[CEdward]

Is there a way to do that for different numbers of clusters? I think for n = 6 clusters the above solution works fine. But I want to equally distribute the number of clusters by the number of steps for different numbers of clusters (e.g. n = 3 clusters is 1 cluster per step, n = 12 clusters is 4 clusters per step).

[CEdward]

Ultimately, I want to create a dataset where intrv = 1 whenever a cluster switches to the intervention at a time step and 0 otherwise. I want to have an even number of clusters switching to the intervention at each step. So in the example below, there are n = 3 clusters and each cluster (and the individuals in that cluster) switch at time 1, 2, and 3, respectively. I want to do the same thing for n = 3, 6, 9, 18, and 36 clusters. Is there a loop of some sort? Note: The u_3, u_2 variables below can be ignored.

clear
input float(cluster u_3 individual u_2 time intrv)
1 .1143134 1 -.7576806 0 0
1 .1143134 1 -.7576806 1 1
1 .1143134 1 -.7576806 2 1
1 .1143134 1 -.7576806 3 1
1 .1143134 1 -.7576806 4 1
1 .1143134 1 -.7576806 5 1
1 .1143134 2 .636223 0 0
1 .1143134 2 .636223 1 1
1 .1143134 2 .636223 2 1
1 .1143134 2 .636223 3 1
1 .1143134 2 .636223 4 1
1 .1143134 2 .636223 5 1
2 .0930611 1 -.50105065 0 0
2 .0930611 1 -.50105065 1 0
2 .0930611 1 -.50105065 2 1
2 .0930611 1 -.50105065 3 1
2 .0930611 1 -.50105065 4 1
2 .0930611 1 -.50105065 5 1
2 .0930611 2 -.3535455 0 0
2 .0930611 2 -.3535455 1 0
2 .0930611 2 -.3535455 2 1
2 .0930611 2 -.3535455 3 1
2 .0930611 2 -.3535455 4 1
2 .0930611 2 -.3535455 5 1
3 .4748023 1 .29760918 0 0
3 .4748023 1 .29760918 1 0
3 .4748023 1 .29760918 2 0
3 .4748023 1 .29760918 3 1
3 .4748023 1 .29760918 4 1
3 .4748023 1 .29760918 5 1
3 .4748023 2 .8398623 0 0
3 .4748023 2 .8398623 1 0
3 .4748023 2 .8398623 2 0
3 .4748023 2 .8398623 3 1
3 .4748023 2 .8398623 4 1
3 .4748023 2 .8398623 5 1
end
[/CODE]

Would it even be necessary to go through the trouble of creating a variable indicating the randomization sequence (or group) for a cluster?

[CEdward]

Here is the solution for anybody that is interested in using this design:

set obs `num_clus'  //*Number of clusters*//
qui gen cluster = _n
qui gen group = 1+mod(_n-1,3) //*Where 3 is the number of steps*//
expand `clussize'
bysort cluster: gen individual = _n
expand 6 //*Number of periods//
bysort cluster individual: gen time = _n-1 //*Set up time variable*//
gen intrv = (time>=group) //*Intervention variable*//