topic How to ensure no statistical differences between groups in covariates when simulating data? in SAS Programming

How to ensure no statistical differences between groups in covariates when simulating data?

whymath — Fri, 30 May 2025 08:10:48 GMT

I am simulating data for a 2-groups trial with 100 subjects in each group. There are four variables: GROUP, SUBJNO, SEX, AGE. SEX and AGE are covariates for following model.

Here is my code:

%let seed=12345;

data tab1;
  call streaminit(&seed.);
  length subjno $4 sex $1;
  do _n_=1 to 100;
    group=1;
    subjno='1'||put(_n_,z3.);
    sex=choosec(rand('table',0.5,0.5),'M','F');
    age=rand('integer',18,70);
    output;
  end;

  do _n_=1 to 100;
    group=2;
    subjno='2'||put(_n_,z3.);
    sex=choosec(rand('table',0.5,0.5),'M','F');
    age=rand('integer',18,70);
    output;
  end;
run;

Unfortunately, The initial random seed, 12345, causes SEX statistical difference on different group, the p value of Chisq is 0.0477.

I have tried new seed value like 123, 1234, 123456, 1234567 and they will not cause SEX statistical difference on different group.

I know there is a possibility that statistical difference of covariates happens. Is there a way to ensure no statistical differences between groups in covariates when simulating data?

Maybe block randomization with covariates as block factor? What about continous covariate variable like AGE?

Re: How to ensure no statistical differences between groups in covariates when simulating data?

Ksharp — Fri, 30 May 2025 09:26:49 GMT

You want this ?

proc plan seed=123;
factors group=2 ordered subj=100 /noprint;
output out=tab1;
quit;
data tab2;
set tab1;
if subj in (1:50) then sex='F';
 else sex='M';
run;
proc freq data=tab2;
table group*sex/ chisq;
run;

Re: How to ensure no statistical differences between groups in covariates when simulating data?

whymath — Fri, 30 May 2025 09:41:17 GMT

Thank you for your reply, would 50:50 be kind of sedulously? And what about continous covariate variable like AGE?

Re: How to ensure no statistical differences between groups in covariates when simulating data?

Ksharp — Fri, 30 May 2025 09:52:25 GMT

Not really,you could use 49: 51 or 48:52 .
About AGE you could try NORMAL distribution other than uniform distribution.