Skip to content

false convergence glmer when quadratic terms are included

3 messages · Stijn Ruiter, ONKELINX, Thierry, Douglas Bates

Original
Stijn Ruiter
# 
Dear all,
Somehow I always get false convergence warnings when including a
quadratic term in a glmer equation. With lmer I have no such problems.
So, my guess is that this has to do with the different algorithm used
for glmer models.
The following example model (which can easily be estimated in
alternative mixed effects programs such as MLwiN or HLM) result in problems:

glmer(y~(1|level2id)+x+I(x^2),family=binomial,data=data)

It leads to:
Warning message:
In mer_finalize(ans) : false convergence (8)

What is the problem here? Why these convergence issues?

Stijn
ONKELINX, Thierry
# 
Dear Stijn,

Why don't you use poly(x, 2) instead? Adding x^2 with large or small
values will lead to even larger or smaller values. That may cause the
model to become unstable. Poly(x, 2) will avoid that by rescaling.

glmer(y~(1|level2id)+poly(x,2),family=binomial,data=data) 

HTH,

Thierry

------------------------------------------------------------------------
----
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature
and Forest
Cel biometrie, methodologie en kwaliteitszorg / Section biometrics,
methodology and quality assurance
Gaverstraat 4
9500 Geraardsbergen
Belgium
tel. + 32 54/436 185
Thierry.Onkelinx at inbo.be
www.inbo.be

To call in the statistician after the experiment is done may be no more
than asking him to perform a post-mortem examination: he may be able to
say what the experiment died of.
~ Sir Ronald Aylmer Fisher

The plural of anecdote is not data.
~ Roger Brinner

The combination of some data and an aching desire for an answer does not
ensure that a reasonable answer can be extracted from a given body of
data.
~ John Tukey

-----Oorspronkelijk bericht-----
Van: r-sig-mixed-models-bounces at r-project.org
[mailto:r-sig-mixed-models-bounces at r-project.org] Namens Stijn Ruiter
Verzonden: dinsdag 19 mei 2009 12:18
Aan: r-sig-mixed-models at r-project.org
Onderwerp: [R-sig-ME] false convergence glmer when quadratic terms are
included

Dear all,
Somehow I always get false convergence warnings when including a
quadratic term in a glmer equation. With lmer I have no such problems.
So, my guess is that this has to do with the different algorithm used
for glmer models.
The following example model (which can easily be estimated in
alternative mixed effects programs such as MLwiN or HLM) result in
problems:

glmer(y~(1|level2id)+x+I(x^2),family=binomial,data=data)

It leads to:
Warning message:
In mer_finalize(ans) : false convergence (8)

What is the problem here? Why these convergence issues?

Stijn

--
Best regards,

Stijn Ruiter
Department of Sociology / ICS
Radboud University Nijmegen
P.O. Box 9104
6500 HE Nijmegen
Netherlands

Phone: + 31 24 361 2272
Fax:   + 31 24 361 2399

Visiting address:
Thomas van Aquinostraat 4.01.71
Nijmegen

website: http://oase.uci.ru.nl/~sruiter

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

Dit bericht en eventuele bijlagen geven enkel de visie van de schrijver weer 
en binden het INBO onder geen enkel beding, zolang dit bericht niet bevestigd is
door een geldig ondertekend document. The views expressed in  this message 
and any annex are purely those of the writer and may not be regarded as stating 
an official position of INBO, as long as the message is not confirmed by a duly 
signed document.
Douglas Bates
# 
On Tue, May 19, 2009 at 5:32 AM, ONKELINX, Thierry
<Thierry.ONKELINX at inbo.be> wrote:
Exactly.

The general advice in a case like this is to add the optional argument

verbose = TRUE

in the call to glmer so that you get a display of the progress of the
iterations.   Take a look at

example(cbpp)

You will see that a line in the trace output is of the form

 16:     100.09586: 0.642264 -1.39853 -0.992327 -1.12866 -1.58032

where the first number is the iteration number, the second is the
current value of the deviance, the third is the standard deviation of
the random effects and the fourth and subsequent numbers are the
values of the fixed-effects parameters.  It is likely that one of the
fixed-effects parameters is very large or very small in your original
form of the model.