Causality: Models, Reasoning, and Inference.
本书由计算机科学家Judea Pearl撰写，系统阐述因果推断的模型与推理方法，挑战传统统计学仅能发现关联的局限，对计量经济学回归因果根源有重要启示。
How does econometrics differ from statistics? In a recent paper, James Heckman (2000) argues that econometrics, unlike statistics, is primarily concerned with causes. Heckman revives an older tradition. The Cowles Commission in the late 1940s and early 1950s – and indeed, as Hendry and Morgan's (1995) anthology demonstrates, most early econometrics – was explicitly causal. The degree of novelty of Heckman's insight is a measure of how far econometrics has drifted from its causal roots towards statistics. Although not an econometrician, in Causality Judea Pearl invites econometrics to reverse course. Pearl is a computer scientist at UCLA. His work is well known to practitioners of artificial intelligence, statistics, and non‐economic social sciences. Causality, the masterwork and capstone to his research program, recently won the Lakatos Prize in the philosophy of science. Unfortunately, Pearl's work is little known to economists and econometricians. In this book, Pearl vigorously opposes the attitude – ultimately traceable to the philosopher David Hume – that the most that we can learn from data are its associations summarised in the likelihood function and that, consequently, it is impossible systematically to infer or use causal relations. The great statistician R. A. Fisher famously argued that the available evidence could not prove that smoking causes lung cancer – a source of comfort to the tobacco companies for years. Even most statisticians reject this conclusion, but in fact Fisher was simply being honest: the standard tools of statistics are inadequate to causal analysis. It is common among both statisticians and econometricians to argue that causality is either too hard or too metaphysical a problem to be analysed.