Prev 1966 / 20628 Next

lmer, poisson family and mcmcsamp

Christopher David Desjardins

Tue, Feb 17, 2009 6:21 AM

I think if you have a large number of observations you can treat them as accurate. However, if your number of observations are small then your approximation of z is poor and your probably best off running MCMC or a bootstrap maybe? Is that you're wondering?

Amelie LESCROEL <lescroel at cebc.cnrs.fr> wrote:

Thanks Chris, I saw this one but I was wondering whether the situation 
changed or was about to change. What about Pr(>|z|)? Are they reliable?
Amelie

Christopher David Desjardins a ?crit :

Regarding mcmcsamp() see 
https://mailman.stat.ethz.ch/pipermail/r-sig-mixed-models/2008q4/001660.html specifically 


...  In particular mcmcsamp is not currently working in that branch.  
My goal is to get mcmcsamp stable and to get all the other parts 
working in this branch then release it for testing and finally release 
lme4-1.0.  The subtext in that message is that only the models I 
describe above will be fit by the lme4-1.0 package.


On Feb 16, 2009, at 9:59 AM, Amelie LESCROEL wrote:

Dear all,

Could someone explain to me why summary(lmer.object) is giving z 
values and Pr(>|z|) when the lmer.object was fitted with 
family=poisson? I used to fit linear mixed models with 
family=gaussian and link="identity" or link="log", then used 
mcmcsamp() and HPDinterval() for computing "confidence" intervals and 
considered that a fixed effect was significant when the 95% HPD 
interval did not include zero. Now, with family=poisson, when I use 
mcmcsamp(), I get the following:

set.seed(101); mcmcfm1 <- mcmcsamp(fm1, n=10000)

Error in .local(object, n, verbose, ...) : Update not yet written

but I get z values and Pr(>|z|). Why is it so? Can we "trust" this 
probability for judging of the fixed effects significance? Would the 
mcmcsamp method be available soon for families other than gaussian?

Thanks in advance,

Am?lie

PS: When will lme4_1.0 be released? Many thanks to Douglas Bates for 
developing lme4 in interaction with users, and for sharing it. Please 
don't give up on making it available in R. lme4 is a wonderful tool 
which is helping me understanding penguin's foraging behaviour (and 
advanced stats!) as it is helping others in plant ecology or human 
psychology.

sessionInfo()

R version 2.7.2 (2008-08-25)
i386-pc-mingw32

locale:
LC_COLLATE=French_France.1252;LC_CTYPE=French_France.1252;LC_MONETARY=French_France.1252;LC_NUMERIC=C;LC_TIME=French_France.1252 


attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base
other attached packages:
[1] lme4_0.999375-28   Matrix_0.999375-17 lattice_0.17-17    gdata_2.4.2
loaded via a namespace (and not attached):
[1] grid_2.7.2   gtools_2.5.0 tools_2.7.2



__________ Information from ESET Mail Security, version of virus 
signature database 3857 (20090216) __________

The message was checked by ESET Mail Security.
http://www.eset.com

_______________________________________________
R-sig-mixed-models at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models