Causal Inference for Data Science

Knowing why something happens gives you the power to change what happens next. Causal Inference for Data Science is an approachable introduction to working out causality and estimating effects with statistics and machine learning. It targets a common business problem: A/B tests and randomized controlled trials are costly and often impractical, so the book teaches methods for finding causes in data even when nobody ran an experiment.

The book rests on a simple approach built around Directed Acyclic Graphs (DAGs), and it does not require advanced statistics or high-level math. Readers learn to model reality with causal graphs, estimate causal effects with statistical and machine learning techniques, and judge whether they have enough variables for an analysis. They also learn to decide when an A/B test, causal inference or machine learning is the right tool, and to explain the objectives, assumptions, risks and limitations of their work. The central idea is that you can predict events without knowing their causes, but understanding causality also lets you intervene to change outcomes.

The examples come from ordinary business settings, such as assessing advertising performance, choosing effective health treatments and setting product prices. The aim is to help readers interpret historical data, understand customer behavior and give management a sound basis for decisions.

The book has three parts. Part 1 covers the basics of causality, working with confounders, applying causal inference, and how machine learning and causal inference support each other. Part 2 covers propensity scores for finding comparable cases, direct and indirect effects with linear models, complex graphs, and the DoubleML library. Part 3 covers instrumental variables, the potential outcomes framework, and the effect of a time-related event. Appendices cover the math behind the adjustment formula, solutions to the chapter 2 exercises, and technical lemmas and a proof for the doubly robust estimator.

It is written for data scientists, ML engineers and statisticians. The author, Aleix Ruiz de Villa Robert, is a data scientist with a PhD in mathematical analysis from the Universitat Autònoma de Barcelona.