[R-sig-ME] Multilevel Ordinal Logistic Regression: clmm warnings

[Rune Haubo]

Nicholas,

I agree with David: this doesn't sound so good and it seems that the
model might not have converged...  However, to provide more qualified
help, I will need to see the result of

summary(Model.mlol)
sessionInfo()

If you can disclose the result of str(your_data) that might also be
helpful. Lastly, I suggest that you install the latest version of
ordinal from R-Forge due to some recent improvements to the fitting
algorithm:

install.packages("ordinal", repos="http://R-Forge.R-project.org")

Cheers,
Rune


On 25 June 2012 00:49, David Duffy <David.Duffy at qimr.edu.au> wrote:
> On Sat, 23 Jun 2012, Nicholas Sabin wrote:
>
>> I am working with repeated measures for subjects and the dependent
>> variable
>> is ordered categories.  So I am working on building a Multilevel Ordinal
>> Logistic Regression model.
>>
>> Example Data:
>>
>> ID      Cycle     X          X2    PerfCat
>> 100         1     3.5        12    1
>> 100         2     7.6        57    4
>> 100         3     6.6        43    3
>>
>> Model.mlol <- clmm(as.ordered(PerfCat) ~
>>      X+X2+(1|ID),
>>      data=ExampleData)
>>
>> "In update.uC(rho) : Non finite negative log-likelihood
>>  at iteration 165"
>
>
> I presume Rune Haubo Christensen will give a better answer but this
> doesn't sound very good.  What did
>
> clmm(as.ordered(PerfCat) ~ X + (1|ID)..
>
> and
>
> clmm(as.ordered(PerfCat) ~ poly(X,2) + (1|ID)..
>
> give?
>
> Also, the rule of thumb is that ordinal variables with five or more levels
> can usually be treated as continuous without inference problems.  That is,
> you should be getting answers close to your LMM result.

Well, in my experience that depends strongly on the problem and the
type of data you are dealing with. I have seen many examples where the
response variable had more than 5 levels and linear (mixed) models
were definitely inappropriate.

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

-- 
Rune Haubo Bojesen Christensen

PhD Student, M.Sc. Eng.
Phone: (+45) 45 25 33 63
Mobile: (+45) 30 26 45 54

DTU Informatics, Section for Statistics
Technical University of Denmark, Build. 305, Room 122,
DK-2800 Kgs. Lyngby, Denmark