[ADMB Users] The importance of parameterizing a nonlinear model well

[dave fournier]

Well if it was just this small example my reaction would be a bit over 
the top.
However it is not just one example.  For many people the world of reasonable
problems just happens to coincide with what is easy to do in R.  The 
standard
response is that R is "easier to use" than ADMB.  But it is only easier 
to use if
you can solve the problem with R.  There is a big world of problems that 
you can not
begin to solve with R and that are simpler with ADMB.

I'll give you two examples.  A while back I posted a toy example of a 
fisheries
management model that at the time took 38 seconds in ADMB and 1.5 hours 
in R.

The response was an ad hominem attack and some obfuscation about how
some of the parameters should have been random effects.  In fact it was
a simpler model of the kind commonly used by fisheries managers. Some of the
real examples of this kind take several hours to fit when written in ADMB.

In spite of this I note in Millar's book he repeats the mantra that R is 
easier
than ADMB and that ADMB should only be used for things that can not be done
in R with the implication that this consists almost exclusively of 
latent variable models.
Of course he only treats completely trivial fish models.

Second is the completely false assertion that you can't make frequentist 
assertions about
functions of fixed and random effects i.e something like a profile 
likelihood in a mixed
effects model.  Of course this is impossible to do in R.  It would be 
pretty easy to add it to
ADMB.  I have finished a sketch of this in the attached pdf.













-------------- next part --------------
A non-text attachment was scrubbed...
Name: simple2.pdf
Type: application/pdf
Size: 96802 bytes
Desc: not available
URL: <http://lists.admb-project.org/pipermail/users/attachments/20130918/bfb24101/attachment.pdf>