When is an A/B test not feasible, and what is DiD?

This tests your grasp of causal inference for non-randomized changes, crucial for measuring large-scale or infrastructural launches. Start with a scenario where user-level randomization is impossible (e.g., a state-level rollout). Introduce Difference-in-Differences (DiD) to compare the *change* in a treatment group vs. a control group over time. Crucially, explain the 'parallel trends' assumption: that both groups would have behaved similarly without the intervention.