Eric walks through the two ways to build the comparison in a geo test, where each one breaks, and how your own history picks between them.
Key moments
- 0:00 How a geo test works: change something in some places and compare them with places that historically behaved alike.
- 0:54 Why it helps for offline and video media, where clicks lose the trail before the sale.
- 1:25 Where matched markets break: local shocks like a storm, or a market like New York City with no twin.
- 2:00 Synthetic control: build the comparison from weighted fractions of many places.
- 3:08 How your own history picks between them: MAPE (how tightly they track) and correlation (whether they move together).
- 4:50 When you can, use both comparisons on the same test.
Read the related guide: Matched Market Test vs. Synthetic Control