topic Re: constrained regression in Statistical Procedures

constrained regression

Amalik — Tue, 27 Dec 2016 21:52:39 GMT

Hi,

I want to run a constrained regression with the following constarints,

US_Small_Val + US_Small_Gr + US_Large_Gr + _US_Large_Val = 1;

0 <= US_Small_Val <= 1; 0 <= US_Small_Gr <= 1; 0 <= US_Large_Gr <= 1; 0 <= _US_Large_Val <= 1;

where the model is,

AMP_AIT = US_Small_Val US_Small_Gr US_Large_Gr _US_Large_Val /noint;

I am attacing the code wci I have used but its not working.

Can smeone please help me out in this?

Regards,

Anum

Re: constrained regression

PGStats — Tue, 27 Dec 2016 22:42:41 GMT

What's the error message?

You could try starting the regression at a feasable point

parms US_Small_Val 0.25 US_Small_Gr 0.25 US_Large_Gr 0.25 _US_Large_Val 0.25;

Re: constrained regression

Amalik — Tue, 27 Dec 2016 23:03:54 GMT

it says

ERROR 22-322: Syntax error, expecting one of the following: ;, ADJSMMV, ALPHA, BLOCK, BREUSCH,
CATNAME, CHOW, COLLIN, CONVERGE, CORR, CORRB, CORRS, COV, COVB, COVBESTIMATOR,
COVS, DATA, DETAILS, DW, DWPROB, DYNAMIC, EPSILON, FASTSOLVE, FLOW, FORECAST,
FSRSQ, GENGMMV, GINV, GMM, GODFREY, GOF, GRAPH, HAUSMAN, HCCME, HESSIAN, HIDE,
I, INTGPRINT, IT2SLS, IT3SLS, ITALL, ITDETAILS, ITGMM, ITOLS, ITPRINT, ITSUR,
JACOBI, KERNEL, LIBNAME, LIST, LISTALL, LISTCODE, LISTDEP, LISTDER, LTEBOUND, M,
MAXERRORS, MAXITER, MAXSUBITER, MEMORYUSE, METHOD, MINTIMESTEP, MISSING, MOD,
MODEL, MODFILE, N2SLS, N3SLS, NAHEAD, NDEC, NDRAW, NEWTON, NODWPROB, NOGENGMMV,
NOHAUSMAN, NOINTGPRINT, NOMEM, NOMEMORYUSE, NONORMAL, NOOUTOBJVALS,
NOOUTVIOLATIONS, NOPRINT, NOREPORTMISSINGS, NORMAL, NOSTARTITER, NOSTORE,
NPREOBS, NVDRAW, OBFUSCATE, OLS, OPTIMIZE,

Re: constrained regression

Rick_SAS — Wed, 28 Dec 2016 00:56:44 GMT

Looks like that error message is coming from the FIT statement.

Re: constrained regression

PGStats — Wed, 28 Dec 2016 03:38:51 GMT

I am just guessing, I think the proper syntax for your proc model invocation should be something like:

proc model data=temp1 noprint outest=Out_Estimates;
parms a 0.25 b 0.25 c 0.25 d 0.25;
bounds 0 <= a b c d <= 1;
restrict a + b + c + d = 1;
AMP_AIT = a*US_Small_Val + b*US_Small_Gr + c*US_Large_Gr + d*_US_Large_Val;
fit AMP_AIT;
run;
quit;

Re: constrained regression

Amalik — Tue, 10 Jan 2017 03:44:01 GMT

Hi PG,

I have used the following code but the restriction a + b + c + d = 1 is not being incorprated in my results.Can you identify the possible issue?

proc nlin data=temp1 noprint outest=Out_Estimates ;
parms a=.25 b=.25 c=.25 d=.25;
bounds 0 <= a b c d<= 1;
restrict a + b + c + d = 1;
model AMP_AIT = a*US_Small_Val + b*US_Small_Gr + c*US_Large_Gr + d*_US_Large_Val;
run;
quit;
title 'out_estimates, loop ' &i;
proc print data=Out_Estimates;run;
run;
quit;

Re: constrained regression

PGStats — Wed, 11 Jan 2017 03:29:28 GMT

restrict is not a valid statement in proc nlin.

You might want to try proc hpnlmod instead.