Let's say I have units that need to be split into two groups for a randomized control trial. One way is I randomly sample units and assign them into 100 groups and take first 20 groups to be my treatment. Second way is I randomly sample units so 20% of units are treatment. Now, the first way sounds like clustered randomization, but it's not. so I wonder, is clustered randomization becomes issues only when each cluster has some shared properties within them? If each cluster are consists of randomly selected units, would it be still considered as clustered randomization even if it looks like it? Want to get more thorough answer.
I tried reading some papers around clustered randomization