In lieu of an abstract, here is a brief excerpt of the content:

Human Biology 73.4 (2001) 621-624



[Access article in PDF]

Book Review

Causality:
Models, Reasoning and Inference


A Rooster Crow Does Not Cause the Sun To Rise: Review of Causality: Models, Reasoning and Inference, by Judea Pearl. Cambridge, UK: Cambridge University Press, 2000. 384 pp. $39.95 (hardcover).

This book is about the formal analyses of cause-effect relationships between sets of observed events and/or underlying variables related to them. Through ten chapters and an exquisitely well-written epilogue, the author, a prolific computer science specialist, essentially "demystifies" the concept of causality and explains its mathematical, statistical, as well as philosophical implications. Along with brief background material on probability theory and graph theory, the first chapter presents the basic paradigms and major problems of causal analysis and sets the tone for what follows in the subsequent chapters. The most difficult question, what constitutes evidence of a cause-effect relationship in observed data, is discussed in chapter 2, ending with the conceptualization of the validity of any such relationship observed. Chapters 3 and 4 get into deeper theoretical treatments of prediction of direct and indirect effects of actions and policies based on data in the presence of an incomplete understanding of the existence of a cause-effect relationship. Identifiability of cause-effect relationship is the central theme of these chapters. The implications of the calculus of intervention, thus developed, are discussed in the context of applications to social and health science problems in chapter 5 and 6, where the popular constructs of structural equations and confounding are presented. In contrast to the graph theory treatment of detecting the presence of confounding and of identifying critical variables that control the effect of confounding (discussed in chapter 3), chapter 6 presents the difficulties of defining and controlling confounding when statistical criteria are used. The theories of counterfactuals and structural models are presented in chapter 7, through which more rigorous definitions of the concepts introduced earlier in the book are obtained. These include concepts such as causal models, action, causal effects, causal relevance, error terms, and exogeneity. The last three chapters (8 through 10) constitute applications of counterfactual analysis. They include methods of the developing bounds of causal relationship from data of imperfect experiments using combinations of graphical and counterfactual models (chapter 8), identification and interpretation of probability of causation (chapter 9), and a formal explication of the notion of "actual cause" (chapter 10).

Paging through formal definitions of "Markovian Parents" (of an ordered set of variables, p. 14), theorems on "observational equivalence of directed acyclic graphs" (p. 19), mathematical distinctions of "back door" and "front door" adjustments for controlling confounding bias (pp. 78-83), and the like throughout the book, readers of this journal at first glance may wonder why this text is at all relevant for our discipline. However, from titles such as "Segregation analysis reveals a major gene effect controlling systolic blood pressure and BMI in an Israeli population" (Cheng et al. 1998), "Effects of intragenic variability at 3 polymorphic sites of the apolipoprotein B gene on serum lipids and lipoproteins [End Page 621] in a multiethnic Asian population" (Choong et al. 1999), and "RH blood groups and diabetic disorders: Is there an effect on glycosylated hemoglobin level?" (Gloria-Bottini et al. 2000) of recent publications in Human Biology, it is obvious that analysis of cause-effect relationships from observational data is an integral element of our research methods. As the title of this review indicates, the thesis of this book is that the inference of cause-effect relationships is not mathematically equivalent to establishing an association between events or variables, although the latter is always observed where a true causal relationship exists. Thus, even though the text is at places highly technical for general readers of this journal, this book is extremely useful for human biologists. Quantitative aptitude is a necessary requirement to understand the logic of the author, although it should not be construed as a drawback of the author's presentation. In contrast, the rigor of the treatment of the subject allows readers to understand scenarios in which association analyses...

pdf