R help - Mar 2011 - how to set the starting value in lme

Hi,
When I fit the mixed-effects model by lme, I got the convergence error code = 1
message = iteration limit reached without convergence (9). Even after I increase
the iteration numbers, it still did not work. The only way I found to avoid the
error message is to increase the number of iterations for the EM algorithm,
which need to be at least 1100. But the weird thing is that the correlation
between two random coefficients is always around 1, which is not true. So I am
thinking the error should be related with the improper starting value. But I do
not know how to set up my desired starting value in lme function. Are there
anybody know it? Or met the same problem with me before and know how to avoid
the error? Thank you for your help.
Bests,
Jia


      
	[[alternative HTML version deleted]]

Hi Jia,

in order to test if the failing parameter estimation really has to do
with wrong initial values, you could first simulate data with your mixed
effects model. Just assume parameters, produce a data set with these
parameters, add noise and look if you can estimate the parameters
correctly with varying initial guesses.



On Tue, 2011-03-15 at 12:34 -0700, jia liu wrote: 
> Hi,
> When I fit the mixed-effects model by lme, I got theconvergence error code
= 1
> message = iteration limit reached without convergence (9). Even after I
increase the iteration numbers, it still did not work. The only way I found to
avoid the error message is to increase thenumber of iterations for the EM
algorithm, which need to be at least 1100. But the weird thing is that the
correlation between two random coefficients is always around 1, which is not
true. So I am thinking the error should be related with the improper starting
value. But I do not know how to set up my desired starting value in lme
function. Are there anybody know it? Or met the same problem with me before and
know how to avoid the error? Thank you for your help.
> Bests,
> Jia
> 
> 
>       
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
-- 
Daniel Kaschek
Physikalisches Institut, Freiburg
Hermann-Herder-Str. 3
79104 Freiburg

Office:  Westbau, 02020
Phone:   +49-761-203-8531
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 490 bytes
Desc: This is a digitally signed message part
URL:
<https://stat.ethz.ch/pipermail/r-help/attachments/20110316/d8356380/attachment.bin>

Daniel Kaschek <daniel.kaschek <at> physik.uni-freiburg.de> writes:
> 
> Hi Jia,
> 
> in order to test if the failing parameter estimation really has to do
> with wrong initial values, you could first simulate data with your mixed
> effects model. Just assume parameters, produce a data set with these
> parameters, add noise and look if you can estimate the parameters
> correctly with varying initial guesses.
> 
  PS.  Correlations between random effects close to 1.0
(or -1.0) are usually a sign of insufficient data.
Further questions along these lines should probably go
to the r-sig-mixed-models list ...