[R-sig-ME] var-covar matrix in ranef

[Douglas Bates]

On Tue, Feb 9, 2010 at 2:50 AM, Hester Lingsma <h.lingsma at erasmusmc.nl> wrote:
> Dear R users,
> If I fit a model with a random slope and a random intercept, the var-covar
> matrix derived from PostVar from the function ranef is a 2 by 2 matrix for
> each upper level subject. I want to use the posterior estimates (from ranef)
> and their standard error from both the slope and the intercepct for each
> upper level subject. Which elemants of the 2 by 2 matrixes to use for the se
> of the posterior estimate?

I'm not sure what "the posterior estimates" means but that term is
probably my fault because the argument name is "postVar" for
"posterior variance".  (Actually I think it is Harold Doran's fault
because he is the one who suggested the term "posterior variance",
which I not realize is a misnomer.)  Even though the argument name is
"postVar", and I would prefer not to change it at this point, I now
refer to the values returned by ranef as the conditional means (for
linear mixed models, in more general models they are the conditional
modes) of the random effects given the observed data and evaluated at
the parameter estimates.  The conditional standard deviations of the
random effects will be the square roots of the diagonal elements of
the 2 by 2 matrices returned in the postVar attribute.