Causal estimands and estimation

This section describes how Meridian defines the primary estimands of interest, including incremental KPI, ROI, marginal ROI, and response curves. These quantities are defined using potential outcomes and counterfactuals, which are the language of causal inference.

With clear estimand definitions in place, you can review the assumptions required for the MMM to provide valid inference. These assumptions help ensure that the model is actually able to estimate these quantities. If assumptions are not met, then estimates can be severely biased.

We recommend that you clearly define causal estimands and required assumptions for any MMM methodology. If this is not done, then the model results are likely to be misinterpreted. Even more impactful, ignoring the required assumptions can render the analysis practically nonsensical due to severe underlying bias.

The definitions in the following section don't rely on any aspects of the Meridian model specification. The same definitions can apply to any MMM. Defining the causal estimand is crucial for any MMM analysis so that the results are interpretable, and to help determine whether a particular model is appropriate for the analysis and under what assumptions.