Abstract
Steve Morgan’s edited volume on causal inference in social science research is a tremendous resource for graduate students and faculty in social and behavioral science disciplines. Given the fact that the volume contains 18 substantive chapters and that Morgan’s excellent introduction discusses the content of each chapter, I will focus primarily on the overall contribution of the volume and how different audiences might benefit from it.
The chapters are organized into six sections covering the historical context of causal analysis (section 1 and parts of section 2), specific causal analysis approaches and designs (sections 2 through 4), social contexts that pose particular challenges (section 5), and what we can learn about causal relationships from studies with suboptimal research designs (section 6). Within each section, most chapters follow a similar structure, beginning with an overview and setting the context, often historical, for the approach discussed, then describing the approach with varying levels of technical detail, and finally providing concrete examples of applications in research. This common chapter structure greatly facilitates reading the volume from cover to cover. Even if one consults this book in a much more selective and targeted manner, as one would a reference manual, the editorial imposition to contextualize, clearly review concepts, and provide examples, will be appreciated by all. Most chapters should be accessible to any graduate student taking quantitative research methods courses. A few will be challenging for those not already well versed in advanced methods and statistics. All are written by highly accomplished authors who have contributed in important ways to recent developments in causal analysis, which have been tremendous.
Beyond covering specific conceptual approaches and methodological techniques, as a whole this volume illustrates that there is no longer just one or even one dominant approach to causal analysis (though the counterfactual approach and directed acyclic graphs (DAGs) have dominated recent scholarship); rather that each among those discussed have strengths and limitations, that there are healthy tensions between proponents of different methods, and that anyone who took research methods more than 10 years ago needs a refresher course. Reading this book would be an excellent start. If nothing else, it defines terminology that is essential to understanding cutting edge social science research published in major journals.
This book also provides valuable material for instructors of graduate research courses. As someone who regularly teaches those courses, I will rely on content from the historical and conceptual chapters to set the stage for my course. Specifically, Barriger et al.’s chapter on the origins of causal analysis research, Freese and Kevern’s chapter on conceptualizations of causes (actionable, proximal versus distal, necessary or sufficient, and ultimate or fundamental), Knight and Winship’s discussion of mechanism in social science research, and Smith’s conceptual chapter and in particular his discussion of the effects of causes and causes of effects, will be a welcome complement to my required reading excerpts from Lieberson and Kish. I think Brand and Thomas’ discussion of treatment effect heterogeneity and Hong and Raudenbush’s chapter on heterogeneous outcomes (including spillover effects and other stable unit treatment value assumption (SUTVA) violations) can also help orient students toward more careful framing of research questions and methodological approaches.
For those teaching overview “toolkit” courses, other chapters include approachable introductions to: fixed, random, and mixed models approaches; heteroscedastic regression models; group differences in generalized linear models (GLMs); non-linear probability models; DAGs; instrumental variable (IV) models; structural equation models; and causal analysis using social network data. Those chapters can also serve as a resource manual for readers (and reviewers) who want a quick overview of the intuition and assumptions behind a method and its limitations. Also of great value, though more technically challenging, is the chapter on mediation analysis by Wang and Sobel, which provides a thorough discussion of why conventional approaches to mediation analysis are often limited in value, and on alternative approaches. This should be required reading for disciples of Barron and Kenny, of which there remain many in the social and behavioral sciences.
Of course, not every issue and perspective relevant to causal analysis can be covered in one book. The range of approaches discussed favors estimating causal effects at the expense of structuring causal explanations, and in particular little discussion of iterative processes between theory, generating predictions, and testing them empirically. The Smith chapter contains some morsels but a more thorough discussion or a chapter on the connections between theory construction, testing, elaboration, and casual analysis would have provided a welcome complement. There is also virtually no discussion of ethnographic approaches to causal inference. This is understandable given the need to keep this volume a manageable length. Harding and Seefelt make a claim about the importance of mixed methods but in a way that reduces qualitative research to informing quantitative research and patching its deficiencies. There is a lot more to say about the way in which ethnographers construct causal explanations.
In sum, this book, while not an exhaustive review of approaches to causal analysis, makes several very important contributions. Practically, it serves as an approachable reference manual for social science researchers trying to keep up with methodological developments. But perhaps most importantly, it provides a very welcome and overdue antidote to what has been the hegemonic econometric paradigm in causal analysis and in fact, as a set of chapters, persuasively argues against the dominance of any one approach. That could have a profound impact on empirical scholarship.