PROC NLIN failed to converge

[LOLO12]

Hi All, 

PROC NLIN failed to converge, I'm fitting around 500 model and this is expected. I want to have  missing values where it fil to converge or just don't get this warning at all. 

[ballardw]

To help us help you it is generally a good idea to at least include the code you are currently running. Better would be to include a LOG entry including the code an all of the messages from the log.

Where do you want to have "missing values"? This is not an obvious, at least to me, requirement.

[LOLO12]

in the estimates variables. it is still generating results in the estimate when the model didn't converge. 

[PGStats]

The variable _STATUS_ in the OUTEST= dataset tells you which estimates are final (properly converged) or not.

[LOLO12]

Please have the log attachment. 

Here is the SAS code 

PROC NLIN DATA=slope1 ;

PARAMETERS L=1 to 3 U=3 to 16 R=.5;

Zl=(parcat3n < R)*(R - parcat3n);

MODEL aval=L + U*(Zl);

BY USUBJID TRTA PARAMN AVISION;

Output out=result parms=L parms=U parms=R

  predicted=predicted

  residual=residual

  sse=sumofsq_err

          stdr = standarderr;

run;

[Rick_SAS · Accepted solution]

Take a look at the article, "Monitor convergence during simulation studies in SAS." It explains how to monitor convergence. You can use post-processing to set the predicted value to missing for models that did not converge.

Note that you cannot suppress WARNINGS, but you can use OPTIONS NONOTES to suppress the NOTEs.

It also might be worth investigating WHY these models are not converging. For example, some  BY groups might have an insufficient number of valid observations (eg, only 2 data points). You can pre-process the data to exclude those situations.

[LOLO12]

I know why it is not converging for some of the data because some of formula that I'm using doesn't fit all observation. I want to know which  data is not coversiong so I can use the right algorithm.