Preface

Why This Book?

The landscape of empirical social science has transformed over the past three decades. The "credibility revolution" reshaped how economists and other social scientists think about causal inference. Machine learning opened new possibilities for prediction and heterogeneity analysis. The replication crisis prompted serious reflection on research practices. And yet, most graduate training still proceeds as if these developments happened in isolation—econometrics in one course, qualitative methods in another (if at all), data science somewhere else entirely.

This book attempts something different. It treats empirical methods as a unified toolkit for answering research questions, with different tools suited to different problems. A researcher studying the effects of a policy might use randomized experiments, natural experiments, or careful observational analysis—and should understand when each approach makes sense and what each can deliver. The same researcher might need to combine quantitative estimates with qualitative evidence to understand mechanisms or assess external validity. And increasingly, they'll need to understand how modern computational methods fit into the traditional empirical toolkit.

The organizing principle throughout is the research question. Before asking "What method should I use?" we need to ask "What am I trying to learn?" Different questions—descriptive, causal, predictive, mechanistic—call for different approaches. The credibility revolution's great contribution was clarifying what it takes to answer causal questions convincingly. But not all important questions are causal, and even causal questions admit multiple valid approaches depending on what data and variation are available.

Who Is This Book For?

This book is designed primarily for graduate students in economics and adjacent social sciences—political science, sociology, public policy, and related fields. It assumes familiarity with basic statistics and regression at the undergraduate level, but develops the material from foundations. Advanced undergraduates with strong quantitative preparation should find it accessible.

Working researchers may also find value here, particularly those looking to expand their methodological toolkit or update their knowledge of recent developments. The treatment of newer methods—modern difference-in-differences estimators, causal forests, double machine learning—reflects the current state of practice, while the fundamentals chapters provide a common foundation.

How to Use This Book

The book proceeds in five parts:

Part I (Foundations) establishes the conceptual and statistical groundwork. Chapter 1 addresses what empirical research can and cannot tell us—a philosophical orientation that shapes everything that follows. Chapters 2 and 3 cover data and statistical foundations. Chapter 4 introduces reproducible programming practices.

Part II (Description) covers methods for characterizing patterns in data. These chapters are often underemphasized in methods training, but description is valuable in its own right and essential groundwork for causal analysis.

Part III (Causation) is the heart of the book, covering the modern causal inference toolkit: experiments, selection on observables, instrumental variables, difference-in-differences, regression discontinuity, and more. This is the longest part, reflecting the field's sustained attention to causal questions.

Part IV (Beyond Averages) extends beyond average treatment effects to mechanisms, heterogeneity, and the integration of machine learning with causal inference.

Part V (Integration and Practice) addresses triangulation across methods, evidence synthesis, and research practice—the practical wisdom that methods courses rarely cover.

Each part includes a programming companion chapter with implementation guidance in R and Python.

The book can be read linearly, but different readers may want different paths. Someone focused on causal inference might begin with Chapters 1 and 3, then proceed directly to Part III. Someone interested in time series would prioritize Chapter 7 alongside Chapter 16. The cross-reference index at the end maps connections across chapters for those pursuing specific topics.

Conventions

Throughout the book:

• Practical Guidance sections offer concrete advice on when methods apply and common pitfalls to avoid

• Qualitative Bridge sections show how quantitative and qualitative approaches complement each other

• Integration Notes connect the current topic to other chapters

• Formal results appear in callout boxes, followed by intuitive explanations

• Five running examples recur across chapters: returns to education, minimum wage effects, microfinance and poverty, monetary policy, and China's post-1978 growth

Mathematical notation follows standard conventions documented in Appendix A. Code examples use R and Python; the programming companion chapters provide detailed implementations.

Acknowledgments

This book has benefited from countless conversations with colleagues and students over the years. The intellectual debts are too numerous to catalog fully, but the influence of the modern causal inference literature—especially the work of Angrist, Imbens, Rubin, Pearl, and their many collaborators and interlocutors—will be evident throughout.

Special thanks to NYU Wagner for providing the environment in which this book took shape.

A Note on AI Assistance

This book was drafted with the assistance of large language models, specifically Claude (Anthropic). The author used AI assistance for:

• Drafting and structuring prose based on outlines and reference materials

• Converting technical notes into accessible explanations

• Generating initial drafts of code examples

• Identifying connections across topics and suggesting cross-references

• Editing for clarity and consistency

All content has been reviewed, revised, and verified by the author. The intellectual framework, pedagogical choices, and any errors remain the author's responsibility.

I believe in transparency about AI use in academic work. These tools are becoming part of the scholarly workflow, much as word processors and literature databases became standard tools in earlier generations. What matters is that the author takes responsibility for the final product and maintains the intellectual standards that scholarly work requires.

The prompts, drafts, and revision history for this book are preserved in the project repository for those interested in how human-AI collaboration shaped the final text.

Contact

Feedback, corrections, and suggestions are welcome:

Laurence Wilse-Samson NYU Wagner School of Public Policy [e-mail]

---

January 2026