An Analytical Review of "The Book of Why: The New Science of Cause and Effect"
In their review of "The Book of Why: The New Science of Cause and Effect" by Judea Pearl and Dana Mackenzie, Peter M. Aronow and Fredrik Savje provide a critical examination of Pearl's exposition on causal inference. Pearl's work underscores the importance of causal diagrams, particularly Directed Acyclic Graphs (DAGs), in advancing the field of causal inference. While the book is positioned as a seminal text in the causal revolution, Aronow and Savje argue that the optimism surrounding Pearl’s framework may overlook essential nuances and alternative methodologies pertinent to causal inquiry.
Key Contributions of the Book
Pearl's narrative asserts that causal inquiry was largely inadequate until the rise of the causal revolution in the 1990s, propelled by the introduction of DAGs. The review acknowledges the transformative potential of Pearl's graphical representation in clarifying causal relationships, particularly among epidemiologists, philosophers, and social scientists. Pearl's framework provides a systematic approach to address confounding and has gained acceptance among those who adopt his language and methodology.
Critiques and Caveats
Aronow and Savje critique the portrayal of DAGs as a comprehensive solution to causal inference challenges. They note that the central problem in causal analysis is often the selection of a valid or reasonable causal model rather than the articulation of the models themselves. Pearl's exposition assumes known causal structures, which is frequently not the case in practice. Consequently, while DAGs aid in illustrating disagreements, they do not resolve disputes regarding the plausibility of causal models, as exemplified in historical debates such as the smoking-lung cancer hypothesis.
The reviewers highlight Pearl's division of causal reasoning into causal discovery and causal analysis, clarifying that causal discovery, the process of determining the true causal model, is inherently complex and perhaps insurmountable. Methods like sensitivity analysis, which do not hinge on the known causal structure, offer an alternative that can adjudicate causal disagreements without full consensus on underlying causal mechanisms.
Methodological Considerations
The book’s treatment of Randomized Controlled Trials (RCTs) is scrutinized for diminishing the methodological rigor that comes from their design, which inherently controls for unmeasured confounding. Pearl juxtaposes RCTs with his causal calculus without acknowledging that the robust implementation of an RCT provides empirical rigor that compensates for theoretical uncertainties inherent in other causal models.
Furthermore, Aronow and Savje advocate for a pluralistic approach, recognizing the roles of various causal models beyond DAGs. For instance, Robins' nonparametric structural equation models, the design-based approach emerging from the work of scientists like Jerzy Neyman and Ronald Fisher, and parametric models that encode functional forms contingent on domain-specific theories, each provide distinct advantages based on differing research contexts and assumptions.
Broader Implications for Causal Inference Research
While Pearl's causal framework represents a significant advancement for instances where causal structures are relatively settled, Aronow and Savje caution about its limitations in broader social science applications, where such structures are often indeterminate. They emphasize the need for complementary methodological approaches that accommodate uncertainty and promote robustness in causal inferences.
Looking ahead, the dialogue on causal inference will likely continue to embrace this methodological pluralism, integrating advances in computational techniques and empirical methodologies. This evolution will require a balanced appreciation for both the structural insights afforded by formal causal frameworks and the pragmatic flexibility offered by alternative approaches.
Overall, Pearl's "The Book of Why" remains a pivotal yet non-exhaustive contribution to the discourse on causal inference. It stimulates ongoing debate regarding the role of causal diagrams and the necessity of diverse methodologies in addressing the nuanced complexities inherent in inferring causation across fields.
