Preface

Causal questions are at the heart of nearly all applied sciences. Does smoking cessation extend life? Does a minimum wage increase reduce employment? Does a treatment work better for some patients than for others? Traditional statistical methods that focus on correlations in observed data cannot address these questions. Drawing valid causal conclusions requires a separate framework that distinguishes correlation from causation. This book is about that framework and the computational tools that can bring it to life.

Concepts are reinforced with worked examples and code implementations in R and Python, using packages such as pgmpy, CausalModels, and causallib. As a result, the reader gains both a solid understanding of each topic and the code needed to implement it. The book is structured to move from foundations to applications. It first develops the mathematical framework of causal inference and then applies it to different case studies.

The first four chapters review probability and Bayes’ theorem, introduce Bayesian networks and their structural properties, and develop the structural causal model framework. The middle chapters focus on relational database representations of probabilistic models, join tree clustering, belief propagation, and variable elimination. Chapters 9 and 10 cover causal inference with structural models and observational data. They introduce the key tools in causal inference, including Pearl’s do-calculus, the back-door and front-door criteria, the potential outcomes framework, and the counterfactual framework.

The final chapters shift from theory to applications, covering causal inference with machine learning (meta-learners) and three case studies drawn from economics (the Lalonde study of the National Supported Work Demonstration), epidemiology (smoking cessation and mortality in the National Health Epidemiologic Follow-up Study), and social science (the Card and Krueger minimum-wage study), all implemented in Python.