Multicell experiments for marginal treatment effect estimation of digital ads

The average treatment effect (ATE) from an A/B test is the standard for measuring the causal impact of an intervention such as an ad campaign. The ATE is the key input into a decision of whether to run a campaign. However, the ATE is not sufficient for decisions of how much to spend on a campaign because it averages over individuals with different levels of responsiveness to advertising. The key quantity of interest for a spending decision is the marginal treatment effect (MTE), which is the ATE for individuals at the margin of being treated. In this paper, we develop a new experimental design, called a multicell experiment, that allows for the estimation of MTEs in the context of digital advertising. The design involves randomly assigning individuals to multiple treatment cells, each with a different level of ad exposure. We show how to use the data from a multicell experiment to estimate the MTE function, which relates the treatment effect to the level of ad exposure. We apply our method to a large-scale multicell experiment for a CPG brand.

Recommended citation: Waisman, C. & Gordon, B. R. (2024). "Multicell experiments for marginal treatment effect estimation of digital ads." Management Science. (forthcoming).
Download Paper