Discovering Causal Structure: Artificial Intelligence, Philosophy of Science, and Statistical Modeling

Abstract

is beautifully produced book. The cover is heavy, the typeface and graphics are clear and easy to read, and the margins are wide for the scholarly reader's notes. The title and subtitle include nine of the most powerful key words imaginable. Unlike most of those who write about causal models, the authors are real philosophers, and they have acquired Foreword by Nobel Prize winner who is seminal contributor to all of the fields they address. The first sentence of the book is This book is about computer program, TETRAD. That characterization seems bit modest, even if this particular computer program (as Herbert Simon says) be viewed as normative theory of how to induce causal models from empirical, nonexperimental data (p. xiii). In reality, however, Discovering Causal Structure is more modest than its title implies. It promises more than it delivers. There are two principal components to the book, each of which can be broken into two subparts. The first component consists of 57-page essay entitled Artificial Intelligence and Nonexperimental Science (chapters 1-3) and 12-page Brief History of Heuristic Search in Applied Statistics (chapter 9). These pages contain what Simon, in his Foreword, applauds as a careful discussion of the discovery process and detailed answers to the objections that have been raised.., .against causal modeling. The second component addresses the computer program directly. Chapters 11 and 12 (60 pages) contain the user's manual for TETRAD, and chapters 4-8 (172 pages) provide detailed descriptions of what the program does, including an introduction to causal modeling and an extensive collection of applications to real and simulated data.