Commentary: What Is the Marginal Value of Analytic Narratives ?

As a political institution that regulated international markets, the ICO [International Coffee Organization] was very rare indeed; few other bodies have effectively governed the behavior of states in the global arena. As with other rare occurrences, the history of the ICO can be described; it possessed a birth, a life and a death. In analyzing this case, I have sought to move beyond mere narration. I have sought to extract systematic knowledge.

—Robert Bates, “The International Coffee Organization,” chapter 5 of Analytic Narratives, 228

The case studies in this volume are motivated by a desire to account for particular events and outcomes. For those committed to the close analysis of particular events, the chapters seek to demonstrate the benefits to be gained from the systematic use of theory. For those committed to the development and use of theory, we seek to demonstrate the returns to be reaped from a close dialogue with case materials. The phrase analytic narratives captures our conviction that theory linked to data is more powerful than either data or theory alone.

Introduction to Analytic Narratives,3

The epigraphs offer a window into a bold new project in social science history. The authors of Analytic Narratives—Robert Bates, Avner Greif, Margaret Levi, Jean-Laurent Rosenthal, and Barry Weingast—intend the most thorough and methodologically self-conscious wedding of formal rational choice theory to historical narrative that has yet been attempted.

Yet the remarks quoted here are also notable for their apparent confusion about what exactly “narratives” are, at least when they are not graced by formal models. At times the authors imply that narratives are facts alone, unstructured and with little generalizable value, mere “data” that must be accompanied by models to extract “systematic knowledge.” At other times narratives are facts arranged in the crudest possible order, with superimposed stages of “life, birth and death.” At their most sophisticated, narratives in this book are sequences of facts chronologically arranged.

I think that this confusion is real—that there is genuine inconsistency across (and within) the chapters about what narratives are and that this vagueness is crucial to understanding the promise and the limitations of the Analytic Narratives project. In what follows, I will outline precisely what the authors mean by “analytic narrative,” what I think “narratives” in their nonanalytic form have to offer social scientists, and the general limitations of the book. My basic point is this: in the rendering of the authors, analytic narratives are brute sequences of events analyzed through extensive form games. This rendering of Analytic Narratives reveals a lot about the promise of this project—much of which has been met in this book, much of which has not—and reveals the idea’s limits as well. Whether formalization “costs too much” at the margin when brought to a narrative of historical change is a question that ought to be asked of any analytic narrative.

I will proceed by addressing these questions: What are analytic narratives, at least as presently conceived? What do analytic narratives miss about narratives? Where has the promise of the book been met, and where has it not? Do these general concerns affect Weingast’s analysis of antebellum U.S. politics or Bates’s interpretation of international coffee institutions?

What Are Analytic Narratives in This Book?

Analytic narrative means lots of things, but in this particular book I think it means the following: analytic narratives are interpretations of history in which chosen historical sequences are structured by, and interpreted through, extensive form games. The driving analytical force of these narratives is the joint rationality of players’ strategies as defined by subgame perfect Nash equilibria (Fudenberg and Tirole 1993: 72–74).

Does it mean anything that the games are extensive form games, not “dynamic” in the technical sense of the word, and that the equilibrium concept is that of subgame perfect Nash as opposed to Bayesian or sequential? Yes, it does, for several reasons.

First...