Skip to content

Development version of lme4 now passes its tests

10 messages · Austin Frank, Bernd Weiss, Martin Maechler +4 more

#
The development version of the lme4 package on R-forge now passes its
tests with linear mixed models, nonlinear mixed models and generalized
linear mixed models.  It should also be capable of fitting generalized
nonlinear mixed models but I haven't written the user-callable
functions for those models yet.

This version represents a substantial redesign of the code (the last
one, I promise) and should be both faster and simpler than previous
versions.  I would recommend trying it if you have large data sets and
complex models (e.g. models with crossed or partially crossed grouping
factors) or if you are working with generalized linear mixed models or
nonlinear mixed models.  For example, Sharon Goldwater recently wrote
to the R-help list about peculiar relationships in her fitted GLMM
models (removing a parameter decreased the deviance) from the
currently released version of lme4.  She was kind enough to provide me
with the data and I could fit all her models, with consistent results,
in the development version of the package.

I'm still working on mcmcsamp so don't expect to get reasonable
results from that function yet.  (But soon, Ben, soon.)

This version of glmer allows the quasibinomial and quasipoisson
families.  I plan to add the Gamma family soon.  Are there any other
important families that I have missed?

I have added web pages documenting the C source code at
http://lme4.R-forge.R-project.org/doxygen.  In general you can use
http://lme4.R-forge.R-project.org/ as the 'home page' for the project.
 There is a link there to the project summary page.  You can browse
the source code for the package under the SCM (source code management)
tab on the project summary page.  The SCM tab provides information on
how you can access the svn repository.  Alternatively, you can wait
for the package builds to occur overnight (in Vienna) and install the
new version with

install.packages("lme4", repos = "http://r-forge.r-project.org")

I would appreciate reports of problems with this version.  Once I get
mcmcsamp running I will call this the alpha release of lme4_1.0.0 It
is much better to catch the bugs during the alpha and beta test phases
than to need to create bug-fix releases so please test it.

I think you should be able to submit bug reports directly if you have
an R-forge login.  There is a link to the Bugs tracker and the Feature
Request tracker on the project summary page.
#
Dr. Bates--

Hello!  Happy New Year!
I've asked on this list before[1], but since you've raised the issue
again, would it be possible to model a dependent variable drawn from a
multinomial distribution?  I've seen the relevant link function called
the "generalized logit".  That function and its log likelihood are
described at [2] and [3].

While the multinomial case is the one I'm most eager to see implemented
the, but there are related functions that could be considered.  Also
useful would be the generalized probit linking function ([4]) for
polychotomous choices, and the cumulative logit and probit linking
functions ([5]) for ordinal multinomial responses.

Thanks for considering these requests!
/au

Footnotes: 
[1]  http://thread.gmane.org/gmane.comp.lang.r.lme4.devel/230

[2]  http://www.statsoft.com/textbook/stglz.html#approach

[3]  http://support.sas.com/rnd/app/da/new/802ce/stat/chap8/sect5.htm

[4]  http://en.wikipedia.org/wiki/Multivariate_probit

[5]  http://www.asu.edu/sas/sasdoc/sashtml/stat/chap29/sect37.htm
#
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Douglas Bates schrieb:

[...]

| I have added web pages documenting the C source code at
| http://lme4.R-forge.R-project.org/doxygen.  In general you can use
| http://lme4.R-forge.R-project.org/ as the 'home page' for the project.
|  There is a link there to the project summary page.  You can browse
| the source code for the package under the SCM (source code management)
| tab on the project summary page.  The SCM tab provides information on
| how you can access the svn repository.  Alternatively, you can wait
| for the package builds to occur overnight (in Vienna) and install the
| new version with
|
| install.packages("lme4", repos = "http://r-forge.r-project.org")


I am sorry if I am missing anything but I was unable to download the
windows version of the development version of lme4. Neither
<install.packages("lme4", repos ...> nor the SCM approach
("win32-latest.zip") works properly.

Some of my datasets are very strange... with respect to multilevel
modelling, of course (N ~ 80.000, K = 15, binary dependent variable).
So, it would be interesting to (1) replicate former results. In
addition, (2) I would like to compare lmer results with MLWwiN and Stata
(xtlogit, gllamm, xtmelogit).

Regards,

Bernd

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFHgjSiUsbvfbd00+ERAsrwAJ4kxHLdn7i39V8Edo40zfIAoweyYQCcDb/m
v95Gz8Rw4RKLM3q4pnDtwmc=
=MX4w
-----END PGP SIGNATURE-----
#
BTW: 

Congratulations, Doug, on your lme4 "completion" !!
BW> -----BEGIN PGP SIGNED MESSAGE-----
    BW> Hash: SHA1

    BW> [...]

    BW> | I have added web pages documenting the C source code at
    BW> | http://lme4.R-forge.R-project.org/doxygen.  In general you can use
    BW> | http://lme4.R-forge.R-project.org/ as the 'home page' for the project.
    BW> |  There is a link there to the project summary page.  You can browse
    BW> | the source code for the package under the SCM (source code management)
    BW> | tab on the project summary page.  The SCM tab provides information on
    BW> | how you can access the svn repository.  Alternatively, you can wait
    BW> | for the package builds to occur overnight (in Vienna) and install the
    BW> | new version with
    BW> |
    BW> | install.packages("lme4", repos = "http://r-forge.r-project.org")


    BW> I am sorry if I am missing anything but I was unable to download the
    BW> windows version of the development version of lme4. Neither
    BW> <install.packages("lme4", repos ...> nor the SCM approach
    BW> ("win32-latest.zip") works properly.

Hmm, yes indeed; the link (on the SCM web page) goes to the
"build log" instead of to the zip file,
and the bin/windows/contrib/2.6/ or current/ directory is
completely empty.

===> a bug in the R-forge setup (or maintenance or ...),
     and I think we've seen the problem before.

--> CC'ing one of the R-forge maintainers..

Martin Maechler, ETH ZUrich
#
You're correct, Bernd, that the versions available by install.packages
from the repository at R-forge.R-packages.org appear to be out of
date.  Even the tar.gz file scm-latest.tar.gz is out of date
(apparently from January 04).

Those who can build a package from the source files can obtain the
latest version from the SVN archive at R-forge (instructions are given
at the SCM tab on the main project page).  I have just submitted the
package to Uwe's win-builder.R-project.org site to create a binary
Windows package, which I will make available on my web site as
http://www.stat.wisc.edu/~bates/lme4-2008-01-06-wiin32.zip

Stefan: Will someone be able to check on what is happening to the
nightly builds?
On Jan 7, 2008 8:18 AM, Bernd Weiss <bernd.weiss at uni-koeln.de> wrote:
#
Douglas Bates wrote:
this is fixed. the permissions were wrong in the svn export directory. 
This was because of some experimenting a few days ago.

the latest binaries are now available as usual.

I'm sorry for the inconvenience caused.

Best,
Stefan
#
On Jan 7, 2008 8:54 AM, Douglas Bates <bates at stat.wisc.edu> wrote:
The Windows package is now available as

http://www.stat.wisc.edu/~bates/lme4-2008-01-06-win32.zip

(The URL I previously gave has win32 spelled wiin32 and won't work.)
#
Hello all,

I'm not sure that things are fixed. See below.
On Jan 7, 2008, at 10:44 AM, Stefan Theussl wrote:

            
> install.packages("lme4", repos = "http://r-forge.r-project.org")
Warning: unable to access index for repository http://r-forge.r-project.org/bin/macosx/universal/contrib/2.6
Warning message:
package 'lme4' is not available
 > sessionInfo()
R version 2.6.1 (2007-11-26)
i386-apple-darwin8.10.1

locale:
C

attached base packages:
[1] grid      splines   grDevices graphics  stats     utils      
methods   base

other attached packages:
  [1] Hmisc_3.4-3          coda_0.13-1          gmodels_2.14.1        
lme4_0.99875-9
  [5] Matrix_0.999375-3    lattice_0.17-4       anacor_0.9-0          
car_1.2-7
  [9] colorspace_0.95      fda_1.2.3            zoo_1.4-1             
scatterplot3d_0.3-25
[13] rgl_0.76             JGR_1.5-8            iplots_1.1-1          
JavaGD_0.4-3
[17] rJava_0.5-1

loaded via a namespace (and not attached):
[1] MASS_7.2-39    cluster_1.11.9 gdata_2.3.1    gtools_2.4.0    
tools_2.6.1

_____________________________
Professor Michael Kubovy
University of Virginia
Department of Psychology
USPS:     P.O.Box 400400    Charlottesville, VA 22904-4400
Parcels:    Room 102        Gilmer Hall
         McCormick Road    Charlottesville, VA 22903
Office:    B011    +1-434-982-4729
Lab:        B019    +1-434-982-4751
Fax:        +1-434-982-4766
WWW:    http://www.people.virginia.edu/~mk9y/
#
Michael Kubovy wrote:
Currently we don't build mac binaries.
The directory
http://r-forge.r-project.org/bin/macosx/universal/contrib/2.6
does not exist yet.

You may install the package from its sources in the meantime.

Best regards,
Stefan
8 days later
#
Douglas Bates wrote:

            
This is very exciting!

I would respectfully add the inverse gaussian as an important family (at 
least to us!).  I take it from the error messages I get when using the 
R-forge version that this family is not supported yet?

Thank you once again for all this remarkable work.

Best

Roger