Randomization

[Cyrus Muriithi]

Hello, I am trying to randomize 100 obs into 3 groups with equal changes of being in either group.
Group 1 = treatment
Group 2= Control 1
Group 3= Control 2
I am trying to improve my current formula but in vain
Here is my Current formula which is only generating only two groups

gen number = uniform()
centile (number), centile(50)
gen treatment=1 if number>=r(c_1)
replace treatment=2 if number<r(c_1)

[Nick Cox]

The centile calculation is wrong in principle as the median you calculate (50% percentile) has no bearing on getting three approximately numerous groups. And you don't need it any way.

The best you can do with 100 is 33 : 33 : 34, e.g. by

set seed 2803 
gen double number = runiform() 
sort number 
gen treatment = cond(_n <= 33, 1, cond(_n <= 66, 2, 3))

How you set a seed is immaterial, but setting a seed is a good idea for reproducibility.

[Cyrus Muriithi]

Thank you Nick Cox, Its working correctly. God Bless You.

[Robin Fatch]

Hi Nick,

I have a similar situation to Cyrus' post, although somewhat different.

I would like to randomly assign 300 observations to 3 groups, using block sizes of 9 for the randomization. (ie. randomize the first 9 observations to the 3 groups, then randomize the second 9 observations, etc) I found some code on the old Stata forum and have modified it here, to work for my situation. I think it is doing what I want, but I just wanted to confirm that this approach seems appropriate.

Thanks!
Robin

clear
set obs 300
egen arm = seq(), to(3)
egen block = seq(), block(9)
set seed 15528
gen random = uniform()
bysort block (random): gen byte seqn = _n
*bysort block (seqn): list seqn arm random 

[Cyrus Muriithi]

Hello Robin, thanks for that i also got a way using:

bysort demo_district demo_treatment: gen random = runiform()
sort random
bysort demo_district demo_treatment: gen insample = _n <= 25
*br demo_district random insample
keep if insample==1