[ADMB Users] restricted maximum likelihood

[Ian Fiske]

Hi all,

The ADMB-RE manual briefly discusses implementing restricted maximum
likelihood (REML) estimation.  The suggested approach is to integrate
out the fixed effects by specifying them as a random_effects_vector.
This makes sense.  However, the usual REML practice is to

(1) estimate variance parameters using the integrated likelihood.
(2) estimate fixed effects conditional on the variance estimate from (1).
(3) iterate between (1) and (2) for nonlinear models.

In contrast, the ADMB-RE method estimates fixed effect parameters
using EBLUPs just like random effects.  Does this turn out to be
equivalent to the usual procedure above because no prior is specified
for fixed effects?  I guess the most obvious difference is that fixed
effects and random effects are estimated simultaneously within the
inner minimization.  Does this matter?  Maybe this is obvious, but I
just want to be sure I know what I'm doing.

Thanks!
Ian

-- 
Ian Fiske
PhD Candidate
Department of Statistics
North Carolina State University