[R] Standard errors of the predicted values from a lme (or lmer)

Dieter Menne dieter.menne at menne-biomed.de
Mon Jun 4 14:34:05 CEST 2007

Fränzi Korner <fraenzi.korner <at> oikostat.ch> writes:

> sorry for not being more specific. I would like to use R to get a prediction
> (with standard error) of the response in a mixed model at selected values of
> the fixed-effects factors. Hence, in a mixed model, say, for response body
> size with, say, fixed factors sex and age, I would like to get a prediction
> of size for each sex and at selected ages such as 5, 10, 15; and I want a SE
> for that prediction as well. 

In that case, estimable in gmodels (by Greg Warnes, as also suggested by Michael
Kubovy) and glht in Thorsten Hothorn's multcomp are best. The first works for
lme out of the box, the very powerful glht can handle lmer(lme4), with strong
support for multiple testing. Too bad the latter does not immediately work with
lme, but it can be tweaked.

In both cases, you have to construct the contrast matrix, which can be
error-prone for complex models. To my knowledge (??), there is no
simple-to-handle package that generates this matrix with an intuitive interface.


More information about the R-help mailing list