Welcome to The Logic of Experiments, the first module of the course. Before any test statistic or p-value, there’s a more fundamental question: why do we run experiments at all? The answer is that an experiment is the only reliable way to move from “these two things move together” to “this change caused that result.” Get this logic right and everything later — sample size, tests, pitfalls — has a solid foundation. Get it wrong and the fanciest statistics just put a confident number on a misleading comparison.

You’ll start by seeing why correlation isn’t causation — and how a hidden confounder makes an observational comparison lie, using a simulation you can run yourself. You’ll meet the control and treatment groups that form the A/B frame, and then see why randomization works: randomly assigning users balances confounders you know about and ones you don’t, so the difference you measure is caused by the change. You’ll put the pieces together into the anatomy of an A/B test, and finish with a guided project running Lumen’s first simulated experiment end to end.

Every demonstration in this module is real, runnable Python on seeded synthetic data — the observational-vs-randomized comparison and Lumen’s first experiment both produce the exact numbers you’ll see, and you can rerun them yourself. Start with Lesson 1 on why we run experiments.