[ADMB Users] Question about admb

[Ben Bolker]

On 11-10-27 10:17 AM, bgeffroy wrote:
> I'm PhD student working on fish personality...
> I'm starting using admb package, and I would like to know if there is a
> possibility to extract the standard error of the random effect in R as
> it exist for the fixed effect in order to get an associated p-value ?
> 
  I believe you are using glmmADMB.

  if your fitted model is (say) MyModel, then MyModel$sd_S will give you
the standard error of the random effect.  However, getting a p-value on
a random effect in this way is extremely dicey -- there's a bit of
material (although not really enough) at http://glmm.wikidot.com/faq on
this topic ... basically, using the standard error on the variance to
test significance assumes that the estimate of the variance has normally
distributed sampling error (which is pretty far from true).  It would be
better if one wanted to do this to estimate the variance on the log
scale (which we might be able to do without terribly much trouble).  It
would be even better (although not without difficulties) to do a
likelihood ratio test (anova()) between models with and without the
random effect.  In general, however, testing the significance of random
effects in the the GLMM world is a bit more difficult and complicated
than it is in the old mixed-model ANOVA world ...

> 
> Coefficients:
>            Estimate Std. Error z value Pr(>|z|)   (Intercept) 
> -0.8901     0.1898   -4.69  2.7e-06 ***
> session      -0.3384     0.0791   -4.28  1.9e-05 ***
> ---
> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
> 
> Number of observations: total=1999, =61, =61
> Random effect variance(s):
> $ind
>            (Intercept)
> (Intercept)     0.85756
> 
> $ind
>         session
> session 0.043752
> 
> Thank you by advance
> 
> Cheers
> 
> ben
>