topic Re: simulating reproducible samples in SAS Programming

simulating reproducible samples

SimRock — Thu, 03 Nov 2016 02:12:25 GMT

Hi,

I am simulating multiple datsets (and concatinating them using the proc append within the macro) using the same seed in the call streaminit but my answers are different. Could anyone suggest a solution

Are they any reasons why my answers my not be reproduced that I am unaware?

Data param;
pr_male=0.7;
pr_all=0.3;
run;

%macro sim (SSB1=, SSB2=, FFV1=, FFV2=,interv=);

Data have&interv ;
set param;
call streaminit(354);
interv=&interv;
do env=1 to 3;
all=rand('bernoulli', pr_all);
do id=1 to 100;
Male = rand('bernoulli', pr_male);
do t=1 to 2;
if t=1 then do;
SSB1 = &SSB1;
FFV1 = &FFV1;
end;
else if t=2 then do;
SSB2 = &SSB2;
FFV2 = &FFV2;
end;
output;
end;
end;
end;
run;

%mend sim;

%sim(SSB1=1, SSB2=0, FFV1=1, FFV2=0,interv=0);
%sim(SSB1=0, SSB2=1, FFV1=0, FFV2=0, interv=1);

thank you

Re: simulating reproducible samples

PGStats — Thu, 03 Nov 2016 03:01:10 GMT

Try it again. Make sure there are no errors or warnings in the log. My own test generates two identical streams for pseudo random variables all and Male.

Re: simulating reproducible samples

Reeza — Thu, 03 Nov 2016 11:08:59 GMT

@PGStats would you get the same streams if run on different system? I assume this is true but can't test at the moment.

Re: simulating reproducible samples

PGStats — Thu, 03 Nov 2016 16:15:00 GMT

Using streaminit is supposed to guarantee deterministic sequences, at least on a given system (hardware/version). But I wouldn't assume it does across systems.

Re: simulating reproducible samples

SimRock — Thu, 03 Nov 2016 17:51:29 GMT

Thank you for your help. I appreciate it