'Conditional' incomplete block design

[spirto]

hi all,

im not sure where to post this so ill start here.

i want to create an incomplete block design. i have 8 treatments A-H. i need to create 3 blocks of size 2.

now here is where im getting it stuck. there are treatment pairs that cannot be separated. in other words, A/B need to stay together, C/D, E/F, and G/H. so out of those four pairs i need to pick 3 pairs and i have to replicate this 50 times. i need all the pairs to be distributed equally and uniformly.

so for example replicate 1 may look like

Block 1	Block 2	Block 3
A	E	C
B	F	D

replicate 2 would may like

Block 1	Block 2	Block 3
E	A	C
F	B	D

any thoughts???

[data_null__]

If the pairs can't be separated doesn't that mean you really only have 4 "treatments"?

[Haikuo · Accepted solution]

Since you haven't stated clearly if the order of treatments matters, so here is the combination version of the code. If it does matter, then use permutation version of it.

data _null_;

   array x(4) $2  ('AB' 'CD' 'EF' 'GH');

   seed = 1234567890123;

   do n=1 to 50;

      call rancomb(seed, 3, of x1-x4);

      put seed= @20 ' x= ' x1-x3;

   end;

run;

Haikuo

[SteveDenham]

Thanks !  This was the perfect code reminder for a completely separate place and project--I had forgotten about rancomb and ranperm, and PROC PLAN was driving me nuts.

Steve Denham

[Rick_SAS]

Not only random permutations, but also ordered sets of permutations:

http://blogs.sas.com/content/iml/2010/10/13/generate-all-permutations-in-sas/

By the way,  in SAS/IML software the ALLCOMB, ALLPERM, RANCOMB, and RANPERM functions work on arbitrary sets, not just integers 1:n.