On Sun, Jan 4, 2009 at 5:28 PM, Bert Gunter <gunter.berton at gene.com> wrote:
Folks:
Lme() is only able to work with nested random effects, not with crossed random effects.
Not quite true. Crossed models **can** be done, albeit clumsily, via pdMatrix objects: the Bates/Pinheiro book even contains an example or two (one on assay plates, I recall, but I don't have my book with me for the reference).
Clumsily indeed and very inefficiently. I would not recommend using lme or nlme to fit models with crossed random effects other than for small toy examples.
Also, lme, not lmer, is currently the only way to implement penalized splines as random effects -- see the lmeSplines package.
And this is relevant because ??? The original question was about the number of denominator degrees of freedom and I will freely admit that lmer, in particular, does not produce the "correct" denominator degrees of freedom. (As many people discover to their chagrin, it doesn't provide denominator degrees of freedom at all.) In my opinion there isn't such a thing as the "correct" denominator degrees of freedom except in small, perfectly balanced, toy examples, which is why I am not that terribly concerned.