Spotify Research introduces an improved synthetic control methodology for measuring product impact when A/B testing isn't feasible. The approach uses forecasting to automatically detect spillover effects and select valid donor controls, eliminating reliance on domain knowledge alone. A new theorem based on proximal causal inference enables data-driven donor selection by testing if pre-intervention forecasts match post-intervention reality. The method also debiases estimates by leveraging excluded donors' pre-intervention data, providing sensitivity analysis frameworks to handle selection errors and latent confounding.

7m read timeFrom research.atspotify.com
Post cover image
Table of contents
IntroductionThe problemConclusionR eferences

Sort: