Reflecting a sea change in how empirical research has been conducted over the past three decades, Foundations of Agnostic Statistics presents an innovative treatment of modern statistical theory for the social and health sciences. This book develops the fundamentals of what the authors call agnostic statistics, which considers what can be learned about the world without assuming that there exists a simple generative model that can be known to be true. Aronow and Miller provide the foundations for statistical inference for researchers unwilling to make assumptions beyond what they or their audience would find credible. Building from first principles, the book covers topics including estimation theory, regression, maximum likelihood, missing data, and causal inference. Using these principles, readers will be able to formally articulate their targets of inquiry, distinguish substantive assumptions from statistical assumptions, and ultimately engage in cutting-edge quantitative empirical research that contributes to human knowledge. "Aronow and Miller have raised the bar when it comes to statistical training for aspiring social and health scientists. Building on modern approaches in statistics, biostatistics, political methodology and econometrics, this introductory graduate coursebook will effectively prepare students to continue on to study many advanced topics in statistical methodology. With its honest, transparent and rigorous presentation of ideas, Foundations of Agnostic Statistics has the potential to influence generations of scholars." James M. Robins, Harvard School of Public Health, Mitchell L. and Robin LaFoley Dong Professor of Epidemiology "This book is a wonderful synthesis of theory and practice and will soon be indispensable to students of political science, statistics, and other quantitative disciplines. The credibility revolution has fundamentally changed empirical work in the social sciences in recent decades, but introductory statistical textbooks have not kept up. Social scientists and statisticians often pay lip service to Box's famous dictum that "[a]ll models are wrong but some are useful," however, it is rare for textbooks to take the dictum seriously. This book does, and it lays the foundation for credible empirical research by introducing readers to a view of statistics that does not assume that generative models are correct. It makes clear that most assumptions are made for mathematical tractability and convenience rather than verisimilitude. The authors communicate well what is most difficult: wisdom and taste." Jasjeet Sekhon, UC Berkeley, Robson Professor of Political Science and Statistics "Foundations of Agnostic Statistics is one of the few books to seriously pursue the goal of providing clarity and transparency about the assumptions underpinning empirical research; this is a critical resource for the social scientist who wants to be both rigorous and honest." Jennifer Hill, NYU Steinhardt, Professor of Applied Statistics and Data Science "I use this book in my graduate statistics course because of the beautifully integrated pedagogical approach. Starting from the first principles of probability theory, it builds up the knowledge necessary to arrive at a modern and grounded perspective on causal inference. Teaching the final chapter, Causal Inference, is a joy. Like the Karate Kid who learns karate without realizing it by practicing cleverly designed exercises, by the time students begin the last chapter they already know what to do." Joel Middleton, UC Berkeley, Assistant Professor of Political Science "The Aronow-Miller book is an imaginative new take on a first course in statistics. It consists of three parts, probability, statistics, and identification. The latter is where the book shines. It discusses concisely, but clearly, modern concepts of bounds, double robustness, and causality that are rarely covered in books for this audience. Aronow and Miller discuss these challenging topics in an accessible way, preparing the reader well for empirical work in the social sciences. I highly recommend this book for any graduate student interesting in doing empirical work using modern statistical methods." Guido W. Imbens, Stanford Graduate School of Business, Professor of Economics "Aronow and Miller have managed to write a book that both students and their professors will find interesting. Reminiscent of Goldberger's classic text, the book is thoroughly modern, self-contained and intellectually coherent. Key concepts in applied probability, statistical inference, regression analysis and causal inference are presented. Included topics are carefully selected and presented in detail. The effect is that students with varying degrees of mathematical and statistical preparation will find the book useful. I certainly plan to recommend this book to my students and will no doubt refer to it when preparing my own lectures." Bryan Graham, University of California Berkeley, Professor of