[R-sig-ME] dealing with the convergence status

[Ben Bolker]

  I'm interested in the answer as well.  I thought about this
briefly,  but the only thing I came up with (which is a terrible
hack) is to use options(warn=2) to turn warnings into errors and
then use try() to catch and evaluate the errors.  One might
also be able to inspect last.warning ... There must
be a better way though ...

  Ben Bolker

Hervé CHAPUIS wrote:
> Hi everyone,
> 
> I am using lmer (and glmer) in a simulation study in order to find the 
> best experimental design.
> Using gaussian deviates I simulate a population and  I analyse the data 
> with a mixed model.
> I repeat these operations many times (about 5000 replicates)  in order  
> to derive trends and distributions of parameters.
> 
> My problem is that  I sometimes run into false convergences. I have a 
> warning message but that's all.
> "Il y a eu 50 avis ou plus (utilisez warnings() pour voir les 50 premiers)"
> 
> Is there a way to obtain the convergence status of the lmer/glmer models 
> in a tractable way so that I  do not account for this replicate in case 
> of false  convergence ? Given the large number of replicates, I do not 
> run the program interactively.
> 
> I hope I made it clear.
> Many thanks.
> 


-- 
Ben Bolker
Associate professor, Biology Dep't, Univ. of Florida
bolker at ufl.edu / www.zoology.ufl.edu/bolker
GPG key: www.zoology.ufl.edu/bolker/benbolker-publickey.asc



-- 
Ben Bolker
Associate professor, Biology Dep't, Univ. of Florida
bolker at ufl.edu / www.zoology.ufl.edu/bolker
GPG key: www.zoology.ufl.edu/bolker/benbolker-publickey.asc