[R] Confidence interval for predictions in mixed effects models

[Douglas Bates]

Montse Rue <rue at jimmy.harvard.edu> writes:

> I would like to know if I can obtain confidence intervals for
> predictions when using nlme.

There is no pre-defined function to do that.  If you are predicting
based on the fixed-effects terms only (i.e. the "population"
predictions) you could use the standard errors of those parameters to
calculate a standard error of your predicted mean response.  If you
are predicting for observed values of the grouping variable(s) it
would be much more difficult to get standard errors of the mean
response.  If I recall correctly the conditional standard deviations
of the conditional modes of the random effects are not passed back to
R from the C code.