A slightly edited section of my comment on the earlier draft:
I lean skeptical about “relative pair-wise comparisons” after participating:
I think people were surprised by their aggregate estimates (e.g., I was very surprised!);
I think later convergence was due to common sense and mostly came from people moving points between interventions and not from pair-wise anything;
I think this might be because I am unconfident about eliciting distributions with Squiggle. As I don’t have good intuition about how a few log-normals with 80% probability between xx and yy would compare to each other after aggregations (probably this is common, see 2a).
After I did my point estimates + my CI via Squiggle for everything alltogether, I think they didn’t match each other that well. Maybe that’s because lognormal is right-skewed and fairly heavy-tailed?
A slightly edited section of my comment on the earlier draft:
Thanks Misha