Impact intelligence, with evidence.

Explore treated and comparison groups, inspect pre-trends, and test assumptions before interpreting estimated programme impact.

DatasetSegmentIntervention at

Observed outcome trajectories

Unit-averaged outcomes across periods, by exposure group

GROUP TRENDS

Treated groupComparison groupIntervention starts

Difference-in-differences

Before/after changes compared across groups

01 / DESCRIBEMeasure the change

Compare mean outcomes before and after the intervention.

02 / VERIFYInspect pre-trends

Check whether trends diverged before the policy began.

03 / CHALLENGEStress-test assumptions

Run a fake intervention date and vary the comparison group.

Group-level trajectories

Changing the intervention date redraws the baseline

Pre-treatment gap

Treated minus control before intervention

Period-by-period observations

Means and sample sizes are calculated from eligible units

The DiD calculation

The group change is estimated with equal weight per unit

Cluster bootstrap

Resamples whole units, not individual time observations

A bootstrap percentile interval is descriptive uncertainty under resampling assumptions. It is not evidence that parallel trends holds.

Relative-period effects

Treated-control gap minus the gap immediately before treatment

Reading the event study

Reference period = −1, set to zero by definition

Dynamic contrasts table

Uncertainty bands are not plotted for these contrasts. They should not be interpreted as confirmed period-specific effects.

Stress-test the counterfactual

Compare placebo, alternative windows and segment-specific estimates

Assumption checklist

These are interpretive requirements, not automatic guarantees

Import your own panel data

CSV with unit, group, period, outcome and segment

Choose a CSV file

Maximum 20,000 rows. Each unit has one observation per period.

An invalid CSV is rejected without replacing your current data.

For sensitive research data, use anonymised records and review the export files before sharing. Computation occurs in your browser.

Research export center

Download the current analysis and reproducibility details

Statistical methodology

Two-group panel DiD with unit-weighted pre/post means; whole-unit cluster bootstrap; event-time differences referenced to the final pre-period; placebo inside the observed pre-period. Requires credible parallel trends, no anticipation, limited spillovers and stable composition.

Synthetic examples are for teaching and software demonstration only. Importing data does not turn associations into verified causal effects.

Data preview

First 10 records currently used in the application