Geo‑Lift testing splits markets into test and control ______.
browser cohorts
geographies (e.g., DMAs)
time buckets
device types
By manipulating spend in spatially separate areas, analysts infer causal impact. Geographic isolation reduces user cross‑over contamination.
2025 guidelines recommend at least ______ geos per cell for adequate statistical power.
20
100
3
2
With fewer than ~20 cells, variance dominates and lift estimates become noisy. More geos stabilise the distribution of local factors.
A **calibration period** ensures test and control behave similarly ______ the intervention.
after
long after
during
before
Analysts check pre‑period sales trends to validate that groups are comparable. Significant pre‑period gaps would bias lift results.
Spillover can bias geo tests when advertising reaches users in ______ regions.
test
control
both test sub‑DMAs
remote
If media spills into control areas, the incremental difference shrinks. Careful media planning or buffered geos mitigate this leakage.
Recommended spend differential between test and control is at least ______ %.
5
70
10
25
Large contrast generates detectable signal over market noise. Under‑powered spend shifts often produce inconclusive lift.
Key KPI is often normalised as sales per geo divided by ______.
GDP
CPM
search volume
population or store count
Normalisation controls for size differences across regions, enabling fair comparison.
After campaign ends, analysts keep measuring for a **washout period** to capture ______ impact.
viewability decay
lagged carry‑over
funnel steps
bot clicks
Some channels keep influencing behavior after spend stops; including washout captures the tail effect.
The primary statistical test for geo lift is often a ______ t‑test across geos.
Welch (unequal variance)
paired
one‑sided Z
Chi‑square
Welch t‑test does not assume equal variance between groups, fitting heterogeneous regional data.
Heterogeneity among geos can be modelled with a ______ regression to estimate lift.
Quantile
Bayesian hierarchical
K‑means
Linear probability
Hierarchical models borrow strength across regions, yielding more stable causal estimates.
If lift is +8 % with 95 % CI of (‑2 %, 18 %), the result is deemed ______.
negative
not statistically significant
positive significant
under‑powered but conclusive
Because confidence interval crosses zero, we cannot reject the null of no effect.
Starter
Review the basics.
Solid
Good job—refine for mastery.
Expert!
Outstanding performance.
