insight to help me?
>
> Thanks again
>
> Billy
>
>
> --
> Gustavo Requena
> PhD student - Laboratory of Arthropod Behavior and Evolution
> Universidade de São Paulo
> Correspondence adress:
> a/c Glauco Machado
> Departamento de Ecologia - IBUSP
> Rua do Matão - Travessa 14 no 321 Cidade Universitária, São Paulo -  
> SP, Brasil
> CEP 05508-900
> Phone number: 55 11 3091-7488
>
> http://ecologia.ib.usp.br/opilio/gustavo.html
>
>
>
> --
> Gustavo Requena
> PhD student - Laboratory of Arthropod Behavior and Evolution
> Universidade de São Paulo
> Correspondence adress:
> a/c Glauco Machado
> Departamento de Ecologia - IBUSP
> Rua do Matão - Travessa 14 no 321 Cidade Universitária, São Paulo -  
> SP, Brasil
> CEP 05508-900
> Phone number: 55 11 3091-7488
>
> http://ecologia.ib.usp.br/opilio/gustavo.html
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>



-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.