"Dunning has produced a useful and remarkably accessible guide for social scientists of all sorts. I especially like his guide to discovering natural experiments." - J. D. Angrist, Department of Economics, MIT
"One of the most exciting developments in contemporary political science is the use of natural experiments to estimate causal effects. In this illuminating and highly readable book, Thad Dunning provides an expert guide to the strengths and weaknesses of this cutting-edge method, demonstrating how researchers can use natural experiments as a powerful tool for causal inference while avoiding common mistakes. I recommend this book to both beginning and experienced researchers." - Alan S. Gerber, Charles C. and Dorathea S. Dilley Professor of Political Science, Yale University
"The biggest problem social scientists face is figuring out what causes what. Does economic growth cause peace or is it the other way round? Do people adopt the values of their friends or just gravitate to others that think like them? Most of the time these questions are unanswerable but every now and then there's a chink in nature's armor. A windfall or crisis throws an economy off course, a fire or flood forces people into new social networks. Natural experimentalists seek out such moments to shine a light on underlying orders. But, as Dunning shows, the natural experimentalist's path is treacherous. In this first serious treatment of natural experiments in social science, Dunning sets down standards and shares techniques to help ensure real learning from such rare moments." - Macartan Humphreys, Professor, Columbia University
"A remarkable synthesis not just of how to do empirical work, but how to do social science. Indispensible." - James Robinson, David Florence Professor of Government, Harvard University
This book provides scholars and students with the first comprehensive guide to the use and evaluation of natural experiments - an increasingly popular methodology in the social sciences. It introduces the key issues in causal inference, including model specification, and emphasizes the importance of strong research design over complex statistical analysis.