I'm uncertain how helpful it will be to give example code, but last week,
this model gave an error message to the tune of "failed to converge" after
about 5 minutes of run-time :
library(nlme)
model.A<- lme (fixed = avbranch~ wk*trt*pop
, random = ~wk|ID/fam/pop, data=branch)
It seemed that failure to converge made sense, since there were many weeks
(wk) and values for "ID/fam/pop". I settled for this model:
model.A2 <- lme (fixed = avbranch~ wk*trt*pop
, random = ~1|ID/fam/pop, data=branch)
However, when I tried the model.A on a different dependent variable this
week, it converged. Since the challenge to convergence (many levels of wk
and ID/fam/pop) was the same as it had been before, I went back and tried
model.A on the other analysis, and it also ran. I then started checking
results for everything I'd done in the past three weeks in packages that use
ML methods (FIML, REML)--and got different outcomes. I've quadruple-checked
to be sure I'm using the same code and the same data (I use .csv files for
simplicity), and see no differences. However, results from nlme and sem
packages are both different. I had not saved detailed output, but had
recorded parameters, model-fit statistics, and convergence failures.
Could some new package I installed could have changed the way that MLE
methods are functioning in the work environment? Everything in the search
path looks as it did before, but could something like Rcmdr (just installed,
but not now in the search path) change other parts of the environment?
Is that possible? If so, how do I check? If not, could anything (besides
user error--obviously the simplest solutions are that the code or the data
really is different or I recorded the results incorrectly before) account
for different results using the same data and code?
My main concern is about which results are correct: convergence seems to be
happening much faster now, and some models are showing better fits, but the
difference makes me nervous.
Any ideas would be helpful.
Thanks,
Claire Lay
[[alternative HTML version deleted]]
______________________________________________
R-help@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.
