Geo Test Power Calculator
How many DMAs does a geo experiment need before it can detect your lift? Tune market count, expected lift, and noise to find out.
Power now 0% Two-sided 5% approximation
Minimum DMAs for 80% 0 At current lift and noise
MDE at current N 0% Rough 80% detectable lift
How it works
What it computes
The calculator models a geo test as n treatment/control market pairs. Each market's outcome carries noise you set as a coefficient of variation (CV). Every number on screen comes from three normal-approximation formulas:
- SE = CV × √(2 / n)
- power = Φ(lift / SE − 1.96), a two-sided 5% test with the far tail ignored
- MDE at 80% power = 2.8 × SE, where 2.8 ≈ 1.96 + 0.84
Assumptions
- Markets are independent, equally weighted, and share the same known CV.
- The lift is one constant relative effect applied to every treated market.
- The estimator is a plain difference in means: no pre-period adjustment, no weighting.
Where it breaks
- Real geo tests usually add pre-period covariates or synthetic-control weighting, which cuts variance. Read the answer here as a conservative upper bound on markets needed.
- No spillover between markets, no shocks that hit every market at once (a national promo, a holiday), no unequal market sizes.
- At the low end of the slider the normal approximation gets rough. Treat small-n readings as directional.