Estimating individual effects in group-level experiments

Michael Weiksner writes,

I [Weiksner] do research on deliberation, where the treatment itself is defined as the interaction with other people (who are inevitably also randomly assigned to the treatment group). Because all the treated individuals interact, I know that the safest course of action is to look only at group level effects. But that’s highly unsatisfying, since you can’t really shed any light on questions about individuals, like does deliberation create better citizens?

It seems that SUTVA [the “stable unit treatment value assumption,” related to the assumption of no interference between units] is an all or nothing thing, which is highly unsatisfying. Is there anyway to correct for this kind of violation in SUTVA? How, if at all, can you report results that those randomly assigned to the deliberation show significant gains in knowledge versus a control group (as an example)?

My first response is: yes, the cleanest approach is to look only at group-level effects. You can still look at individual-level outcomes: for example, does a certain type of deliberation increase or decrease the average knowledge of the participants. There’s no problem with looking at measurements on individuals and then combining them to get group-level outcomes, so that “n” is the number of groups in the experiment.

Beyond this, you can certainly set up models for the interactions of individuals within a group. It doesn’t sound like this is necessary for the questions you’re interested in, though.