Why Do Old Regressions Still Work in Experimentation - Regression Adjustments and CUPED in Modern A/B Testing

Explore the enduring relevance of traditional regression methods in modern experimentation through this 34-minute conference talk by Mark Eltsefon from Meta, delivered at the Data Science Festival Game On event 2025. Discover how regression adjustments, CUPED (Controlled-experiment Using Pre-Experiment Data), and their variations continue to provide value in contemporary data science practices. Learn how these established techniques effectively leverage pre-experiment data to reduce variance and increase experimental sensitivity, even alongside advanced machine learning and causal inference methods. Examine the theoretical foundations underlying regression adjustments and gain practical insights into CUPED applications. Compare these traditional approaches with newer methodologies in terms of robustness, interpretability, and scalability. Understand effective implementation strategies for these techniques while recognizing their inherent limitations. This introductory-level session requires some technical knowledge and provides valuable insights for students and practitioners seeking to understand why classical statistical methods remain relevant in modern experimental design and analysis.