From: Spencer Graves <spencer.graves_at_pdf.com>

Date: Sun 18 Jun 2006 - 02:46:41 EST

> Error in deviance(.Call("mer_coefGets", x, pars, 2, PACKAGE =

*> "Matrix")) :
*

*> Leading minor of order 2 in downdated X'X is not positive
*

*> definite
*

*>
*

*> Similar problem if I use quasibinomial.
*

*>
*

*> If I use glm, of course it thinks I have roughly twice the number of
*

*> subjects so the standard errors would be smaller than they should be.
*

*>
*

*> I used SAS's NLMIXED and it converged without problems giving me
*

*> parameter estimates close to what glm gives, but with larger standard
*

*> errors. glmmPQL(MASS) gives very different parameter estimates.
*

*>
*

*> Is it reasonable to fit a mixed effects model in this situation?
*

*>
*

*> Is there some way to give starting values for lmer and/or glmmPQL?
*

*>
*

*> Rick B.
*

*>
*

*> ______________________________________________
*

*> R-help@stat.math.ethz.ch mailing list
*

*> https://stat.ethz.ch/mailman/listinfo/r-help
*

*> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
*

R-help@stat.math.ethz.ch mailing list

https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Sun Jun 18 02:54:23 2006

Date: Sun 18 Jun 2006 - 02:46:41 EST

'lmer' RETURNS AN ERROR WHEN SAS NLMIXED RETURNS AN ANSWER

sessionInfo()

Version 2.3.1 (2006-06-01)

i386-pc-mingw32

attached base packages:

[1] "methods" "stats" "graphics" "grDevices" "utils" "datasets"
[7] "base"

lme4 lattice Matrix

"0.995-2" "0.13-8" "0.995-10"

When I ran your example limited only to the observations you listed, I got a different result:

lmer(response~(1|id),data=gdx,family=binomial)

**STARTING VALUES?
**
The help page for "lmer" describes the following argument:

start: a list of relative precision matrices for the random effects.

This has the same form as the slot '"Omega"' in a fitted model. Only the upper triangle of these symmetric matrices should be stored. To understand this, it might help to experiment with one ofthe examples on this help page:

> fm1 <- lmer(Reaction ~ Days + (Days|Subject), sleepstudy)

*> fm1@Omega
*

$Subject

2 x 2 Matrix of class "dpoMatrix"

(Intercept) Days

(Intercept) 1.0746247 -0.2942832

Days -0.2942832 18.7549595

Just taking a guess, I tried the following:

fm1. <- lmer(Reaction ~ Days+(Days|Subject), sleepstudy,

start=list(Subject=diag(1:2)))

fm1.c <- lmer(Reaction ~ Days+(Days|Subject), sleepstudy,

start=list(Subject=array(1, .1, .1, 2), dim=c(2,2))) Both of these returned the same answer. Starting values areonly required for the random effects, because these models are all "plinear" in the sense described in the "nls" documentation.

For "glmmPQL", I was able to get the "update" function to work. Thus, if you can get an answer with one model, you can use that as an input to "update". I would expect "update" to try to use the parameter estimates from one model fit as starting values for the modification, where it can find a way to do that.

Hope this helps. Spencer Graves

Rick Bilonick wrote:

> I'm using FC4 and R 2.3.1 to fit a mixed effects logistic regression.

*> The response is 0/1 and both the response and the age are the same for
**> each pair of observations for each subject (some observations are not
**> paired). For example:
**>
**> id response age
**> 1 0 30
**> 1 0 30
**>
**> 2 1 55
**> 2 1 55
**>
**> 3 0 37
**>
**> 4 1 52
**>
**> 5 0 39
**> 5 0 39
**>
**> etc.
**>
**> I get the following error:
**>
*

>> (lmer(response~(1|id)+age,data=gdx,family=binomial))

> Error in deviance(.Call("mer_coefGets", x, pars, 2, PACKAGE =

R-help@stat.math.ethz.ch mailing list

https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Sun Jun 18 02:54:23 2006

Archive maintained by Robert King, hosted by
the discipline of
statistics at the
University of Newcastle,
Australia.

Archive generated by hypermail 2.1.8, at Tue 20 Jun 2006 - 00:11:01 EST.

*
Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-help.
Please read the posting
guide before posting to the list.
*