Run this notebook: Open in Colab Open in Kaggle

Phase 27: Causal Inference — Start Here

Move beyond correlation — learn to estimate true cause-and-effect relationships from data using experiments, DAGs, and counterfactual reasoning.

Correlation vs. Causation

Standard ML finds correlations: “users who see this ad also buy product X”.

Causal inference asks: “Does showing this ad cause users to buy product X?”

This distinction matters enormously for decisions, policy, and AI alignment.

Notebooks in This Phase

Notebook	Topic
01_causal_fundamentals.ipynb	Potential outcomes, counterfactuals, ATE
02_causal_graphs_dags.ipynb	DAGs, d-separation, backdoor criterion
03_experimental_design.ipynb	Randomized experiments, A/B testing
04_observational_methods.ipynb	Propensity scores, matching, IV
05_advanced_topics_applications.ipynb	Heterogeneous treatment effects, causal ML
06_quasi_experimental_designs.ipynb	Diff-in-diff, regression discontinuity

Key Concepts

Concept	Description
Potential outcomes	Y(1), Y(0) — what would have happened
ATE	Average Treatment Effect
DAG	Directed Acyclic Graph — causal structure
Confounders	Variables affecting both treatment and outcome
Propensity score	P(treatment
IV	Instrumental Variables — for unmeasured confounders
Diff-in-diff	Before/after comparison with control group

Prerequisites

• Statistics and probability

• Linear regression

• Python/pandas/statsmodels

Learning Path

01_causal_fundamentals.ipynb         ← Start here
02_causal_graphs_dags.ipynb
03_experimental_design.ipynb
04_observational_methods.ipynb
06_quasi_experimental_designs.ipynb
05_advanced_topics_applications.ipynb

Recommended Tools

• dowhy — Microsoft’s causal inference library

• econml — Causal ML (HTE estimation)

• causalml — Uplift modeling

• pgmpy — Probabilistic graphical models