Research: measure before you attribute

This collection compares marketing-measurement methods through the decision they can support, their data requirements and their limits. A contribution attributed by a platform is not automatically a causal effect.

Start with the decision

Describe whether the task is to explain a variation, estimate an incremental effect, forecast a scenario or allocate a constrained budget. These questions require different counterfactuals, time horizons and units. Do not select a method from its label alone.

Check the data before the model

Document time grain, economic units, media calendars, price, promotions, availability, distribution and measurement breaks. A long history does not repair missing variation or a definition change. Inspect the documented dataset before modelling.

Compare evidence, not dashboards

Attribution, controlled experiments and Marketing Mix Modeling can disagree without one necessarily being wrong. Compare their population, timing, comparison group and uncertainty. Read the measurement framework before moving a budget.

Collection status

The collection is being expanded. Each future analysis must identify a primary source, state what it observes and separate a documented method from a reproduced empirical result.