glmmPQL / "system is computationally singular"

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

glmmPQL / "system is computationally singular"

David Reitter
Hi,

I'm having trouble with glmmPQL from the MASS package.
I'm trying to fit a model with a binary response variable, two fixed  
and two random variables (nested), with a sample of about 200,000  
data points.

Unfortunately, I'm getting an error message that is difficult to  
understand without knowing the internals of the glmmPQL function.

> model <-  glmmPQL(primed ~ log(dist) * role , random = ~ dist |  
> target.utt / prime.utt , family=binomial(link = "logit"),  
> data=data.utts, niter=5, verbose = TRUE)
> Loading required package: nlme
> iteration 1
> iteration 2
> iteration 3
> Error in solve.default(pdMatrix(a, fact = TRUE)) :
>         system is computationally singular: reciprocal condition  
> number = 8.65949e-32
> In addition: Warning messages:
> 1: Singular precision matrix in level -1, block 4
> 2: Singular precision matrix in level -1, block 4
> 3: Singular precision matrix in level -1, block 4
> 4: Singular precision matrix in level -1, block 4
> 5: Singular precision matrix in level -1, block 4

Any suggestions? Will a larger dataset (possible) solve the problem?

Thanks
David

--
David Reitter - ICCS/HCRC, Informatics, University of Edinburgh
Blog: http://www.davids-world.com   Homepage: http://www.david- 
reitter.com

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
Reply | Threaded
Open this post in threaded view
|

Re: glmmPQL / "system is computationally singular"

Spencer Graves
          Did you try "traceback()"?  What do you get?

          I've had good luck with problems like this in listing the function
then using "debug" to review while I walk throught the code line by line.

          This may not be the issue here, but with "family=binomial", if the
model being fit can achieve perfect separation, software of this type
might generate error messages similar to what you describe.

          hope this helps.
          spencer graves

David Reitter wrote:

> Hi,
>
> I'm having trouble with glmmPQL from the MASS package.
> I'm trying to fit a model with a binary response variable, two fixed  
> and two random variables (nested), with a sample of about 200,000  
> data points.
>
> Unfortunately, I'm getting an error message that is difficult to  
> understand without knowing the internals of the glmmPQL function.
>
>
>>model <-  glmmPQL(primed ~ log(dist) * role , random = ~ dist |  
>>target.utt / prime.utt , family=binomial(link = "logit"),  
>>data=data.utts, niter=5, verbose = TRUE)
>>Loading required package: nlme
>>iteration 1
>>iteration 2
>>iteration 3
>>Error in solve.default(pdMatrix(a, fact = TRUE)) :
>>        system is computationally singular: reciprocal condition  
>>number = 8.65949e-32
>>In addition: Warning messages:
>>1: Singular precision matrix in level -1, block 4
>>2: Singular precision matrix in level -1, block 4
>>3: Singular precision matrix in level -1, block 4
>>4: Singular precision matrix in level -1, block 4
>>5: Singular precision matrix in level -1, block 4
>
>
> Any suggestions? Will a larger dataset (possible) solve the problem?
>
> Thanks
> David
>
> --
> David Reitter - ICCS/HCRC, Informatics, University of Edinburgh
> Blog: http://www.davids-world.com   Homepage: http://www.david- 
> reitter.com
>
> ______________________________________________
> [hidden email] mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html