Explanation in Causal Inference: Methods for Mediation and Interaction

Explanation in Causal Inference: Methods for Mediation and Interaction

Explanation in Causal Inference: Methods for Mediation and Interaction

Explanation in Causal Inference: Methods for Mediation and Interaction

Synopsis

The book provides an accessible but comprehensive overview of methods for mediation and interaction. There has been considerable and rapid methodological development on mediation and moderation/interaction analysis within the causal-inference literature over the last ten years. Much of this material appears in a variety of specialized journals, and some of the papers are quite technical. There has also been considerable interest in these developments from empirical researchers in the social and biomedical sciences. However, much of the material is not currently in a format that is accessible to them. The book closes these gaps by providing an accessible, comprehensive, book-length coverage of mediation. The book begins with a comprehensive introduction to mediation analysis, including chapters on concepts for mediation, regression-based methods, sensitivity analysis, time-to-event outcomes, methods for multiple mediators, methods for time-varying mediation and longitudinal data, and relations between mediation and other concepts involving intermediates such as surrogates, principal stratification, instrumental variables, and Mendelian randomization. The second part of the book concerns interaction or "moderation," including concepts for interaction, statistical interaction, confounding and interaction, mechanistic interaction, bias analysis for interaction, interaction in genetic studies, and power and sample-size calculation for interaction. The final part of the book provides comprehensive discussion about the relationships between mediation and interaction and unites these concepts within a single framework. This final part also provides an introduction to spillover effects or social interaction, concluding with a discussion of social-network analyses. The book is written to be accessible to anyone with a basic knowledge of statistics. Comprehensive appendices provide more technical details for the interested reader. Applied empirical examples from a variety of fields are given throughout. Software implementation in SAS, Stata, SPSS, and R is provided. The book should be accessible to students and researchers who have completed a first-year graduate sequence in quantitative methods in one of the social- or biomedical-sciences disciplines. The book will only presuppose familiarity with linear and logistic regression, and could potentially be used as an advanced undergraduate book as well.

Excerpt

This book developed out of a course and set of lectures on the topic of methods for mediation and interaction that I have offered at the Harvard School of Public Health. the course was structured so as to be accessible to second-year graduate students in applied disciplines—such as epidemiology and the social and behavioral sciences—who had had a one-year introductory sequence in statistics. the lectures and the present book approach these topics from a counterfactual-based perspective on causal inference. Many of the students attending the lectures at Harvard had previously acquired an introductory knowledge of causal inference and counterfactuals in other courses at the university. in trying to bring the material in this book to a broader audience, a decision needed to be made as to whether to also presuppose, from the reader, a similar background in such introductory principles of causal inference.

After some thought, along with conversations with colleagues, I decided that, so as to attempt to extend the reach of this book as broadly as possible, I would not presuppose such a background, but that I would rather describe as many of the methods and assumptions as possible without requiring specific appeal to the notation of counterfactual-based logic. the methods presented in the book are shaped by such ideas, but the reader is not required to have had any previous exposure to them. the reader may end up acquiring a background in counterfactuals and causal inference simply by reading the book, but this would not be required to benefit from the book. There are many fine book-length introductions to counterfactuals and causal inference elsewhere (Morgan and Winship, 2007; Pearl, 2009; Hernán and Robins, 2015; Imbens and Rubin, 2015). Occasionally, when appeal to counterfactual notation is necessary (as occurs in some of the later chapters in the book), the reader is explicitly warned and can pass over these sections if desired. I believe that in almost all such cases, the reader could skip over these sections without jeopardizing comprehension of the material that appears subsequently in the book.

To the best of my knowledge, this book is unique in addressing the topics of mediation and interaction from a counterfactual-based perspective on causal inference. the book, I believe, would be unique in addressing either topic from a counterfactual-based perspective. While there have been a few book-length . . .

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.