Improvement in Minimum Detectable Effects in Randomized Control Trials: Comparing User-Based and Geo-Based Randomization

In digital advertising, randomized control trials (RCTs) are a fundamental method for evaluating campaign effectiveness. The randomization approach, whether user-based or geo-based—can substantially impact the statistical power and precision of detecting treatment effects. This paper explores the influence of these randomization methods on the Minimum Detectable Effect (MDE), focusing on how intra-cluster correlation (ICC) in geo-based randomization inflates variance and increases sample size requirements. Through mathematical modeling and simulation, we demonstrate that user-based randomization is more efficient, particularly for detecting small advertising effects, and offer recommendations for optimizing experimental designs in digital advertising contexts. Keywords: Causal inference, Randomized Control Trials, Digital Advertising, Minimum Detectable Effect