From: "James Bailey" <James_Bailey@Emory.org>

Subject: minimization termination

Date: Thu, 26 Jul 2001 17:33:29 -0500

 

To all

 

I am using the conditional estimation technique and get the following

error message

 

MINIMIZATION TERMINATED DUE TO PROXIMITY OF LAST ITERATION EST TO A

VALUE AT WHICH THE OBJ FUNC IS INFINITE

 

Any advice on how to interpret the message and what to do.

 

Thanks

 

Jim Bailey

 

*******

 

From: "bvatul" <bvatul@ufl.edu>

Subject: Re: minimization termination

Date: Fri, 27 Jul 2001 09:52:42 -0400

 

Hello James

Inorder to get around your error message you can do the following:

1. Check if you are overparameterising your model.

2. Check your initial estimates (You can run some simulations and check your

initial estimates)

3. Have a look at the etas and thetas in your output. SOme of them might be

really bad. IN that case you can fix the eta and rerun the analysis. perhaps

theta too intially. It all depends on your output numbers.

4. You may want to minimise the number of etas in your model. In case you

dont have a dense data structure and if you are using a peripheral

compartment too then you can use on eta for both k12 and k21 (ie.,

distribution rate constants) or something like that.

5. You might also change the initial estimates of your etas.

6. In FOCE your error structure can also influence your outcome. You can

change the error model.

7. If you have data rich situation you might even try FO.

 

I am not sure of the exact interpretation of the error message (in terms of

the mathematical steps in the algorithm) but these things will get you

around the problem.

 

I hope this helps

 

Atul