Hello all. I'm currently working with mixed models, and have noticed a curious difference between the nlme and lmer packages. While I realize that model selection with mixed models is a tricky issue, the two packages currently produce different AIC scores for the same model, but they systematically differ by 2. In looking at the logLik values for each method, I find that they indeed differ by 1. So, the following code:
utils::data(npk, package="MASS") library(lme4) a<-lmer(yield ~ 1+(1|block), data=npk) logLik(a) library(nlme) b<-lme(yield ~ 1, random=~1|block, data=npk) logLik(b) produces a df of 2 for a, and a df of 3 for b. I'm guessing that lmer is not accounting for the level-1 variance. Is this the case, and, if so, will this be fixed? I see that this issue was brought up sometime back. Is there a reason it has not been addressed? https://stat.ethz.ch/pipermail/r-help/2006-March/102520.html Incidentally, I'm also curious what folk think about the approach to using the conditional AIC value as posted here https://stat.ethz.ch/pipermail/r-help/2008-February/154389.html Thanks! -Jarrett ______________________________________________ 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.
