Many thanks to Mariusz Lubomirski and Spencer Graves for their replies.
Mariusz has hit the nail on the head, and his explanation and solution has
solved this problem that I have been struggling with.
Gratefully,
Eric Lim
-----Original Message-----
From: Lubomirski, Mariusz
(DMS) [Contr] [mailto:msl@css.ncifcrf.gov]
Sent: 16 August 2004 15:48
To: 'Eric Lim'
Subject: RE: What does this LME error mean?
Eric
If I understand correctly you want
different variances per stratum where "gp"
has only two levels 0 or 1. If you read Pinhero and
Bates page 211, varPower should not be used with
covariates that may assume value 0.
Try instead :
weights = varIdent(form= ~ 1 | gp)
Mariusz
DMS at National Cancer Institute
USA
-----Original Message-----
From: Eric Lim [mailto:eric.lim@cvsnet.org]
Sent: Sunday, August 15, 2004 2:51 AM
To: s-news@lists.biostat.wustl.edu
Subject: What does this LME error mean?
Dear S-news,
I am an S Plus 6 user on Windows XP, and
would be very grateful for any help with regards to this problem that I have
been suffering by fitting a linear mixed effects model of fev1 (a measure of
lung function) with time (timem - months) and group (gp - surgery or not) by patient (id).
lm.3 <- lme
(fev1 ~ timem + timem^2 + timem^3 + gp + gp:timem
+ gp:timem^2 +gp:timem^3, data="" random = ~ timem
| id, na.action=na.omit)
S Plus fits the model perfectly, but when
I try to create a heteroscedastic model by group using:
lm.3v <- update (lm.3, weights = varPower(form
= ~ gp))
I get the error:
Problem in .C("mixed_loglik",: subroutine mixed_loglik:
Missing values in argument 1
Firstly, I would most grateful if anyone
could tell me what this means? I've checked the Pinheiro
and Bates book (as hard as I could), and have copied the command very carefully
from page 216 of their book. (My Christmas wish is for Pinheiro
and Bates would make the lme / nlme
error messages more user friendly...)
Secondly, if anyone could
tell me how to correct it? (I hope
Prof Brian Ripley is around...)
Kind regards,
Eric Lim
Papworth Hospital, UK