stratified block randomization with proportional allocation

[hyunjeehale]

I am needing to develop a randomization table using stratified block randomization with proportional allocation for an outside group to use. We know the general population proportions among the sample we are screening participants from for the two strata variables we want to use.

Treatment groups: 3

Strata 1: 2 levels (proportions of each level: 50%, 50%)

Strata 2: 3 levels (proportion of each level: 17%, 36%, 47%)

Total participants: 30,000

I have been using PROC PLAN to develop my randomization table, but could not find a solution that allowed me to account for the proportional breakdown of each strata. I found some references for PROC SURVEYSELECT and proportional allocation, however I could not find a way to incorporate blocks.

Using PROC PLAN, I am able to generate a randomization table of most of what I need, but the treatment groups are distributed evenly across the unique strata combination

What I get:

Treatment 1 (same for treatments 2 and 3)

                  Strata 2

                   Level 1       Level 2         Level 3

Strata 1

Level 1        1680         1680              1680

Level 2        1680         1680               1680

What I Need

Treatment 1 (same for treatments 2 and 3)

                  Strata 2

                   Level 1       Level 2         Level 3

Strata 1

Level 1        571        1210              1579

Level 2        571         1210               1579

Thank you for any help!

[ballardw · Accepted solution]

See if this come close to what you want:

data want;
   set have;
   strata1 = rand('table',.5,.5);
   strata2 = rand('table',.17,.36,.47);
run;

The rand('table') function uses a list of proportions to randomly assign values of 1 , 2, 3 etc depending upon how many parameters are used. Since it is random you may not get exactly the number of records per cell though with large data set and not really small percentages you should get close. If not close enough re-run until they are .