8 Jan
2008
8 Jan
'08
5:27 p.m.
Hi Gregor,
You are correct that the predicted values should be draws from the
normal distribution. Thanks for catching this!
Additionally, the code benefits from your clean-up proposals ...
pulling definitions out of loops, adding stop() errors, updating
dependencies, etc. I appreciate your edits. My code can virtually
always benefit from an editor!
Finally, the log-normal support is a straightforward extension.
Thanks for the additional utility.
So, in short, thank you for your careful attention and contribution
to the public good.
Much appreciation,
Delia
************************************************
Delia Bailey
Washington University in St. Louis
School of Law, CERL
Campus Box 1120
One Brookings Drive
St. Louis, MO 63130
dbailey(a)wustl.edu
http://delia.wustl.edu
************************************************
Begin forwarded message:
From: Gregor Gorjanc
<gregor.gorjanc(a)bfro.uni-lj.si>
Date: January 8, 2008 1:03:25 PM EST
To: zelig(a)lists.gking.harvard.edu
Subject: [zelig] Re: Predicted values for lm.mixed + patches
Hi!
I fixed the issue with predicted values for lm.mixed plus some
other minor
tweaks. I uploaded the files as noted bellow. Please let me know
what you
think about these fixes/additions.
Regards, Gregor
----------------------------------------------------------------------
----
http://gregor.gorjanc.googlepages.com/qi.mixed.R
- fixed issue with predicted values
- added two errors i.e. stop() if family or link not provided by
qi.mixed.R is being used. Additionally, I added support for log-
normal
model to above file, so that one can use
z.out1 <- zelig(income ~ education + age + female + tag(1 | state),
data=voteincome, model="ls.mixed", family=gaussian
(link="log"))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
----------------------------------------------------------------------
----
http://gregor.gorjanc.googlepages.com/ls.mixed.Rnw
- added note about support for log-normal mixed model
----------------------------------------------------------------------
----
http://gregor.gorjanc.googlepages.com/param.mixed.R
- removed duplicated code so that it is easier to understand what
is done
----------------------------------------------------------------------
----
http://gregor.gorjanc.googlepages.com/gamma.mixed.Rnw
http://gregor.gorjanc.googlepages.com/logit.mixed.Rnw
http://gregor.gorjanc.googlepages.com/poisson.mixed.Rnw
http://gregor.gorjanc.googlepages.com/probit.mixed.Rnw
- changed Matrix package to lme4, since lmer is now back in lme4
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)lists.gking.harvard.edu
[un]subscribe Options: http://lists.gking.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
************************************************
Delia Bailey
Washington University in St. Louis
School of Law, CERL
Campus Box 1120
One Brookings Drive
St. Louis, MO 63130
dbailey(a)wustl.edu
http://delia.wustl.edu
************************************************
29 Jan
29 Jan
9:23 a.m.
New subject: Predicted values for lm.mixed + patches
Hello again!
I was bitten today by Zelig ls.mixed model, since due to the nature
(collinearity) of my data one coefficient (silently!) drops out from the model
fit. The dropout is done by lmer(), but Zelig's sim() can not work with such an
object.
qi.lmer is using the following code to get the coefficient names:
fTermsNames <- colnames(model.matrix(parsefml$fixed, data = D))
However, in my case above code gives one coefficient to many. I believe that
the following code is:
- more generic
- simpler
- more efficient as there is no need to set up the model.matrix and fixed
effects coefficient names are already in simpar or betas objects availale
in qi.lmer.
fTermsNames <- dimnames(betas)[[2]]
As previously I upload the changed qi.mixed.R to
http://gregor.gorjanc.googlepages.com/qi.mixed.R
I tested (R CMD check with 2.6.1) the package with modfified version of
qi.mixed and there is not problems. Btw. codetools says "qi.survreg: no
visible binding for global variable ‘arr.ind’".
-----------------------------------------------------------------------------
Very similar code as above is also used for random effects, but I was not able
to fully understand it and I did not dare to touch it.
Regards, Gregor
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)lists.gking.harvard.edu
[un]subscribe Options: http://lists.gking.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
10:15 a.m.
New subject: Predicted values for lm.mixed + patches
Thanks, Gregor. We are going to release a new version soon (hopefully by
the end of this week) with your contribution and other improvements.
Thanks for your patience!
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu/
---------------------------------------------------------
On Tue, 29 Jan 2008, Gregor Gorjanc wrote:
Hello again!
I was bitten today by Zelig ls.mixed model, since due to the nature
(collinearity) of my data one coefficient (silently!) drops out from the model
fit. The dropout is done by lmer(), but Zelig's sim() can not work with such an
object.
qi.lmer is using the following code to get the coefficient names:
fTermsNames <- colnames(model.matrix(parsefml$fixed, data = D))
However, in my case above code gives one coefficient to many. I believe that
the following code is:
- more generic
- simpler
- more efficient as there is no need to set up the model.matrix and fixed
effects coefficient names are already in simpar or betas objects availale
in qi.lmer.
fTermsNames <- dimnames(betas)[[2]]
As previously I upload the changed qi.mixed.R to
http://gregor.gorjanc.googlepages.com/qi.mixed.R
I tested (R CMD check with 2.6.1) the package with modfified version of
qi.mixed and there is not problems. Btw. codetools says "qi.survreg: no
visible binding for global variable ‘arr.ind’".
-----------------------------------------------------------------------------
Very similar code as above is also used for random effects, but I was not able
to fully understand it and I did not dare to touch it.
Regards, Gregor
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)lists.gking.harvard.edu
[un]subscribe Options: http://lists.gking.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
5955
days inactive
5976
days old
2 comments
3 participants
participants (3)
-
Delia Bailey -
Gregor Gorjanc -
Kosuke Imai