Very good question and a lot of insightful answers.
Reposted from Seema Jayachandran
Question about std errors:
In an RCT, treated people are randomly assigned to "pods" and interact (e.g., treatment = WhatsApp group membership). The norm (e.g. Cai-Szeidl) is to cluster on indiv for control group + on pod for treatment group. Why is clustering needed if any ICC is due to treatment?

Comments