R help - Feb 2008 - p-value for fixed effect in generalized linear mixed model

Dear R-users,

I am currently trying to switch from SAS to R, and am not very familiar with R
yet, so forgive me if this question is irrelevant.

If I try to find the significance of the fixed factor "spikes" in a
generalized linear mixed model, with "site" nested within
"zone" as a random factor, I compare following two models with the
anova function:

model1<-lmer(aphids~spikes+(1|zone:site), method="ML",
family=quasipoisson)
model2<-lmer(aphids~(1|zone:site), method="ML",
family=quasipoisson)
anova(model1,model2)

This gives me a p< 2.2e-16 ***, concluding that "spikes" has a
highly significant effect on "aphids". However, when I look at the
summary of model1:

summary(model1)

I find a t-value for "spikes" of  -0.1166 which is really
insignificant...

When I try model1 in SAS with proc glimmix, corrected for overdispersion with
"random _residual_", it also gives a p-value for "spikes" of
0,985. So if "spikes" is not having a significant effect on
"aphids", then why the above mentionned p-value generated by anova in
R?

Can anyone explain this?

Please find the dataset in the attachment.

Many thanks beforehand,

Martijn.









-- 
Martijn Vandegehuchte 
Ghent University 
Department Biology 
Terrestrial Ecology Unit 
K.L.Ledeganckstraat 35 
B-9000 Ghent
telephone: +32 (0)9/264 50 84
e-mail: martijn.vandegehuchte at ugent.be 

website TEREC: www.ecology.ugent.be/terec
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: aphids.txt
Url:
https://stat.ethz.ch/pipermail/r-help/attachments/20080220/2d75cb9f/attachment.txt

The lme4 package is still in development.  I can't guarantee the
evaluation of the deviance for the quasipoisson family.  There is a
certain amount of mystery about the role of the dispersion parameter
and the null deviance in families like quasipoisson and quasibinomial.
 For the time being don't base tests on a value of the deviance for
such models.

On Feb 20, 2008 8:19 AM, Martijn Vandegehuchte
<martijn.vandegehuchte at ugent.be> wrote:
> Dear R-users,
>
> I am currently trying to switch from SAS to R, and am not very familiar
with R yet, so forgive me if this question is irrelevant.
>
> If I try to find the significance of the fixed factor "spikes" in
a generalized linear mixed model, with "site" nested within
"zone" as a random factor, I compare following two models with the
anova function:
>
> model1<-lmer(aphids~spikes+(1|zone:site), method="ML",
family=quasipoisson)
> model2<-lmer(aphids~(1|zone:site), method="ML",
family=quasipoisson)
> anova(model1,model2)
>
> This gives me a p< 2.2e-16 ***, concluding that "spikes" has a
highly significant effect on "aphids". However, when I look at the
summary of model1:
>
> summary(model1)
>
> I find a t-value for "spikes" of  -0.1166 which is really
insignificant...
>
> When I try model1 in SAS with proc glimmix, corrected for overdispersion
with "random _residual_", it also gives a p-value for
"spikes" of 0,985. So if "spikes" is not having a
significant effect on "aphids", then why the above mentionned p-value
generated by anova in R?
>
> Can anyone explain this?
>
> Please find the dataset in the attachment.
>
> Many thanks beforehand,
>
> Martijn.
>
>
>
>
>
>
>
>
>
> --
> Martijn Vandegehuchte
> Ghent University
> Department Biology
> Terrestrial Ecology Unit
> K.L.Ledeganckstraat 35
> B-9000 Ghent
> telephone: +32 (0)9/264 50 84
> e-mail: martijn.vandegehuchte at ugent.be
>
> website TEREC: www.ecology.ugent.be/terec
>
> ______________________________________________
> 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.
>
>