Message-ID: <4E088FEE.2030502@gmail.com>
Date: 2011-06-27T14:13:02Z
From: Ben Bolker
Subject: problem of convergence of simulated data
In-Reply-To: <4E088F45.7070905@uea.ac.uk>
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 06/27/2011 10:10 AM, Jim Maas wrote:
> I've faced the same problem with fitting models before. I'm simulating
> different permutations of data sets, starting off with the baseline that
> has almost no variance at all, hence it is difficult for the algorithm
> to converge. Having said that, the estimates will be more than adequate
> if I could relax the convergence criteria even a little. Is there a way
> to do such with the lme procedure?
>
> Thanks for any suggestions.
>
> J
>
>
> Error in lme.formula(nmalor ~ 0 + nmatr1 + nmatr3, ~1 | trtpair, data =
> fednmadat, :
> nlminb problem, convergence error code = 1
> message = singular convergence (7)
>
You could play with the "tolerance" and "msTol" parameters (see
?lmeControl), although I don't know if that will actually help.
Alternately, try try() if you're willing to filter out failures.
Ben Bolker
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iEYEARECAAYFAk4Ij+4ACgkQc5UpGjwzenOh6QCePP6ssyl9kSeEBX26Mwrb+aXO
HggAniwM0LGMhATOF5kVJo40F/Lky0EB
=8a+0
-----END PGP SIGNATURE-----