Upload your market-by-period data with a treated flag and a before/after flag. Get the difference-in-differences estimate with a 95% interval, the relative lift, an event study, and an honest parallel-trends verdict. Free.
Free analyses run on up to 10,000 rows. Larger files are randomly sampled to that size — sign up to analyze your full dataset.
Measuring incremental lift against your control markets...
Sent to — the incremental lift with a 95% interval, relative lift, treated-vs-control trends, an event study, the difference-in-differences 2x2, a parallel-trends verdict, R code, and AI insights.
Analyze another fileThe incremental effect is the difference of two differences: how much the treated units changed from before to after, minus how much the untreated control units changed over the same periods. It is fitted as a two-way fixed-effects regression of the outcome on a treated-by-after indicator with a fixed effect for every unit and every period, so persistent size differences and shocks common to all units are absorbed first. Standard errors are clustered on the unit by a hand-computed CR1 sandwich, and the whole result is gated on a parallel-trends test fitted to the pre-intervention window alone.
Use it when some units were deliberately held back from a change and you need to know how much of the movement was caused by it rather than by everything else happening at the time.
Not for a single time series with no control group — that is an interrupted time-series question, and the counterfactual there is far weaker. Not when the treated units were chosen because they were already moving differently.
Built for: Marketers, growth teams and analysts reading out geo holdout or rollout tests
Typical data source: A market-by-week table of sales or conversions with a column marking which markets were treated and a column marking when the change started
One row per market per week, with the two flags:
Minimum 12 rows · Best with 20-200 units x 10-100 periods, with at least a third held back as controls
Standard-library analysis: how much of the lift was actually incremental. You turned something on in some markets and deliberately held others back — this measures the difference-in-differences between the two arms, so the untreated markets carry whatever would have happened anyway. Two-way fixed-effects estimate with unit-clustered 95% intervals, a parallel-trends integrity check that says out loud when the estimate is not credible, an event study showing when the effect started, and the 2x2 of cell means underneath it all.
The two arms plotted side by side — parallel before the intervention, separating after it, is what a valid test looks like.
When the effect actually started. Anything showing up before the intervention is a trend, not an effect.
The four cell means the estimate is built from, including how much the control markets moved anyway.
The integrity gate: whether the assumption the causal claim rests on survives contact with the pre-period data.
Plain-English interpretation — what the numbers mean, what's significant, and what to do next.
How much of the lift was actually incremental?
You turned advertising on in some markets and held others back. Map the market, period, outcome, treatment flag and after flag. You get the difference-in-differences estimate with a 95% interval, the relative lift, and — before any of it — whether the two arms were already diverging, which decides if the number can be read causally at all.
See our FAQ for details on pricing, data privacy, and how the analysis works. Every report includes a Methodology section showing the statistical test, assumptions checked, and diagnostics run.
Run any analysis on your own data — validated R analyses, interactive reports, AI insights, and PDF export.
Try Free — No Credit CardTell us what went wrong, in your own words. We capture the page you're on automatically, so no need to describe where you are.