Skip to content
Prev 7379 / 20628 Next

lme4 and sample size

Dear professor Bates,

Is it possible to put the iteration information in an R object (this is printed when verbose=TRUE in the lmer function)? 

I like to monitor this output during simulations.

Kind regards, 
Walter



________________________________________
Van: dmbates at gmail.com [dmbates at gmail.com] namens Douglas Bates [bates at stat.wisc.edu]
Verzonden: vrijdag 13 januari 2012 17:05
To: Bouwmeester, W.
Cc: R-mixed models mailing list
Onderwerp: Re: lme4 and sample size

On Fri, Jan 13, 2012 at 9:58 AM, Bouwmeester, W.
<W.Bouwmeester at umcutrecht.nl> wrote:
Yes, but do bear in mind that the cvg indicator is from the optimizer,
which is nlminb in the case of the released lme4.  We have encountered
difficulties with nlminb failing to converge or giving the false
convergence message or getting stuck at boundary values.  We later
switched to the bobyqa optimizer from the minqa package and then to a
local implementation of the Nelder-Mead simplex optimizer.

Failure to converge is a property of the optimizer being used, not the
overall design of lme4.  It happens that good optimizers that are
available to Open Source projects are difficult to come by.