Causal Inference

My running notes on causal inference, in the order I’d hand them to someone starting out.

Most of what I write here circles back to one question: when can you actually call something a cause, and not just a correlation that happened to show up in the data?

These are the posts where I work through that, roughly in the order I’d read them.

Start with the cheatsheet for the vocabulary and the assumption behind each method. Then assessing overlap goes deep on one specific diagnostic, whether your treatment and control groups are even comparable to begin with. From there, omitted variable bias asks the uncomfortable follow-up: how much could a confounder you never measured be moving your answer. And recommendations as treatments is where it gets applied, treating a recommender system as a policy you can study with the same tools.