[R-sig-ME] Error message in ordinal() package running cumulative link mixed models function clmm()

[Sverre Stausland]

Dear all,

I have a data set where subjects gave a 1-7 rating to pairs of items,
and there's a set of numeric independent variables. The data set has
the following set-up:

Subject	Pair	Trial	Response	A	B	E
AM25Y	x,y	54	1	2	4	3.13
AR82M	x,y	65	3	2	4	3.13
BA89W	b,n	27	1	5	6	5.17
CK44O	b,n	19	1	5	6	5.17
etc.

Given the ordinal dependent variable, I need to run a ordinal
regression model. Since I want to model Subject and Pair as random
effects, I am trying Rune Haubo's clmm() function from the ordinal()
package.

First, running a cumulative link model with clm() works fine:

> clm(Response~Trial+A+B+C,data=dataset)->FullModel.clm
> summary(FullModel.clm)
Call:
clm(location = Response ~ Trial + A + B + C, data = dataset)

Location coefficients:
             Estimate Std. Error z value  Pr(>|z|)
Trial	-0.0010   0.0009    -1.0638 0.28743
A	-0.1953   0.0247    -7.9144 2.4841e-15
B	-0.1446   0.0240    -6.0166 1.7809e-09
C	-0.2151   0.0204   -10.5287 < 2.22e-16

No scale coefficients

Threshold coefficients:
      Estimate Std. Error z value
1|1.5  -2.7595   0.0915   -30.1696
1.5|2  -2.7513   0.0914   -30.0925
2|3    -1.8590   0.0874   -21.2622
3|4    -1.1453   0.0853   -13.4291
4|5    -0.4202   0.0849    -4.9511
5|6     0.6346   0.0897     7.0737
6|6.5   2.3828   0.1289    18.4908
6.5|7   2.3942   0.1293    18.5119

log-likelihood: -9104.93
AIC: 18233.86
Condition number of Hessian: 342793.76


But when I try to run a cumulative link mixed model with Subject as a
random effect, I get errors:

> clmm(Response~Trial+A+B+C,random=Subject,data=dataset)->FullModel.clmm
There were 50 or more warnings (use warnings() to see the first 50)
> warnings()
Warning messages:
1: In rho$updateU(rho) : Non finite negative log-likelihood
  at iteration 112
2: In rho$updateU(rho) : Non finite negative log-likelihood
  at iteration 112
3: In rho$updateU(rho) : Non finite negative log-likelihood
  at iteration 112

> summary(FullModel.clmm)
Cumulative Link Mixed Model fitted with the Laplace approximation

Call:
clmm(location = Response ~ Trial + A + B + C, random = Subject, data = dataset)

Random effects:
             Var  Std.Dev
Subject 2.241108 1.497033

Location coefficients:
             Estimate Std. Error z value Pr(>|z|)
Trial	-0.0015      NaN        NaN NA
A	-0.2581      NaN        NaN NA
B	-0.1833      NaN        NaN NA
C	-0.2833      NaN        NaN NA

No scale coefficients

Threshold coefficients:
      Estimate Std. Error z value
1|1.5 -3.6036      NaN        NaN
1.5|2 -3.5921      NaN        NaN
2|3   -2.3596      NaN        NaN
3|4   -1.3958      NaN        NaN
4|5   -0.4317      NaN        NaN
5|6    0.9001      NaN        NaN
6|6.5  2.8441      NaN        NaN
6.5|7  2.8557      NaN        NaN

log-likelihood: -8022.125
AIC: 16070.25
Condition number of Hessian: NaN
Warning message:
In summary.clmm(FullModel.clmm) :
  Variance-covariance matrix of the parameters is not defined

I would appreciate if anyone could point out to me what I am doing
wrong in my use of the clmm() function.

Thanks
Sverre