Matched market testing
Matched market testing measures causal impact by pairing geographic markets with similar historical behavior, running a campaign in the test market while holding out its matched control, and attributing the post-period difference to the campaign. It is the practical workhorse for offline and channel-level incrementality where user-level randomization is impossible — closely related to geo experiments and the synthetic control method.
How markets are matched
You select candidate markets (cities, regions, DMAs) and pair them by similarity in historical outcomes — baseline sales, traffic, seasonality. The campaign runs in the test market; its matched control receives no change.
Because the pair tracked together before the test, their pre-period similarity is the basis for expecting them to continue together absent the campaign.
Reading the lift
After the campaign, you compare the test market's outcome to the control's. The difference, beyond the pre-period baseline relationship, is the estimated incremental effect. Difference-in-differences and synthetic control are common ways to formalize that comparison and build a more robust control from many markets.
The method's validity rests on the match: if test and control diverge for unrelated reasons, the estimate is biased — so pre-period parallelism is checked before trusting the result.
- Pair markets by similar pre-period behavior
- Run campaign in test, hold out matched control
- Post-period difference net of baseline = lift
How it appears in analytics and logs
A sustained post-period gap between matched test and control markets is causal evidence of the campaign's effect, net of shared baseline trends.
Diagnostic use case
Measure incrementality for TV, radio, or whole-channel changes by comparing a test region against a matched holdout region.
What WebmasterID can help detect
WebmasterID's first-party, geo-aware traffic and conversion data can serve as the regional outcome metric for test and control markets.
Common mistakes
- Pairing markets that did not track together beforehand.
- Ignoring concurrent local events that break the match.
- Reading the raw difference without the baseline relationship.
Privacy and accuracy notes
Markets are compared at aggregate region level, with no individual tracking. Educational, not legal advice on experiment design.
Related pages
- Geo experiments for measurement
A geo experiment divides geographic regions into a treatment group (which sees a media change) and a control group (which does not), then compares outcomes between them. Because assignment is at the region level rather than the user level, geo experiments measure incremental effect without needing cookies, device IDs, or per-person attribution — making them a privacy-resilient complement to touch-based models.
- Synthetic control method
The synthetic control method estimates causal impact by constructing a 'synthetic' version of the treated unit — a weighted blend of comparison units that closely matches its pre-intervention behavior. The gap between the real treated outcome and its synthetic counterfactual after the intervention is the estimated effect. It is widely used in geo-experiments where a single market is treated. This page explains the construction and its assumptions.
- Difference-in-differences for measurement
Difference-in-differences (DiD) is a quasi-experimental method that estimates the causal effect of an intervention — like turning a campaign on in some regions — by comparing how a treated group changed against how an untreated control group changed over the same time. By differencing out both pre-existing gaps and shared time trends, DiD isolates the incremental effect. This page explains the method, its key assumption, and where it fits in measurement.
- Multi-site analytics
Compare regional outcomes for test and control markets.
Sources and verification notes
- Google — Geo experiments methodology (research paper)Foundational method for matched-market geo measurement.
Last reviewed 2026-06-24. Facts are checked against primary/official sources where available; uncertain specifics are marked “Data not yet verified” rather than guessed.