When I wrote the first edition of my textbook on Formal Models of Domestic Politics, I made a conscious decision not to include models of autocracy. The literature was too new, the big picture insufficiently clear. There was enough to cover on more established topics. Nondemocracy could wait.
That first edition went to press in 2012. In the eight years since, there has been an explosion of interest in authoritarian politics. I see it at academic conferences, where panels on autocracy draw overflow audiences. Creative empirical work has cast new light on the governance of nondemocratic regimes. Simultaneously, and in dialogue with empirical research, political scientists and economists have explored the institutions of autocracy using the language of game theory. The themes of the theoretical literature are now sufficiently clear to deserve a chapter in the second edition of my textbook.
Of all the topics in the text, this is the one in which I have worked most directly. I have drawn substantially on a review article I published a few years ago with Konstantin Sonin and Milan Svolik, though I haven’t stopped there. In summarizing the literature, I have tried to be a faithful ambassador, including important models that may or may not reflect my personal views on the nature of autocracy. Nonetheless, where the literature has branched—in how to model information manipulation, for example—I have made some choices that inevitably reflect my personal modeling experience, even as I provide signposts to other parts of the literature.
With that as background, let me offer a brief survey. The chapter begins with Daron Acemoglu, Georgy Egorov, and Konstantin Sonin’s model of coalition formation in nondemocracies. Theirs is an explicitly institution-free setting: politics is governed only by the relative power of competing factions. The model thus serves as a benchmark against which more institutionally rich models can be evaluated. Anybody who has seen the brilliant film The Death of Stalin will immediately recognize the environment.
The chapter then moves to incorporate institutions into the analysis. The selectorate model of Bruce Bueno de Mesquita, James Morrow, Randolph Siverson, and Alastair Smith appears here, transplanted from the chapter on coalitions in which it previously appeared. The basic idea is to show how coalition choice and economic policy depend on the institutional environment, as measured by the relative size of the “selectorate” and the “winning coalition.” A complementary approach is to ask where institutions come from to begin with. Here, I present the simplified version of Roger Myerson’s model of institutions as a commitment device that Konstantin, Milan, and I developed for our review piece.
Myerson’s model brings information into the story; this is the focus for the remainder of the chapter. To set the stage, the chapter includes a general discussion of Bayesian persuasion, the explicit or implicit framework for many models of information manipulation in autocracies. What distinguishes this framework from the cheap-talk model of Crawford and Sobel is the assumption that the sender can commit to a probability distribution over signals for every state of the world—say, to send the signal “economy is weak” with probability 0.6 and the signal “economy is strong” with probability 0.4 if the economy is in fact weak, and to send the signal “economy is strong” with certainty if the economy is in fact strong. Is this assumption reasonable? Here is what I write:
The commitment assumption that characterizes models of Bayesian persuasion is not without controversy. It is reasonable to ask in which settings it is likely to hold. In their seminal contribution, Kamenica and Gentzkow (2011) provide the example of a prosecutor attempting to persuade a judge of a defendant’s guilt. The prosecutor might call a witness, not knowing exactly what she will say. Or she might order a DNA test, hoping for a positive result but constrained by law to share exculpatory evidence. In either case, the prosecutor de facto commits to a distribution of signals for each possible state of the world. Similarly, an autocrat might choose to disqualify candidates from an election (Ma, 2020), not knowing with certainty the distribution of voters’ preferences and thus the mapping from candidates to election outcomes. Alternatively, we can think of commitment as reflecting delegation to an actor of more or less sympathy with the sender’s point of view (think of how reporting on incumbents’ economic performance is affected by the identity of cable news personalities), where the actor is costly to replace (e.g., because charismatic hosts are in short supply). The latter environment may reflect the reality that any dictator relies on others to disseminate her message.
As this discussion illustrates, the commitment assumption is useful in two important environments in nondemocracies: government control of the media and electoral manipulation. The chapter includes some discussion of each, focusing on my model of media control with Konstantin Sonin and on work by Arturas Rozenas and by Alberto Simpser and me on electoral manipulation. This leads naturally to the final section of the chapter—a simplified version of Sergei Guriev and Daniel Treisman’s model of “informational autocracy,” which incorporates information manipulation into the models of political agency covered elsewhere.
As with other chapters, there is a lot of theory in the exercises as well as the main text. Alongside various extensions to the models discussed above, I include a model of “ex post” (i.e., after the election result is known to the incumbent) electoral fraud borrowed from a recent paper by Zhaotian Luo and Arturas Rozenas. (The full paper jointly examines ex ante and ex post manipulation.) Building on work by Gary Cox and by Andrew Little, I also consider authoritarian elections as a mechanism for gathering information about potential challenges to the incumbent regime.
Someone recently thanked me for my “service” in writing Formal Models of Domestic Politics. Oddly, I have never thought of it in those terms. Writing a textbook is one of the most enjoyable things I have done in my academic career. It satisfies various compulsions: to figure things out, to tinker with models, to write as clearly and efficiently as possible. As Gregory Mankiw has recently argued, it’s not for everyone—but it is for me. All the better that others have found the text useful. If things go as planned, the next edition will be out next summer.