[ADMB Users] Problems estimating variance component(s)

H. Skaug hskaug at gmail.com
Tue May 4 12:50:16 PDT 2010

> As for the commented-out code:
> totL += -.5*norm2(u);
> e=tau*u;
> I found it a bit confusing as well, but this is how it is listed in the
> ADMB-RE .pdf manual that I downloaded.

Correct, the relative order of these 2 lines does not matter, but you
have used "e" above
in your code where it has not been assigned a value. This is the
reason why this formulation does not work for you.

> "nan", and never recovers).  Thus, my question persists.  Is there a way to
> control ADMB's evaluations of the random effects vectors such that I can
> limit its search for estimates of the "e" vector?

Not anything built in (there may be a random_effects_bounded_vector
floating around, but I do not think it works).

Some suggestons:
- find the range of tau values for which you model works, and restrict
yourself to this range. The range
  in itself provides some useful information about the problem.
- use mfexp() rather than exp(). This may help.
- Transform e via a logistic transformation to provide a bounded
random effect. This may not solve the problem,
  it may have undesired effects.

Im am not sure if any of these will get you somewhere.


More information about the Users mailing list