Subtract the control-region change from a before-and-after result

A region’s conversion total can rise while a campaign contributes little to the change. This lab calculates an additive difference-in-differences contrast. It deliberately stops short of calling the result causal, because parallel trends and stable populations require evidence outside a four-cell worksheet.

Define the calculation before using it

Enter treatment-region outcomes before and after the intervention, plus equivalent control-region outcomes. Subtract the treatment baseline from its later value, do the same for the control, then subtract the control change from the treatment change. Keep periods equally defined and do not change the measured conversion or geographic boundary halfway through the exercise.

Work through the synthetic example

The teaching example moves the treatment region from 100 to 140 conversions and the control from 80 to 100. The raw treatment increase is 40, but the additive contrast is 20 after subtracting the control’s increase. A simultaneous product release or differential seasonal effect could still explain some or all of that contrast.

Use the artifact and preserve its assumptions

Open the editable calculator to change the inputs and inspect the sensitivity view. The CSV records synthetic inputs and expected outputs; the JSON fixture keeps the equations available for reproduction. These calculations have been checked against the stated example. No measured campaign data is included.

The sensitivity rows vary only ca by 20% below and above the entered value. They are scenarios, not confidence limits or a forecast distribution. A row outside the model’s constraints is labelled rather than turned into a plausible-looking result. Save the chosen inputs with the decision so another reader can distinguish a changed assumption from a changed formula.

Evidence and limits

Google’s export schema separates the reporting date, UTC timestamp and event parameters. The worksheet uses simplified aggregate inputs; it does not claim that an export already contains a correctly reconciled cohort. See Google Analytics BigQuery Export schema, especially “event_date, event_timestamp, event_value_in_usd and event_params fields”.

This is an unnormalised count model. Unequal scale, changing population, pre-trend differences and interference between regions need a more appropriate design before the result informs a causal budget decision.

Background: Geo-lift tests: a practical guide for UA teams and Incrementality tests you can afford: geo holdouts, ghost ads, and what breaks. These existing articles provide context; the present calculation does not verify every archived claim.