Skip to content

glme output

2 messages · Daniel Frese, Ben Bolker

Original
Daniel Frese
# 
Begin forwarded message:

From: Ben Bolker <bbolker at gmail.com<mailto:bbolker at gmail.com>>
Subject: Re: glme output
Date: June 7, 2015 at 8:29:12 PM CDT
To: Daniel Frese <daf3366 at vet.k-state.edu<mailto:daf3366 at vet.k-state.edu>>

Could you please send this to r-sig-mixed-models at r-project.org<mailto:r-sig-mixed-models at r-project.org>?

 thanks
  Ben Bolker
On Sun, Jun 7, 2015 at 11:40 AM, Daniel Frese <daf3366 at vet.k-state.edu<mailto:daf3366 at vet.k-state.edu>> wrote:
Dr. Bolker,

I am in the process of learning R, and converting over from SAS.

I have the following model and I am having a difficult time finding code
that will get me the lumens output of the model.  What I am looking for is
an equivalent command to the SAS code of  LSMEANS within the PROC GLIMMIX.

Model is the following

fm14bf<-glmer(I(Steps+1)~Treatment*BFStrata*treathour  + time hour +
(1|Block) + (0+treatdate|Steer), family=poisson,data=datahourly)

summary(fm14bf,ddf="Kenward-Roger")
qqnorm(resid(fm14bf),main="QQ Model 14")
plot(fm14bf,main="Residual Model 14")
hist(residuals(fm14bf))


How would I go about getting LSMEANS and differences output from R

Thanks

Dan

Daniel Frese DVM
Graduate Teaching Assistant
Beef Cattle Institute
College of Veterinary Medicine
Kansas State University
Ben Bolker
# 
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 15-06-08 11:23 AM, Daniel Frese wrote:
[snip]
I would suggest that you take a look at the 'lsmeans' and 'effects'
packages for R -- I'm pretty sure they can both handle glmer models.

Some further questions/suggestions about your model:

* I'm not sure that summary(...,ddf="Kenward-Roger") is going to
do anything at all for a glmer model -- presumably you are using
the lmerTest package, which augments the summary method in this
way, but relies on the pbkrtest package, which does *not* implement
K-R for GLMMs (Stroup 2014 says that despite the lack of theoretical
grounding K-R *does* seem to give reasonable results for GLMMs).

* why are you using Steps+1 as your response variable in a Poisson model?
That seems odd -- a Poisson model *should* be able to handle Steps==0
responses perfectly well.  (I think you can actually get away without
I() for a model _response_)

* do you have a good reason to suppress the intercept variation among
Steers, i.e. are the starting values for each Steer constrained to
be exactly identical?

* you might want to consider adding an observation-level random effect
to allow for overdispersion ...

- ----
Stroup, Walter W. ?Rethinking the Analysis of Non-Normal Data in Plant
and Soil Science.? Agronomy Journal 106 (2014):
1?17. doi:10.2134/agronj2013.0342.








-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iQEcBAEBAgAGBQJVdfB0AAoJEOCV5YRblxUHkRYH/2iGN3ddo/Ji0X0Q7a4bqfLr
YmAH9kDCueqCYOcGoBF0GEkY+yOpAs0kxhDcpcfvCSwoKrAwUlUCL11f7RjOg7p6
NE5coZs78aEtjRUpSTWlVLveCgIYWKb9H6Zc6OsniVr105S/x1KFiGiz5mJKzGBS
LwxrKLnq522K8kOML9QfBoF0lfFXLHiYEBa2HovowETd1gMUzPXtshPBQRkwsywz
VHByJW+la1iL2Nb7cbDl6fU7eCD80V4dtkolNEXd9qV3T91UuaKRsI+Iz6n9/zJy
coO1KDXfo6Ev537UuN0FrZ0WbO0ybY1m9T8LVvFWeSacNUqbxfHpbzs/PIZ3F+I=
=IDIw
-----END PGP SIGNATURE-----