The problem here is that parameter estimates from a prior successful procedure run are not being overwritten when the NLMIXED code fails to converge. Rather than creating a copy of the output data set prior to executing NLMIXED in which parameter estimates are initialized with some null value, I would delete the file which contains the parameter estimates after appending the parameter estimates to the master file.

Thus, I would structure code something like the following:

%macro mysim(...); /* ... indicates macro parameter specification */

... /* ... indicates any macro processing steps */ /* that are performed prior to iterations */ /* in which data are generated and the */ /* NLMIXED procedure is executed. */

%do iter=1 %to 1000; <simulated data generation>

/* execute NLMIXED against simulated data */ ods output ParameterEstimates=Parms_iter; proc nlmixed data=sim; ... /* ... NLMIXED code for fitting the model */ run;

/* Test for existence of Parms_iter. Existence implies */ /* successful completion of the NLMIXED procedure. If */ /* the file exists, attach the current iteration number */ /* to the file, append the file to the master results */ /* file, and then delete the current iteration results. */ /* */ /* If the file does not exist, we just go on to the */ /* next iteration. */ %if %sysfunc(exist(Parms_iter)) %then %do; data parms_iter; retain iteration &iter; set parms_iter; run;

proc append base=Parms data=Parms_iter; run;

proc datasets; delete Parms_iter; quit; %end;

... %mend;

Dale

--------------------------------------- Dale McLerran Fred Hutchinson Cancer Research Center mailto: dmclerra@nFoHsCpRaCm.org Ph: (206) 667-2926 Fax: (206) 667-7004 ---------------------------------------

--- On Wed, 1/26/11, Data _null_; <iebupdte@GMAIL.COM> wrote:

> From: Data _null_; <iebupdte@GMAIL.COM> > Subject: Re: Outputting parameter estimates in NLMIXED > To: SAS-L@LISTSERV.UGA.EDU > Date: Wednesday, January 26, 2011, 9:44 AM > Before each call to NLMIXED create > the data set that is expected. > Make it with 0 obs and variables. Then the SET should > work but not > pick up the wrong data. > > On Wed, Jan 26, 2011 at 11:33 AM, SUBSCRIBE SAS-L > Ameribrit > <FRP3@pitt.edu> > wrote: > > HI Guys: > > Is there any way in the ods output statement to only > tell NLMIXED to > > output the file if the procedure converges? > > > > I am trying to accumulate all the estimates from a > simulation > > by using the set statement on the previous output. But > what I think is > > happening is if NLMIXED fails with "no valid parameter > points found' the > > output file is not created then I set the output file > on this iteration > > (which now dosnt exist) with the output file from an > iteration which did > > run and I end up with a run of identical parameter > estimates.This makes it > > look like I am running the proc on the same data set > and that the > > randomness is not present. > > > > So is there a better way to output results from a proc > and append to the > > output file only if the proc converges and has results > to output? > > > > thanks > > Francis > > >