[R-sig-ME] lme error message

[Ken Beath]

On 22/07/2009, at 8:22 AM, Stuart Luppescu wrote:

> On 水, 2009-07-22 at 08:15 +1000, Ken Beath wrote:
>> On 22/07/2009, at 7:54 AM, Stuart Luppescu wrote:
>>
>>> Running a nested model with lme, and getting this message:
>>>
>>> Error in MEEM(object, conLin, control$niterEM) :
>>> Singularity in backsolve at level 0, block 1
>>>
>>> Can anyone explain this?
>>>
>>
>> Usually means that some of the random effects are highly correlated  
>> or
>> possibly zero.
>
> Random effects? Only the intercept is random, and in a more restricted
> model there's tons of variance.

If by a more restricted model, you mean one with less covariates, then  
either the covariates are reducing your random effects variance to  
zero or causing some other collinearity problem.

Try control option for verbose output, and see if the numbers are very  
large.

Ken