[R-sig-ME] mixed models without residual term

[Douglas Bates]

On Fri, Feb 22, 2008 at 2:57 AM, De Maeyer Sven <sven.demaeyer at ua.ac.be> wrote:
> Dear List members,

>  I'm a recently new user of R and lme4. In some of my research designs I used to estimate mixed models without a residual term. Only random effects were added, but no residual. I did this in MlwiN. Now I'm trying to do the same analyses with lmer, but I can't seem to fix the residual to zero. Is this currently possible in lme4?

I don't think so.  The models that can be fit with the lme4 package
are all based on the assumption that the response vector is an n
dimensional random variable.  I'm not sure how to define maximum
likelihood estimates for parameters based on observed data without a
probability model for the distribution of the response vector.