Philip G. Wright, directed acyclic graphs, and instrumental variables
回顾了赖特1928年著作中关于因果推断的原创贡献，包括结构方程模型、工具变量估计需求弹性、有向无环图识别以及反事实福利分析，并指出其远超当时统计学和计量经济学的发展水平。
This Special Issue celebrates the seminal contributions of Philip G. Wright to causal inference in economics and puts them in a modern perspective. At its core is a reproduction of Online Appendix B in Wright’s 1928 book The Tariff on Animal and Vegetable Oils.1 This book deals with demand for and supply of oils and butter. In its Online Appendix B, Wright made several fundamental contributions to causal inference. He introduced a structural equation model of demand and supply, established the identification of demand and supply elasticities via the method of moments and directed acyclical graphs (DAGs), developed empirical methods for estimating demand elasticities using weather conditions as instruments, and proposed methods for counterfactual analysis of the welfare effects of imposing tariffs and taxes. Moreover, he took all of these methods to data. Wright’s (1928) Online Appendix B was far ahead of, and much more profound than, any contemporary theoretical and empirical developments on causal inference in statistics or econometrics. In the years leading up to its publication, economists had developed an appreciation for the identification problem in economics, with a focus on identifying demand and supply curves from price and quantity data.2 Wright himself was one early contributor to this literature, with his 1915 review of Henry L. Moore’s book Economic Cycles.3 In this book, Moore provided an extensive analysis of one important source of economic fluctuations, rainfall, and subsequently analysed the statistical relations between percentage changes in observed prices and quantities of various goods over time. Moore argued that these ‘statistical laws of demand’ described ‘average changes that society is actually undergoing’ (page 77) and were therefore more useful for prediction than Marshallian demand curves, with their dependence on abstract ceteris paribus conditions. He moreover emphasized that, for some goods, his statistical laws of demand substantially differed from typical Marshallian demand curves (page 126):