[R-sig-ME] Passing lme4 optimizer arguments to pbmodcomp and afex/mixed

[Henrik Singmann]

I am also currently working on a binomial GLMM on which I encounter this problem regularly (depending on which model I fit). But as my data is somewhat large (12000 observations) and fitting takes a while I wouldn't consider it a nice demonstration of the issue.

However, I did try to increase the number of steps in pwrssUpdate by altering the source of lme4 and, as predicted by Ben, this does not help at all (it just reaches the new boundary).

But, if nevertheless someone is interested to try it out for him or herself, you can install a version of lme4 1.1.0 that is identical to the development version as of yesterday that only stops after 10000 pwrssUpdate steps from my github:
library("devtools"); install_github("lme4",user="singmann")

Best,
Henrik


Am 25.09.2013 15:21, schrieb Ben Bolker:
>    At the moment it's not possible to increase the number of iterations;
> that's on the list to allow, but my guess is that that's not going to
> help. Unfortunately, our test cases for lme4 did not include a lot of
> binomial GLMMs with complete separation and/or extreme values of the
> beta parameters (say |beta|>5).  I suspect that what's going on here is
> similar to the examples listing in these issues ..
>
> https://github.com/lme4/lme4/issues/134
> https://github.com/lme4/lme4/issues/124
>
>    Is it in fact the case that if you fit this model with lme4.0 (or an
> older version of lme4) you get large values of beta?
>
>    Looks like we're going to have to put higher priority on fixing this ...
>
>    Ben Bolker
>
>
> On 13-09-25 09:13 AM, Tom Wenseleers wrote:
>> Dear all, Just  a quick question re pbmodcomp (I'm using version
>> 0.4-0 of pbkrtest, and latest devel version of lme4). I am calling
>> pbmodcomp via afex, but I am getting the dreaded error "3 nodes
>> produced errors; first error: pwrssUpdate did not converge in 30
>> iterations" (my call is
>> mixed(ALATES~ANT_TENDED*MELEZITOSE+NR_START+(1|CLONES)+(1|PLANT)+(1|REPLICATE),type=3,family=poisson,method="PB",args.test
>> = list(nsim = 10,cl=cl),data=data_alates) ).
>>
>> I would now like to try passing some optimizer arguments to the lmer
>> calls. To do this I first called options(glmerControl=list(optimizer
>> = c("bobyqa", "bobyqa"),tolPwrss = 1e-03
>> ,optCtrl=list(maxfun=1000))) But then I still get " pwrssUpdate did
>> not converge in 30 iterations". Does one of you happen to know how I
>> could perhaps change the nr of maximum iterations to >30? And also if
>> it is possible to pass such arguments in the call to mixed and/or
>> pbmodcomp?
>>
>> Cheers, Tom
>>
>>
>> _______________________________________________________________________________________
>>
>>   Prof. Tom Wenseleers P      Lab. of Socioecology and Social
>> Evolution Dept. of Biology Zoological Institute K.U.Leuven
>> Naamsestraat 59, box 2466 B-3000 Leuven Belgium +32 (0)16 32 39 64 /
>> +32 (0)472 40 45 96 //tom.wenseleers at bio.kuleuven.be
>> http://bio.kuleuven.be/ento/wenseleers/twenseleers.htm
>>
>>
>>
>

-- 
Dipl. Psych. Henrik Singmann
PhD Student
Albert-Ludwigs-Universität Freiburg, Germany
http://www.psychologie.uni-freiburg.de/Members/singmann