[R-sig-ME] package "ordinal" - failure to converge and slow calculations

[Diana Michl]

Dear Rune Haubo,

I'm using 'ordinal' to model my data - thank you for making it possible! 
But I'm running into a problem and would be very grateful for any 
response on this.

I have an ordered rating response ranging from 1-5 (Wert). The 
predictors are sex (M_W), origin (Richtung), education (Bildung2), and 
age (Alter) of the participants who rated 122-244 items (ItemId). Age is 
numeric, the rest is helmert or sum-contrast coded, Wert is an ordered 
factor. Please see the table below.

Now, I have 3 more data tables that look practically the same and the 
models run okay with those. But with the one below and this code:

ord.allnwo <- clmm(Wert ~ M_W + Alter + Richtung + Bildung2 + (1|ItemId) 
+ (1|Id), data=spnwounsc,                   model=T, Hess=T, 
link="logit", na.action=na.omit, threshold="flexible",         
               control=clmm.control(grtol=1e-6))

I always get the message "Error: optimizer nlminb failed to converge". I 
tried changing the command "control=clmm.control(grtol=5e-4))" to 
numbers between 5e-4 and 1e-9, as you suggested to someone else with the 
same problem. Changing the optimizer to'ucminf' is impossible, according 
to the R error message.

head(spnwounsc)    X  Id M_W Alter Bundesl Richtung            Bildung Liste beide.L ItemId Wert Bildung2
1 1 265   2    55  Ba-Wue        3 Hoch/Fachhochschul     2      ja      1    3        c
2 1 265   2    55  Ba-Wue        3 Hoch/Fachhochschul     2      ja      2    5        c
3 1 265   2    55  Ba-Wue        3 Hoch/Fachhochschul     2      ja      3    5        c
4 1 265   2    55  Ba-Wue        3 Hoch/Fachhochschul     2      ja      4    4        c
5 1 265   2    55  Ba-Wue        3 Hoch/Fachhochschul     2      ja      5    4        c
6 1 265   2    55  Ba-Wue        3 Hoch/Fachhochschul     2      ja      6    5        c
.
.
.
770 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     38    2
771 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     39    3
772 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     40    1
773 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     41    5
774 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     42    5
775 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     43    1
776 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     44    3
777 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     45    4
778 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     46 <NA>
779 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     47 <NA>
780 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     48 <NA>
781 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     49 <NA>
782 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     50 <NA>
783 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     51 <NA>
784 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     52 <NA>
785 4 273   1    46       Hamburg        1 Hoch/Fachhochschul     1    nein     53 <NA>
.
.
.


Do you know what's going on and how I can remedy this?

Also, R takes several minutes until it finally spits out either model or 
error message. Is this a reason to worry, maybe indicating false 
results? Or is it only because my data frames contains 26000-31000 rows?

Many thanks in advance and kindregards

Diana Michl

-- 
Diana Michl, M.A.
PhD candidate
International Experimental
and Clinical Linguistics
Universität Potsdam
www.ling.uni-potsdam.de/staff/dmichl


	[[alternative HTML version deleted]]