[ADMB Users] multiple random effects vectors with correlated Gaussian priors

[Ian Fiske]

Thanks for the responses, Hans.  Here's a brief update as I'm trying
to figure this out.  I have been following the packaged spatial
example instead of the web site example because my data are not on a
regular grid.  In Cressie's terminology, I'm using geostatistical
methods rather than lattice methods.

>>
>> (*) Is it possible to specify correlated priors for multiple vectors
>> of random effects?
>
> I am not sure. Only Dave would know. It would be nice
> if the answer is yes.
>
>

Because of the experimental nature of this feature, I've started to
dig into the source a bit to try and figure out the problem with
multiple normal_prior vectors.  Looking at the source, line 18 of
fquadpri.cpp

const int df1b2quadratic_prior::max_num_quadratic_prior=100;

which makes me think that we can have up to 100 normal_prior vectors.
If I'm interpreting this right, then my measly 2 vectors should work.

>> (*)  Is it possible to construct a concentrated likelihood for the
>> fixed effects when computations are done inside of a
>> SEPARABLE_FUNCTION?
>
> I take "concentrated likelihood" to mean "integrated likelihood".
> Is this also for the spatial model, or is it for a separate model?
>
> Hans
>

I was talking about the "concentrated likelihood" that you get by
solving for one parameter in terms of others and plugging in to get
rid of a parameter.  But this is of less interest to me than the
correlated priors, so it's kind of tangential.

And I'd be happy to update the manual -- once I get this working ;) .

Best,
Ian

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