[R-sig-ME] Basic question confidence intervals glmer

[Douglas Bates]

On Thu, Mar 24, 2011 at 11:20 AM, Phillip Chapman
<pchapman at stat.colostate.edu> wrote:
> Does the "confint-MASS" function from the MASS library do the profile
> likelihood confidence interval you are recommending?  From the documentation
> it appears to.  I just upgraded to R version 2.12.2.  In the previous
> version of R it was just called "confint", but apparently it was changed to
> avoid confusion.

Before you can create confidence intervals from the profiled deviance
you need to profile the deviance.  Profiling for the parameters in a
linear mixed-effects model is available in the development version of
lme4, currently called lme4a, but not for generalized linear mixed
models.

> On 3/24/2011 8:15 AM, Douglas Bates wrote:
>
> On Wed, Mar 23, 2011 at 7:31 PM, Sandra M. Durán <sduran at ualberta.ca> wrote:
>
> Hi, I want to evaluate if bird species respond strongly to management,
> vegetation or both. I am running different models with different species.
> Here is the overall formula here:
>
> print(fm1 <- glmer(birds ~ ownership + (1 | site), family=poisson,
> data=wma))
>
> And I want to know the syntax to get the confidence intervals (lower and
> upper limit) for the parameters in the fixed effects. I have used:
>
> confint(fm1), but I get an error message, although this code has worked for
> me when I used glm. It is possible to specify the confidence level(0.90,
> 0.95)?
>
> A way of getting approximate confidence intervals is to use the
> coefficients table from the summary for the model, as shown in the
> enclosed.
>
> These confidence intervals are approximations.  A more reliable
> approach would be to profile the deviance with respect to the
> parameters of interest but that has not yet been implemented.
>
> Note that in cases like this example and for your example confidence
> intervals on coefficients associated with contrasts for a factor may
> not be of too much interest by themselves, because they are expressed
> with respect to a somewhat arbitrary reference level for the factor.
> Also these do not take into account multiple comparisons.
>
>
>
> Thanks so much for any insights.
>
>
>
> The output I get is:
>
>
>
> Generalized linear mixed model fit by the Laplace approximation
>
> Formula: birds ~ ownership + (1 | site)
>
>   Data: wma
>
>   AIC   BIC logLik deviance
>
>  171.3 184.7 -81.63    163.3
>
> Random effects:
>
>  Groups Name        Variance Std.Dev.
>
>  site   (Intercept) 0.099663 0.31569
>
> Number of obs: 211, groups: site, 36
>
>
>
> Fixed effects:
>
>                     Estimate Std. Error z value Pr(>|z|)
>
> (Intercept)            0.1209     0.1497   0.807  0.41947
>
> ownershipCrownland     0.5362     0.1977   2.712  0.00669 **
>
> ownershipPrivateland   0.3902     0.1956   1.995  0.04608 *
>
> ---
>
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
>
>
> --
> Sandra M Durán
> University of Alberta
>
>        [[alternative HTML version deleted]]
>
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>