[R-sig-ME] Glmulti, clm, computationally singular system

[Ben Bolker]

Liz Hare <doggene at ...> writes:

> 
> Hello,
> 
> I am using glmulti with clm. I get an error that "system is 
> computationally singlar: reciprocal condition number = 2.100..."
> 
> I've searched the web for this kind of problem and found suggestions to 
> use other methods (QR decomposition, Cholesky) to find the inverse, but 
> the only option in clm is NR.
> 
> Is there anything else I can change about the clm runs or glmulti before 
> I start taking variables out?

  Any chance of a reproducible example? http://tinyurl.com/reproducible-000
  (It would be helpful to tell us that clm is in the ordinal package
and that "glmulti" has its own package.)

  Can you expand on "the only option in clm is NR"?  I assume "NR"
means Newton-Raphson, which puzzles me a bit since that's not really
want lme4 is doing anyway, and I thought clm was built on lme4 ...
Are your failures coming from clm or from glmulti?  I assume what's
happening is that glmulti is trying to fit a submodel that's failing
within clm.  It would be helpful to isolate the particular submodel
that's failing (if that's the case).

  I may be totally off here, just trying to guess -- please correct
any misconceptions.