Based on what I’ve learned from Zoe, and what I’ve read to fill in the gaps, I think the two big differentiators are “how much statistical rigor do you require” and “how many different variations of the question do you need to ask?”