Causal Methods
A track of P8 · Causal Inference & Experimentation.
What to do when you cannot randomize: confounders, diagrams, matching, difference-in-differences and uplift — causal claims earned, not assumed.
Randomization is the gold standard, and most of the time you cannot have it. You cannot randomly assign people to smoke, cities to a minimum-wage law, or customers to a recession. This track is the craft of recovering a causal claim from data the world handed you un-randomized — the difference between "these move together" and "this causes that".
It opens with the map: causal diagrams that make confounding visible, and the one rule that catches most beginners — control the confounders, but never control a collider, because doing so manufactures a correlation that was not there. From there, three workhorses: matching, which compares like with like; difference-in-differences, which cancels a shared time trend to read an effect off a policy change; and uplift, which reframes the question from "did it work?" to "for whom?" — separating the persuadable from the sure thing and the lost cause. Each method buys its honesty with an assumption, and the track is as much about seeing those assumptions break as about applying the formulas.