Date:         Wed, 22 Oct 2003 16:02:20 +0200
Reply-To:     Andy <uio2@RZ.UNI-KARLSRUHE.DE>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Andy <uio2@RZ.UNI-KARLSRUHE.DE>
Organization: University of Karlsruhe, Germany
Subject:      PARMS in wide-form not importing init values (NLMIXED)
Content-Type: text/plain; charset=iso-8859-1

Hi,

I'm trying to fit a model with NLMIXED, starting from a grid of init values.
I would like to import the inits, instead of hard-wiring them in the code.

As soon as I introduce a random effect "u1" in the model, none of the init
observations expect the first one are read properly: From the 2nd init
observation on, the warning

"WARNING: Parameter vector X is not processed due to invalid or missing
values"

is printed to the log file. If I either remove "u1" or include the inits in
the code, everything works fine again. Model see below.

Has anyone encountered and solved the same or similar problem?

Thanks in advance,
Andreas

model:
%MACRO RunP1SB_010 (file=,lib=,CPO=);
 OPTIONS MLOGIC;
 %IF &CPO = 1 %THEN %LET CPOstr=CPO; %ELSE %LET CPOstr=;
 ODS OUTPUT ParameterEstimates=&file.work_&CPOstr.pe;
 ODS OUTPUT FitStatistics=&file.work_&CPOstr.fit;
 ODS OUTPUT Dimensions=&file.work_&CPOstr.dim;
 PROC NLMIXED DATA=&lib..&file.&CPOstr.DATA
  METHOD=GAUSS NOAD QFAC=2000 QMAX=4000 QTOL=0.5 TECH=TRUREG
  FDH=CENTRAL EBOPT OPTCHECK ITDETAILS;
  %IF &CPO = 1 %THEN %DO;
   BY datagroup;
  %END;
  PARMS / DATA=&lib..&file.&CPOstr.INIT;
  /*PARMS mu=10 12 14 16, sigma1= 0.75 1.0 1.25 sigma2=1.0 1.2 1.4 1.6;*/
  ARRAY sigma[2] sigma1 - sigma2;

  ll = LOGPDF('LOGNORMAL', y, mu + u1, sigma[subject]);
  MODEL y ~ general(ll);
  RANDOM u1 ~ normal(0, s2u) SUBJECT = subject OUT=&file.work_&CPOstr.u;
 RUN;
%MEND RunP1SB_010;