What we’ve learned in this book
Back in Chapter 1, we started our causal journey by asking about the reasons to use causal modeling rather than traditional machine learning, despite the tremendous success of the latter.
We defined the concept of confounding and showed how it can lead us astray by producing spurious relationships between causally independent variables. Next, we introduced the Ladder of Causation and its three rungs – observations, interventions, and counterfactuals. We showed the differences between observational and interventional distributions using linear regression.
After that, we refreshed our knowledge of the basic graph theory and introduced graphs as an important building block for causal models. We discussed three basic conditional independence structures – forks, chains, and colliders, and showed that colliders have a special status among the three, allowing us to infer the direction of causal influence from the data.
Next, we...
