[ADMB Users] a difference in estimates between versions?

[dave fournier]

Well actually it doesn't run very well as you are discovering.
You should try and formulate it in R and submit it to r help, that would 
be fun,  but I digress.

These models are difficult even if you do them right, because often the 
data is just not informative.
To improve this model.

1.) read my semi implicit formulation of the differential equation 
solution. It is much more stable
than the explicit version you are using.

2.) With a time step of 1 the value or r you are estimating is probably 
not very close to the
value of r you would get from an exact solution of the DE.  You should 
modify the
code to have n steps in the solution and see what happens when you 
increase n.

3.)  B1 and K are highly correlated. You should parameterize B1    as  
B1=bmult*K

4.) You should always rescale the data to be close to 1.

   That would be a start.