Predictive Incrementality by Experimentation (PIE) for Ad Measurement
Published in Working Paper, available at arXiv, 2023
The measurement of advertising effectiveness is a central problem in marketing. The gold standard for measurement is a randomized controlled trial (RCT), which provides an unbiased estimate of the average treatment effect (ATE). However, RCTs can be expensive and time-consuming. In this paper, we propose a new method, Predictive Incrementality by Experimentation (PIE), that combines the strengths of experimental and non-experimental approaches. PIE uses a small-scale RCT to build a predictive model of the individual-level treatment effect, which is then used to estimate the ATE in a larger, non-experimental population. We show that PIE can provide more precise estimates of the ATE than a small-scale RCT alone, while being less biased than purely non-experimental methods. We demonstrate the performance of PIE using a large-scale field experiment on Facebook.
Recommended citation: Gordon, B. R., Moakler, R., & Zettelmeyer, F. (2023). "Predictive Incrementality by Experimentation (PIE) for Ad Measurement." arXiv preprint arXiv:2304.06828.
Download Paper