[R] interpreting glmer results

[Bert Gunter]

Comment at end Below.

Bert Gunter
Genentech Nonclinical Biostatistics
 
 -----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On
Behalf Of Kingsford Jones
Sent: Monday, October 05, 2009 10:30 AM
To: Umesh Srinivasan
Cc: r-help
Subject: Re: [R] interpreting glmer results

On Mon, Oct 5, 2009 at 8:52 AM, Umesh Srinivasan
<umesh.srinivasan at gmail.com> wrote:
> Hi all,
[snip]
>
> Fixed effects:
>             Estimate Std. Error z value Pr(>|z|)
> (Intercept) -138.8423     0.4704  -295.1  < 2e-16 ***
> SpeciesCr     -0.9977     0.6259    -1.6  0.11091
> SpeciesDb     -1.2140     0.6945    -1.7  0.08046 .
> SpeciesHk     -2.0864     1.2134    -1.7  0.08553 .
> SpeciesPa     -2.6245     1.2063    -2.2  0.02958 *
> SpeciesPs      1.3056     0.4027     3.2  0.00119 **
> distancen    121.7170     0.3609   337.3  < 2e-16 ***
> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
> Correlation of Fixed Effects:
>          (Intr) SpcsCr SpcsDb SpcsHk SpecsP SpcsPs
> SpeciesCr -0.423
> SpeciesDb -0.391  0.295
> SpeciesHk -0.223  0.169  0.152
> SpeciesPa -0.222  0.170  0.153  0.088
> SpeciesPs -0.732  0.507  0.458  0.262  0.263
> distancen -0.648 -0.020 -0.002 -0.003 -0.006  0.085
>
> Here, clearly, distance from the tree has an effect, but I want to
> know whether the identity of the species influences seedling numbers
> in general. I am unable, however, to make much sense of the output.


As with other linear model type functions in R the summary method
returns tests based on a factor's contrasts (treatment by default,
comparing other levels to a baseline level).  To get an omnibus test
of a factor, one option is to create a model with and without the
factor and perform an LRT:

library(lme4)
example(glmer)
gm0 <- glmer(cbind(incidence, size - incidence) ~ 1 + (1 | herd),
family = binomial, data = cbpp)
anova(gm0, gm1)

> Also, what does correlation of fixed effects really tell me?
>

These can be of interest for inference (e.g. a confidence region for
two of the coefficients is an ellipse with eccentricity defined by
their correlation).

-- ... and if the correlations are "high" it tells you that your model may
be near unidentifiable = the model parameters may not be effectively
estimated from the data. To understand what "high", "near" and "effectively"
may mean for your data, CYLS ("Consult your local statistician")

(If you really wish to use sophisticated tools like glmer, you really need
to understand what you're doing. There is no guarantee of immunity from the
consequences of ignorance.)

-- Bert

hth,

Kingsford



> Many thanks for any help.
>
> Cheers,
> Umesh Srinivasan,
> Bangalore, India
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>

______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.