How do you model social feedback loops when data is limited?

I study how social norms and institutional rules feed back on each other, and I often have to build models with much less data than I would like.

I am curious how others in sociocybernetics decide when a model is good enough to publish or to use in policy advice.

What has worked for you when you need to justify a model built on sparse or qualitative data?