Dynamics of Conflict

Dynamics of Conflict Ronald A. Francisco Dynamics of Conflict 123 Ronald A. Francisco Department of Political Scie...

Author: Ronald A. Francisco

26 downloads 843 Views 484KB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Dynamics of Conflict

Ronald A. Francisco

Dynamics of Conflict

123

Ronald A. Francisco Department of Political Science University of Kansas 1541 Lilac Lane Lawrence, KS 66044 USA [email protected]

ISBN 978-0-387-75241-9 e-ISBN 978-0-387-75242-6 DOI 10.1007/978-0-387-75242-6 Library of Congress Control Number: 2008941246 c Springer Science+Business Media, LLC 2009 All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Printed on acid-free paper springer.com

For my favorite mathematicians, Christopher and Cynthia

Acknowledgments

This project is wholly dependent on data, so my first debt is to the institutions who provided the funds to code European and American data. The National Science Foundation (SBR-9631229), the University of Kansas General Research Fund and the Department of Political Science at the University of Kansas were all important contributors to the data collecting effort. Equally as critical to the development of our data were the students who worked on the project. The original four graduate students did the needed basic technical work as well as coding: Phil Huxtable, Astrid Obst, Uwe Reising and William Yarrow. Later graduate students coded: David Brichoux, Federico Ferrara, Steve Garrison, Taehyan Nam and Alana Querze. In addition, several (then) undergraduates coded as well: Amanda Boatright, Aimee Cox, Ian Ostrander and Erin Simpson. See the project codebook for more information: http:\\web.ku.edu/ronfran/data/index.html. Several colleagues have aided my effort in crucial ways. Paul Johnson provided invaluable help in virtually every aspect of the project, from estimation to formatting and bibliographic assistance. Erik Herron read the chapter on dictatorships and provided meaningful improvements. Philip Schrodt helped enormously in the early stages of data coding and with dynamic models. Michael Lynch in my department and Ted Juhl in economics helped to unravel knotty econometric problems. Mark Lichbach and Christian Davenport at the University of Maryland encouraged the data collection and this project and I am indebted to them. Two editors at Springer Verlag were critical to this project. Barbara Fess initiated the project and brought it to fruition while Jon Gurstelle saw it to completion. I thank both of them for their contributions. Our son, Christopher Francisco, helped me to render Chapter 1 mathematically correct. It was he also who provided the solution insights into lower and upper triangle mysteries in Jacobian matricies. I could not have done this on my own and remain in his debt. For over a decade my wife Deborah has been stalwart through time-consuming data collection efforts and this project. She is an able editor who has eased the plight of every reader, as she eases every day for me. I take responsibility for all errors contained in this volume. Lawrence, Kansas, USA

Ronald A. Francisco

vii

Contents

1 Introduction to the Problem Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 A Summary of Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.3 Model and Equilibrium Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.4 Model of Choice I: Lotka–Volterra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.5 Model of Choice II: The Competing Species Model . . . . . . . . . . . . . . . 6 1.6 Model Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.7 Dynamic Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.8 Mathematical Outcomes Arising in Estimation . . . . . . . . . . . . . . . . . . . 8 1.9 Institutional Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.10 The Dynamics of Conflict . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 2 The Dynamic Relationship Between Protest and Repression in Democratic Countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Salient Differences Among Formal Theorists . . . . . . . . . . . . . . . . . . . . 2.3 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 The Relationship Between Protest and Repression in Differing Contexts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.1 When Does Protest Generate Repression? . . . . . . . . . . . . . . . . . 2.5.2 When Do Protest and Repression Interactively Accelerate Each Other? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.3 What Happens When Repression Is Absent? . . . . . . . . . . . . . . . 2.6 Analytic Results in Democratic Countries . . . . . . . . . . . . . . . . . . . . . . . 2.7 Survey of the West European Democracies and Illinois . . . . . . . . . . . . 2.8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

13 13 14 14 14 15 15 15 16 16 27 30 31

ix

x

Contents

3 The Dynamics of Protest and Repression in Dictatorships and Democratic Transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Cases and the Context of Dictatorship . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Mobilization Under Dictatorship and Harsh Repression . . . . . . 3.2.2 What Happens to Repression When Mobilization Grows to a High Magnitude? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Empirical Results on Dictatorship Periods . . . . . . . . . . . . . . . . . . . . . . . 3.4 Empirical Results from Transition Periods . . . . . . . . . . . . . . . . . . . . . . . 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

33 33 34 34 35 36 43 48 49

4 Varied Dynamics of Bandwagon Mobilization . . . . . . . . . . . . . . . . . . . . . . 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

51 51 52 53 62 64

5 Dynamics and Stability in Civil Wars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 The Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 The Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

65 65 67 69 70 71 76 78

6 Conclusion: Stability in Conflict . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Stability is the Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Varieties of Repression in Democracies and Dictatorships . . . . . . . . . . 6.3 Convergence in Estimations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Correction of Time-Series Pathologies . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5 When Repression is Absent or Rare . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.6 What Have We Learned? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

81 81 82 84 84 84 85 85

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

Chapter 1

Introduction to the Problem Set

The effects of chance are the most accurately calculable, and therefore the least doubtful, of all the factors of an evolutionary situation R.A. Fisher

1.1 Introduction We explore the dynamics or mechanisms of conflict in this book. This is possible at last because we have data that permit dynamic analysis. In many fields, the mechanism of a physical or social process is the first factor to investigate and is usually prior to any other work. But in much of social science, the sequence has been reversed, simply because the underlying data in the field of protest and repression and many others for decades were insufficient for the tests necessary to find dynamic properties: equilibria, divergence, or oscillation. This kind of mathematical work was for the most part relegated to game theory, which needs no data. Now we have the resources to estimate dynamic models. We do not assume rationality of the individuals underlying our data, but we do consider them self-interested and risk-averse. After all, who wants to be arrested, injured, or killed in pursuit of a public good? The likelihood of protest participation is low. We know that not more than five percent of any local population ever participates in protest (Lichbach, 1995). Not many need to take part, however, in a densely populated area, to make an impact upon the neighborhood and even public policy. Nonetheless, it is important to keep in mind that we are for the most part working with fervid minorities in this book. Even bandwagon mobilization (Chapter 4) does not bring the level of mobilization to high levels. Game theorists have long been puzzled about why anyone would act for a public good (Riker and Ordeshook, 1973). Lichbach extended collective action theory so that it would function with risk and probable inefficacy (Lichbach, 1995, 1996). We are interested in the people who decide to act, or more accurately, how many choose to act. They are represented in our data. We seek principally to find out what happens when they act and the state responds to them or simply ignores them. This is the process or mechanism that we attempt to discover. Why should we consider the mechanism of conflict important? The answer to this question is that it is the basis of understanding two- or more-sided conflict and competition. We know a great deal about the correlates of conflict. We also know what R.A. Francisco, Dynamics of Conflict, DOI 10.1007/978-0-387-75242-6 1, C Springer Science+Business Media, LLC 2009

1

2

1 Introduction to the Problem Set

happens to dissent when repression grows harsh (Francisco, 2004). But prior to these concerns is the fundamental problem of the dynamic process (Morrison, 1991). Unless we understand the mechanism, we can understand nothing else about the problem. Only four things can happen in dynamic analysis: (1) stability or equilibrium; (2) divergence, that is, exponential growth or decay; (3) oscillation, a wave function that signifies instability if it is not damped; and (4) a saddlepoint, or unstable equilibrium. A saddlepoint result is shaped as a hyperbolic paraboloid and is only stable at the midpoint (0,0,0) and nowhere else. It certainly makes a difference to people in a conflict which of these mechanisms emerges. Anything but stable equilibrium brings disadvantages that might expand to severe levels. Divergence means sharp escalation or decay, something that is uncomfortable for a community. So too is oscillation, since things might be pleasant one moment and a cauldron of conflict in another. The saddlepoint, as in game theory, is stable as long as none of the instrumental players shifts position, but in the real world, position shifts are common over time with scores of players, and then the saddlepoint becomes unstable as well. All of these alternatives to equilibrium are noxious, especially at the local level and for participants (Kauffman, 1993). One of the more noteworthy discoveries in this arena was the finding by the 19th century mathematicians Laplace and Lagrange that the solar system had to have negative eigenvalues or it would fly apart into space (Tabak, 2004). Another relevant topic is that equilibrium in game theory is different from other equilibria that we uncover with empirical data (Riker, 1982, 45). Stuart Kauffman stresses the beauty and order of the natural biological world, a world infinitely more complex than is the general context of protest and repression (Kauffman, 1993).1 So complex and dynamic processes can maintain order. Equilibrium or stability is essentially what is called a “steady-state or evoluative” context, one that allows transactions without great surprises or systemic shocks (Morton, 1999, 83). Another definition of statistical equilibrium is “the condition of a macroscopic system when we observe no change over time” (Coleman, 1975, 15). Still another maintains that stability is achieved “if the system returns to equilibrium when it is pushed slightly away from equilibrium” (Roughgarden, 1998, 239). In practical terms, equilibrium best implies a return to origin, or zero, after protest, repression, or interaction is completed. In this volume we attempt to use the proper (interval-level) data for conflict and also the standard procedures of interactive time-series data estimation. Recent papers have attempted to do this, but have relied on ordinal data or artificial manufacturing of interval data from the ordinal origin, for example, Carey (2006). They have also neglected the interaction inherent in protest and repression. To model interaction of two sides, one needs at least two equations for estimation (see King, 1989). Below we introduce two models that will be the workhorses for estimation of parameters and discuss how we analytically estimate parameters. 1 Kauffman even suggests that in an ecosystem, many players can be frozen in Nash equilibrium while other players continue to adapt and evolve (Kauffman, 1993, 256). We expect that our contexts are much simpler. Certainly we do not have to worry about the carrying capacity of the environment, which matters heavily for ecologists and biologists.

1.2 A Summary of Objectives

3

1.2 A Summary of Objectives We have two primary objectives in this volume. First we seek to test interval conflict data in dynamic analysis to explore what happens when mobilization leads to protest and then some form of repression in a variety of institutional, contextual, and government-form contexts. The problem we confront is how protest and repression differ in different governmental contexts. We investigate what happens in different forms of democracy and then what happens in dictatorship and authoritarian forms of government. An added focus is what happens when a dictatorship collapses and political transition begins. Bandwagon mobilizations happen in both types of government. These events are different from normal protest and repression. They accelerate rapidly and repression may be applied at the beginning of mobilization, the end, or not at all. We consider six of these events in Chapter 4. Civil war marks our final substantive topic. Civil war is certainly a definite and incongruous means of conflict that warrants separate treatment. We estimate seven unrelated civil wars in Chapter 5. As noted later, the nature of civil war requires a different model form from the rest of the volume. The second goal of the book is to test theory. We view this enterprise as part of the empirical tests of formal models. The theory under examination is Mark Lichbach’s rational choice theory (Lichbach, 1995, 1996). Lichbach’s theory is an extension of Mancur Olson’s (1965) collective action theory. Lichbach was able to apply the theory to mobilization in conflict by using dozens of solutions grouped in four areas: market, community, contract, and hierarchy. We will also have occasion to test the theory of mobilization in conflict by DeNardo. In the 1980s, he crafted a rational choice model for conflict mobilization (DeNardo, 1985). And in the chapter on dictatorship, we will invoke the theory of Wintrobe, who constructed a dictatorship strategic theory (Wintrobe, 1998). Our problem in this context is whether the theory available conforms to empirical analytic results. If so, we confirm theory. If not, how not? Does theory need revision in terms of advancing the Collective Action Research Program (Lakatos, 1970)? So we will be mindful of theory as we proceed. Another set of assumptions we tackle is that democracies are not repressive and that dictatorships repress heavily. A separate question that has heretofore not been tested empirically is whether inconsistent repression accelerates protest (see Lichbach, 1987). We have occasion to test this game-theoretic conjecture in Chapter 2. We examine the German governments’ treatment of leftist protesters, mainly university students. This allows us to focus on one group for a long period of time to see what happens between dissidents and the state when the state represses only occasionally. Each system can be characterized by an underlying mechanism. It is that underlying form of movement that provides the foundation of a process or conflict. Scientists study stable systems in experimental settings and then perturb them to see if they remain in equilibrium. The most simple example (Luenberger, 1979, 184) is a vertical stick: secured at its bottom, it is in unstable equilibrium; secured at its top, it is in stable equilibrium, since any perturbation will move it, but it will always return to the vertical position. A system that remains stable under perturbation also remains

4


predictable and can be studied with conventional methods (see Merkin, 1997). However, a system that veers away from stability makes life more difficult for scholars, to say nothing of participants in a conflict. In order to uncover the mechanism of our conflicts, we need models. In fact we need multiple equation models, because protest and repression as well as civil war sides are all interactive and have reciprocal causative actions. The workhorse model for our context is the Lotka–Volterra (predator–prey) model, one that was developed during World War I and whose mathematical properties have been deeply explored (Murray, 1993). Originally, the model was designed to simulate the action of predators and prey in the natural world. In the decades since the introduction of the model, it has been used a great deal for human conflict, wars, or especially protest and repression (Tsebelis and Sprague 1989). Because the Lotka–Volterra model focuses on one side eliminating or repressing another side, it works well for our application. We put dissidents in the prey or protest equation and the state in the predator or repression equation. Since their relationship is dynamic and interactive, this mechanism captures a context that we can estimate with time-series data. For most of our chapters, Lotka–Volterra will be the model of choice. As we see later, civil wars require a different process model. We estimate these models with empirically coded data. These estimates provide parameter estimates that are unique for every process of protest and repression. We know what parameters theoretically move a system from stability to oscillation or divergence, but we are interested in the parameter estimates that conform to the process of conflict in every system in our sample (Morton, 1999, 137).

1.3 Model and Equilibrium Estimation We use daily aggregated interval data to estimate our models. Because the models (see later) are exactly identified, their parameters form a square, 2 × 2 matrix. Once we generate parameter estimates, we can determine stability by computing this 2×2 Jacobian matrix’s eigenvalues. There are other, more specific tests for equilibrium (e.g., Greene, 2003, 659 or Tsebelis and Sprague, 1989), but we use the general test for difference equations. Generally, in difference equations, equilibrium requires that the resulting eigenvalues be numbers wholly in the real system (i.e., not complex conjugate, and not on the complex plane) and be bounded by −1 and 1 (Elaydi, 1996, 139; Goldberg, 1986, 171–172).2 Many possibilities arise when a mechanism diverts from a soldier’s ideal view, that is, it is neither stable nor even asymptotically stable. If battles are intermittent or

Assume A is any k by k matrix. Then as n goes to infinity, the limit of An = 0 if and only if λ is bounded by −1 and 1 for all eigenvalues λ of A. Clearly, if every eigenvalue’s absolute value is less than one, fractional exponents moving to infinity will reduce to a limit of zero as the series progresses.

2

1.4 Model of Choice I: Lotka–Volterra

5

periodic, then the system of civil war may oscillate, diverge exponentially, or form an unstable saddle point. In game theory, saddle points are Nash equilibria, but in a civil war’s systemic mechanism, saddle points are necessarily unstable.3 All of the possibilities that may emerge from a Jacobian matrix are known in mathematics (see Boyce and Diprima, 1977; Elaydi, 1996; Goldberg and Potter, 1998; Merkin, 1997).4 The website http://www.math.iupui.edu/∼mtc/Chaos/ phase.htm shows graphically everything that might happen when a system is either stable or unstable. Please note, though, that this site concerns differential equations, which have somewhat different requirements for stability analysis. We have followed recommended statistical practice in ecology for estimating our data (see Roughgarden, 1998). We use seemingly unrelated regression (SUR) with models modified from differential equations to difference equations, since we cannot with certainty assume continuity. Other estimation methods largely use instrumental variables; these are problematic within our models, especially in achieving normalization (see Maddala, 1977). So SUR is a more conservative estimation method (see Maddala, 1977). These SUR estimations generally result in series convergence. We also test for and correct serial correlation and heteroscedicity.

1.4 Model of Choice I: Lotka–Volterra Our most important consideration is to select a model that, when exactly identified, allows us to estimate analytically the relationship between protest and repression (Morton, 1999). We choose the predator–prey model because it has been successful for nearly a century in estimating relationships. It is a widely used model and will play the role of the workhorse model in this volume. The Lotka–Volterra model consists to two simultaneous differential equations: dP = aP t − g(Pt × Rt ) dt dR = bR + h(Pt × Rt ) dt where P is the number of rebels, R is the repression of the state (i.e., arrests, injuries, and deaths), a is the rate at which protest declines in the absence of repression, b is the rate at which repression declines in the absence of protest, g is the rate at which the interaction of protest and coercion affects the decline of protest, and h is the rate at which the interaction of protest and repression increases repression. The dynamic mathematical properties of the Lotka–Volterra model are better known than for almost any other mathematical model. Because of this and 3 4

The requirement that no one defect or change positions is unreasonable in systemic mechanisms.

If these concepts are wholly new to a reader, then I suggest reading Acheson (1997) before the other, more technical sources.

6


because the model has been estimated with data, we understand a great deal about the dynamic possibilities it presents (May, 2001; Murray, 1993). There is a general perception about a gap between formal models and empirical estimation (Morton, 1999). The situation in our context is that the model is fully dynamic, in other words, any of the four outcomes can happen in the predator–prey model. The hypothesis inherent in this exercise is that repressive states are more likely to arrest, injure, and kill dissidents than are democratic, less repressive states. Whether this is true depends on the estimation results we generate from our data through a dynamic model. The basic reason that predator–prey presents a good model for protest and repression, however, is that it has a parameter for protest arising without repression, another for repression arising without protest, in addition to two interactive parameters, one for the change of protest and one for the shift of repression. Protest is measured as the number of people mobilized per day. Repression is a simple and nonweighed sum of arrests, injuries, and deaths of dissidents (generally arrests overwhelm the other categories). Because we cannot assume continuity even with daily aggregated data (protesters go home, after all), we transform the model to difference equations: Pt = aP t−1 − g(Pt−1 × Rt−1 ) + ε Rt = bR t−1 + h(Pt−1 × Rt−1 ) + ε where P is the number of rebels, R is the repression of the state in the form of arrest, injuries, and deaths, a is the rate at which protest declines in the absence of repression, b is the rate at which repression declines in the absence of protest, g is the rate at which the interaction of protest and coercion affects the decline of protest, h is the rate at which the interaction of protest and repression increases repression, and ε is the error term.

1.5 Model of Choice II: The Competing Species Model While the Lotka–Volterra model works best in almost all chapters, the civil war chapter is different. In a civil war, one side is not necessarily repressing the other; rather, each side tries to eliminate the other and either win the war to control the state or secede and gain territorial independence. This is a fundamentally different mechanism, and it calls for a different model: the competing species model (Murray, 1993): dR = aR t − m(Rt × St ) dt dS = bS t − n(Rt × St ) dt

1.7 Dynamic Estimation

7

where R is the number of casualties and deaths of state forces caused by rebels, S is the number of casualties and deaths of rebels caused by state forces, a is the rate at which protest declines in the absence of repression, b is the rate at which repression declines in the absence of protest, m is the rate at which the interaction of protest and coercion affects the decline of rebel forces, and n is the rate at which the interaction of protest and repression dampens state forces. Once again, we employ difference equations because we cannot assume continuous processes, even in civil war. The civil war model thus becomes Rt = aS t−1 − m(Rt−1 × St−1 ) + ε St = bR t−1 − n(Rt−1 × St−1 ) + ε where R is the number of state forces captured, injured, or killed, S is the number of rebel forces captured, injured, or killed, a is the rise in state casualties without interaction, b is the rise of rebel casualties without interaction, m is the dampening of state casualties with interaction, n is the dampening of rebel casualties with interaction, and ε is the error term.

1.6 Model Identification Both of our models have two equations and two variables on the right side of the equal sign for parameter estimation, so both are exactly identified (see Maddala, 1977). Exact identification provides a unique estimate for each parameter, which is what we seek as an analytic outcome. Under-identification may preclude estimates and over-identification may provide multiple parameter estimates, so we have chosen models that are both appropriate for our application and exactly identified. For the mathematical definition of the identification assumption, see Greene (2003, 541).

1.7 Dynamic Estimation Because we have two equation models and they are exactly identified, we have the right situation to estimate dynamics from the data. The parameter estimates are partial derivatives, and because there are four in each model, we have a square 2 × 2 Jacobian (i.e., partial derivative entries) matrix for each country tested. Because the Jacobian matrix is square, we can take its determinant and compute eigenvalues. By the Fundamental Theorem of Algebra (also the Cayley–Hamilton theorem is pertinent here), we know that we should get two eigenvalues from each model estimation, since the degree of each characteristic equation in the Jacobian matrix is two. If these eigenvalues (solutions to the characteristic equation of the matrix) are real numbers, bounded by −1 and 1, then we have stability. If the eigenvalues are real numbers outside the range of −1 and 1, we have instability. And if the eigenvalues

8


are complex conjugate numbers, then we have either oscillation or exponential growth or decline in the system estimated (see Elaydi, 1996; Goldberg, 1986). All of the time-series tests in this volume had some expected difficulty with serial correlation and heteroscedicity at least in one of the equations. We ran the Bruesch–Pagan test for heteroscedicity and the Bruesch–Godfrey and Durbin h tests for serial correlation as detection measures. As a consequence, all of the parameter estimates in the chapter tables are corrected for one or both of these problems and the eigenvalues are computed from the Jacobian matrix formed by the corrected parameter estimates. Heteroscedicity is corrected with generalized autoregressive conditionally heteroscedicity, while serial correlation was overcome by a series of standard econometric techniques.

1.8 Mathematical Outcomes Arising in Estimation Throughout, we assume still for simplicity that we are working with two equations. There are several situations we may confront in estimating dynamic models with empirical data. Two of the most important of these are on follows: (1) when the Jacobian matrix (i.e., a matrix of parameter estimates) is upper or lower triangular and (2) the case of instability and oscillation. In the first problem, if the parameter estimates not on the main diagonal are zero (or their product is very close to zero), and consequently their t-values are near zero, then a result from linear algebra simplifies life considerably. When the parameter estimates are exactly zero, the eigenvalues of the matrix are the same as the parameter estimates on the main diagonal of the Jacobian matrix; if the product of the parameter estimates is very near zero, then the eigenvalues will be very close to the diagonal entries.5 When we confront an unstable, diverging, or oscillating case, we want to use mathematics to understand the behavior of the solutions. This can be rather complicated, and one needs to work with both real and complex numbers. When we have oscillation, the eigenvalues are complex numbers that are not real, and they occur in complex conjugates; that is, if α + iβ is one eigenvalue, the other is α − iβ, where α and β are real. As Elaydi (1996, 133) points out, the oscillation can be represented with sine and cosine functions. Suppose we have complex conjugate eigenvalues α ± iβ. Let r = α 2 + β 2 ; this is the modulus of the eigenvalues and represents the length of the vector defined by α + iβ or α − iβ in the plane. Denote the first and second eigenvalues by ξ1 and ξ2 . Then, the solution x(n) to the system is

5

The proof of this is straightforward: Assume a matrix A that has the form We find the determinant of λ × I − A. This new matrix is

λ − a −b

a b 0 c

.

. 0 λ−c Take the determinant; it is (λ − a) × (λ − c) − 0 × b = (λ − a) × (λ − c). We set this equal to 0, and the eigenvalues are a and c.

1.9 Institutional Theory

9

x(n) = r n [(cos n⌰)ξ1 − (sin n⌰)ξ2 ] + ir n [(cos n⌰)ξ2 + (sin n⌰)ξ1 ]

Here is an example drawn from parameter estimates in a case outside of our sample. We have complex conjugate eigenvalues and therefore an unstable system:

x(n) = (0.0000000172816 + 0.00000003733552i)4,381 + (0.000000526676 0.001354438i 0+ 0.00000000005762306i − + 0+ 0.00000001113676i 0.02114537

It is easy to see that this system oscillates. Also, the exponent (n) of 4,381 is sufficiently large that the modulus of the complex number will be very close to zero. Note that the eigenvalues and eigenvectors are all well below one. Therefore, while this system oscillates, it dampens toward stability. The limit of the system is zero, or the origin; therefore, the system would eventually move to stability.

1.9 Institutional Theory While our interest in conflict is paramount, we must have a broader view of protest and repression, civil war or even mobilization. Some countries have a greater amount of conflict than do others. This indicates that the form of government might play a role in the level of street conflict. So we will consider forms of government beyond our delineation in Chapters 2 and 3 on democracies and dictatorships, respectively. Two kinds of theory dominate in our context. First, the number of veto players in a government logically matters for groups (see Tsebelis, 2002; Doering, 1995). If veto players are maximized then groups usually will have access to at least one type of veto player. Mobilization on the street should then be limited. In contrast, where there is effectively a single veto player (e.g., United Kingdom and France), access to politics is minimal and we would expect a larger amount of mobilization at the street level. Another theory that makes sense for us is federalism (see King, 1982). There are differing degrees of federalism. Austria and Germany are both federal countries, but in each local power is dwarfed by the power United States’ states, counties, and cities. Switzerland, in contrast, is a country that has greater local than central political power. More local autonomy should increase the standard deviation of conflict in a country. Some local governments might be benign in the face of opposition while others might actively engage opponents (see degree of centralization in King, 1982, 122–123). We will test this with our four federal countries: Austria, Germany, Switzerland, and the United States as represented by the state of Illinois (see Chapter 2).

10


1.10 The Dynamics of Conflict The problem set of this volume we have attempted to portray consists of (1) choosing appropriate models to estimate for the different contexts in our data; (2) estimating mechanisms of our data with appropriate models; (3) testing theory (especially Lichbach’s collective action theory) along the way as we generate results; and (4) dealing with statistical and time-series econometric as well as linear algebra problems. This chapter introduces all of these concepts. We will not revisit most of them specifically in the rest of the volume, since we will concern ourselves there with the actual outcomes of data estimation.

Bibliography Acheson, David. 1997. From Calculus to Chaos. New York: Oxford University Press. Carey, Sabine C. 2006. “The Dynamic Relationship Between Protest and Repression.” Political Research Quarterly 59(1):1–11. Coleman, Stephen. 1975. Measurement and Analysis of Political Systems: A Science of Social Behavior. New York: John Wiley and Sons. DeNardo, James. 1985. Power In Numbers: The Political Strategy of Protest and Rebellion. Princeton, NJ: Princeton University Press. Doering, Herbert. 1995. Parliaments and Majority Rule in Western Europe. New York: St. Martin’s Press. Elaydi, Saber N. 1996. An Introduction to Difference Equations. New York: Springer Verlag. Francisco, Ronald A. 2004. “After the Massacre: Mobilization in the Wake of Harsh Repression.” Mobilization 9(2):107–126. Goldberg, Jack and Merle C. Potter. 1998. Differential Equations: A Systems Approach. Upper Saddle River, NJ: Prentice-Hall. Goldberg, Samuel. 1986. Introduction to Difference Equations. New York: Dover Publications. Greene, William H. 2003. Econometric Analysis. Upper Saddle River, NJ: Prentice Hall. Kauffman, Stuart A. 1993. The Origins of Order. New York: Oxford University Press. King, Gary. 1989. Unifying Political Methodology. New York: Cambridge University Press. King, Preston. 1982. Federalism and Federation. Baltimore, MD: Johns Hopkins University Press. Lakatos, Imre. 1970. “Criticism and the Growth of Knowledge.” Chap. Falsification and the Methodology of Scientific Research Programmmes, New York: Cambridge University Press. Lichbach, Mark I. 1987. “Deterrence or Escalation? The Puzzle of Aggregate Studies of Repression and Dissent.” Journal of Conflict Resolution 31:266–297. Lichbach, Mark I. 1995. The Rebel’s Dilemma. Ann Arbor: University of Michigan Press. Lichbach, Mark I. 1996. The Cooperator’s Dilemma. Ann Arbor: University of Michigan Press. Luenberger, David G. 1979. Introduction to Dynamic Systems. New York: John Wiley & Sons. Maddala, G.S. 1977. Econometrics. New York: McGraw-Hill. May, Robert M. 2001. Stability and Complexity in Model Ecosystems. Princeton, NJ: Princeton University Press. Merkin, David R. 1997. Introduction to the Theory of Stability. New York: Springer Verlag. Morrison, Foster. 1991. The Art of Modeling Dynamic Systems. New York: Wiley Interscience. Morton, Rebecca B. 1999. Methods and Models: A Guide to the Empirical Analysis of Formal Models in Political Science. New York: Cambridge University Press. Murray, J.D. 1993. Mathematical Biology. New York: Springer Verlag. Olson, Mancur. 1965. The Logic of Collective Action. Cambridge, MA: Harvard University Press. Riker, William H. 1982. Political Equilibrium, Chap. A Reply to Ordeshook and Rae, pages 41–46. Boston: Kluwer Nijhoff.

Bibliography

11

Riker, Wiliam H. and Peter C. Ordeshook. 1973. An Introduction to Positive Political Theory. Engelwood Cliffs, NJ: Prentice-Hall. Roughgarden, Jonathan. 1998. Primer of Ecological Theory. Upper Saddle River, NJ: Prentice Halll. Tabak, John. 2004. Algebra: Sets, Symbols and the Language of Thought. New York: Facts on File. Tsebelis, George. 2002. Veto Players: How Political Institutions Work. Princeton: Princeton University Press. Tsebelis, George and John Sprague. 1989. “Coercion and Revolution: Variations on a PredatorPrey Model.” Mathematical and Computer Modelling 12:547–559. William E. Boyce and Richard C. DiPrima. 1977. Elementary Differential Equations. New York: John Wiley & Sons. Wintrobe, Ronald. 1998. The Political Economy of Dictatorship. New York: Cambridge University Press.

Chapter 2

The Dynamic Relationship Between Protest and Repression in Democratic Countries

A bill of rights, as Jefferson remarked, was “what the people are entitled to against every government on earth, general or particular, and what no just government should refuse, or rest on inference.” Hannah Arendt, On Revolution

2.1 Introduction This chapter explores the relationship between protest and its repression in its most feasible contexts: democratic regimes. I catalog both past knowledge and recent empirical tests of formal theoretical conjectures using data on European democratic countries as well as the state of Illinois in the United States. The purpose of this journey is to give firm basis to our present knowledge in order to determine, first, what is known and second, what remains uncertain about the dynamics of protest and repression. We also have a unique perspective of the street to evaluate human rights and democracy. All the countries in this chapter are democratic. But how democratic? Do citizens retain rights, even when they exercise their rights to speech and assembly? We do not focus primarily on institutions, but at the level of the street. This task is not as daunting as it may seem. Most of the previous work in the 1960s through the 1980s used yearly or perhaps quarterly or monthly ordinal data rather than today’s interval daily data. Moreover, earlier work tested hypotheses with linear or multiple regression rather than estimating multiple equation models for interaction (King, 1989). If anything in politics is interactive, it is protest of dissidents and repression by the state. We have to cover then a small amount of more valid testing as well as new testing of theoretical conjectures by formal theorists, especially (Lichbach, 1987, 1995, 1996; DeNardo, 1985). Our first priority is to settle the wrangling among our formal theorists. Once we have them at peace, we state our assumptions and then consider the most important interactive questions. These require past and present tests of different contexts and countries; in our case all are in Europe, except for the state of Illinois in the United States. Once answered or found unanswerable, we shift to varied contexts, essentially, “what happens to y when x occurs?” All of this is designed to portray

R.A. Francisco, Dynamics of Conflict, DOI 10.1007/978-0-387-75242-6 2, C Springer Science+Business Media, LLC 2009

13

14

2 Dynamic Relationship Between Protest and Repression

the complex relationship between these two highly interactive concepts—or at least inform us of the research that must be done to discover the necessary knowledge.

2.2 Salient Differences Among Formal Theorists The greatest conflict among the formal theorists: DeNardo’s (1985) assumption that nearly anyone might be mobilized in a given policy space versus Lichbach’s (1995 1996) argument that at most five percent of any population can be mobilized. Empirical evidence in this instance supports Lichbach. Mobilization for dissent, even without repression, rarely achieves even five percent—at least until bandwagons develop, and even they rarely reach five percent of the population. DeNardo limits his incentives to ideology or state spatial policy, whereas Lichbach considers a full range of selective incentives and solutions to the collective action problem. Lichbach (1987) proved with game theory that inconsistent repression accelerates protest. These are abstract arguments that we will test as closely as possible to discover what really happens in a protest and repression context.

2.3 Assumptions Our assumptions are simple and few. First, we assume that our daily interval data are valid and reliable. Data come from 500 sources and are coded daily and sub-daily, that is, when more than one event occurs on a single day (http://web.ku.edu/ronfran/ data/index.html). Second, we assume that the countries we test are representative at least of most types of democratic countries. For presidential systems we shift to the United States and examine 21 years of Illinois data. Finally, we have 16 years of daily data (except for Germany’s special case). We presume that these years, 1980 through 1995, represent a typical time of context. For Eastern Europe, this period covers regime transitions from communist dictatorships toward democratization. Hence, all the former communist states appear in Chapter 3 with interrupted time-series tests, before and after regime transition. In terms of expectations of our predator–prey model estimation, truly democratic countries should not have much conflictive interaction on the street. So, we expect that in most cases, the parameter a would be statistically significant, and that in some cases the repression (without interaction) parameter b might be statistically significant, but that neither of the interactive parameters would achieve statistical significance in most cases (parameters g and h).

2.4 Cases The vast majority of our cases come from our European NSF coding fund (see the URL earlier). They comprise Austria, Belgium, Denmark, France, Germany, Greece, Iceland, Ireland, Italy, Luxembourg, Netherlands, Northern Ireland (as a

2.5 The Relationship Between Protest and Repression in Differing Contexts

15

region), Norway, Portugal, Spain, Sweden, Switzerland, and the United Kingdom. We also test 21 years of coded data from the United States, in this case, the state of Illinois, from 1985 to 2005. Illinois allows us to test a federal, presidential form of government that does not exist in Europe.

2.5 The Relationship Between Protest and Repression in Differing Contexts We begin our investigation with what our formal theorists predict would happen under the most feasible conditions; we also explore empirical questions of the relationship between protest and repression.

2.5.1 When Does Protest Generate Repression? Protest that generates or accelerates repression reduces the government’s capacity to enforce its laws and repress dissent. Even democratic countries require demonstration and rally permits. Riots, ironically, are least repressed in North America, mainly because standard police procedure is for police to flee the scene. In Europe, however, riots emerge from the dregs of demonstrations and rallies. They are almost always violent and therefore generate squads of specially trained riot police. Repressive countries are more concerned with threats of competing democratic organizations. Any such movement is more repressed than a few standard dissidents occupying a square or a building.

2.5.2 When Do Protest and Repression Interactively Accelerate Each Other? The first thing to recognize in the interactive situation is that both dissidents and the state police think and adapt. Oliver and Myers (2003) show that social movements and states co-evolve. With both sides adapting, repression should reduce deaths and injuries over time, while protest becomes more conventionally difficult. Sometimes, though, states and dissidents remain fixed in their tactics. In such a context, the dissidents and police are in a dependent, Lotka–Volterra (predator-prey) interaction, with protest “dependent” on repression and vice-versa. In general, we expect that thinking and adaptation would minimize interaction in democratic countries. We would expect that protest rises in the absence of repression, but that little else occurs. There is a famous theorem, however, proved with game theory, that inconsistent repression accelerates protest Lichbach (1987). We will test this theoretical conjecture empirically in this chapter. The test requires precisely the sort of data we have: interval-level protest and repression data with protesters identified. This allows us to isolate a single group, warrant that repression is inconsistent, and then look at protest levels.

16


2.5.3 What Happens When Repression Is Absent? If everyone is satisfied, little mobilization should take place. In the absence of repression, grievances always arise, if only because dissident entrepreneurs seek to mobilize in their own interest (Lichbach, 1995, 1996). In reality, a lack of repression and general citizen satisfaction should translate to difficulty in mobilizing. We have not had a situation to test this, but nearly total satisfaction and brutal repression should be the two contexts in which mobilization is challenging. We have seen consistent violent repression. Can we find a context in which trivial repression occurs?

2.6 Analytic Results in Democratic Countries We now move to the test of the Lotka–Volterra model for all our cases. We present them alphabetically in Europe and then add the Illinois case from the United States at the end of this section. Each table provides parameter estimates, t-values and the probability of t, as well as eigenvalues as a stability measure and the size (N ) of the data sample. Austria is our first case and gives us an interesting basis (Table 2.1). Not only are the protest and repression without interaction statistically significant, but the interaction repression parameter, h, is significant as well. We need to keep in mind that while Austria is a parliamentary democracy, it is a federal system with a good deal Table 2.1 Austria Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 5, 464 ∗

Estimate ∗

0.064873 0.085409∗ 0.00069 −0.0000525∗ λ1 = 0.0657683427

t-value

p(t)

4.8 5.54 0.1 2.65 λ2 = −0.0009478427

0.0001 0.0001 0.9189 0.0081

indicates statistical significance Table 2.2 Belgium Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 )

Parameter

Estimate

t-value

p(t)

a b g h Eigenvalues N = 5, 661

0.3065∗ 0.0759∗ 0.0151∗ −0.00000004837 λ1 = 0.302713951

21.76 5.67 8.98 0.57 λ2 = 0.003786001

0.0001 0.0001 0.0001 0.5708

∗

indicates statistical significance

2.6 Analytic Results in Democratic Countries

17

Table 2.3 Denmark Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 5, 884 ∗

Estimate ∗

0.314 0.00334 0.0137∗ −0.0000000191 λ1 = 0.016220921

t-value

p(t)

24.32 0.23 7.0 0.07 λ2 = 0.002820941

0.0001 0.8165 0.0001 0.9416


of local autonomy. However, its sign is negative, indicating that interaction reduces rather than increases repression. Belgium (Table 2.2) gives us a similar impression, but in this case it indicates that protest increases with interaction. This may well be a result of ethnic conflict in Brussels (Flemish versus Walloons), rather than a basic difficulty with police. In fact, it is likely to represent cases when police attempt to break up Flemish-Walloon fights. Denmark (Table 2.3) presents results that are close to its Scandinavian neighbors. Both of the parameters in the protest equation are statistically significant, indicating both lively generation of protest and some interaction between police and dissidents. The sign of the interactive parameter (g) is negative (see model). This means that in Denmark, interaction between state agents and protesters dampens protest. There is absolutely no indication of repression in Denmark. Both of the repression equation parameters are near zero and have exceptionally small t-values. Moreover, the sign on the interactive (h) parameter is negative, meaning that interaction with police tends to reduce, not increase, repression. For France, Table 2.4 shows us a remarkable situation where all parameter estimates of the Lotka–Volterra model are statistically significant. T -values are high, especially for the repression parameters b and h, and yet the interaction of dissident and police is stable. What we see here might be a response to a form of government. In Western Europe, France is unique in having a unitary semi-presidential form of government, a totally new form of democracy invented by Charles DeGaulle and Michel Debre in 1958. This form of government has both a president and premier, Table 2.4 France Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.0161∗ 0.0738∗ 0.000234∗ 0.000002217∗ λ1 = 0.107703078

8.06 49.99 2.42 31.91 λ2 = −0.001625248

0.0001 0.0001 0.0156 0.0001

∗


18


the latter functionally a prime minister. Premiers serve at the president’s pleasure, and the legislature in these kinds of government is weaker than in any other form of democratic government, since they are dominated by the premier. The choice of meaningful actions for a dissident is bleak: lobbying a legislator is not helpful; the President and Premier are inaccessible, so the street is the most logical choice. Clearly, many choose the street and find police present. We return to the French case below after considering the rest of the established West European democracies. The Federal Republic of Germany is a long-standing democratic country; it features a constitution with human rights as well as democratic elections and federalism that allows a degree of local control. Again, we test the Lokta–Volterra model with daily aggregated data on protest and related repression. The German situation differs from most of our cases, because West and East Germany merged in 1990. In order to estimate purely democratic data, we limit the data to 10 years, 1980–1989. Table 2.5 shows the parameter estimates and tstatistics for coefficients of the predator–prey model. Germany appears to be truly democratic on the street in the relationship it fosters between protest and repression. The only statistically significant parameters are a, showing the positive rise of protest in the absence of repression and b, the positive increase in repression without protest. Protesters in Germany feel no compunction about moving out into the streets. Repressive behavior on the part of police forces escalates in the absence of protest, but at far higher levels than in the Netherlands. Neither of the interactive parameter estimates is statistically significant. The German profile is another type we expect for a democratic government: People move out on the streets with grievances, but generally are not punished for doing so. The repression appears to be inconsistent Lichbach (1987), but probably depends on the contexts of mobilization. The possibility of inconsistency allows us to test empirically an important theoretical conjecture proved by game theory, but not yet estimated with actual data. The group that allows us to find out whether inconsistent repression increases protest in Germany is leftists. Leftists protested during every year of the decade we have. In the early 1980s, they were repressed regularly. Thereafter, they were repressed inconsistently with the same tactics, for example, demonstrations. Leftist protest was one of the strongest in all of the German data. Ecological protesters dominated all other groups, but this was three years of occupying an airport site to Table 2.5 (West) Germany Lotka–Volterra results, 1980–1989 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.2077∗ 0.2957∗ −0.000411 0.000000020564 λ1 = 0.2071132064

12.36 17.16 1.95 0.18 λ2 = 0.0005868142

0.0001 0.0001 0.0507 0.8543

∗



19

prevent a new runway. There were 274 separate leftist protests. After 1980, about two-thirds of the time leftist events drew no repression. But in one-third they did, even using the same tactics in the same place. The best way to see this is graphically. Figure 2.1 displays German leftist protest over time (1980–1989), while Figure 2.2 50000

40000

30000

20000

10000

0 14 28 42 56 70 84 98 112 126 140 154 168 182 196 210 224 238 252 266

Fig. 2.1 German leftist protest, 1980–1989 300

250

200

150

100

50

0 14 28 42 56 70 84 98 112 126 140 154 168 182 196 210 224 238

Fig. 2.2 German leftist repression, 1980–1989

266

20


depicts repression (arrests, injuries, and deaths) over the same period. It is plain to see that protest rose substantially over this period of inconsistent repression. It is also clear that repression was inconsistent and certainly did not match the protest graph. The zero-order Pearson correlation between the leftists and their repression by state agents is −0.05656. If one lags protest one day, then the same correlation shifts to 0.033. But as one can see from Figs. 2.1 and 2.2, there is an increase in protest and only episodic repression. The two time series do not correlate, but Lichbach’s theorem holds in graphic reality. Figure 2.3 represents another view of the same data. It shows the ratio of the number of leftist protesters to the number arrest, injured, or killed. It is clear that repression was episodic and inconsistent on the same group using most the same tactics all the time. Let us consider Greece, a country on the geographic margin of Europe. Ruled by a military dictatorship during the mid 1970s, Greece thereafter became a democratic parliamentary government with little repression. We have 16 years of daily data on protest and repression starting in 1980, five years after Greek democracy was restored. Table 2.6 shows the results of the predator–prey mechanism operating in Greece. The parameter estimates show that both a and b are highly statistically significant, whereas the interactive parameters g and h are not. Greek protest arises freely. Repression arises on an anniversary of student clashes against the military dictatorship (when there are riots in Athens), but not on most other days. Greece has more repression than the average European country, but the estimation results are similar to Belgium’s (Table 2.2). The most significant difference is that the interaction term represented by g is statistically significant, indicating much more interaction between dissidents and police. Much of this interaction 1400 1200 1000 800 600 400 200 0 14 28 42 56 70 84 98 112 126 140 154 168 182 196 210 224 238 252 266

Fig. 2.3 The ratio of leftist German protest to repression, 1980–1995


21

Table 2.6 Greece Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 5, 884 ∗

Estimate ∗

0.11459 0.08397∗ 0.03767∗ 0.0000004723 λ1 = 0.06823068

t-value

p(t)

8.7 5.99 9.11 1.61 λ2 = 0.04635885

0.0001 0.0001 0.0001 0.1074


encompasses annual fall clashes in Athens between anarchist students and riot police. Like many protest events, this is an anniversary of repression in the early 1970s military junta. The Greek protest interaction parameter is negative (see model), so that even here interaction dampens action. Iceland is a small Scandinavian country with a small population. Once again this means that we would not expect much interaction and that there would be little statistical significance. This is what Table 2.7 tells us, but it also presents us with our first linear algebra lower triangle situation. Because the interactive components are near zero, the eigenvalues in this special form of a Jacobian matrix are the same as the main diagonal parameters. Iceland’s “repression” consists only of arrests; in 16 years there were neither injuries nor deaths of dissidents. Our Ireland data does not reflect conflict in Northern Ireland, which is part of the United Kingdom. Rather, it comprises mostly labor disputes and some social problems, especially abortion after rape and divorce in a country with a Catholic religion. The Irish results demonstrate what we would expect to find in a democracy. Parameter a alone in Table 2.8 is statistically significant, with all other t-values near the zero mark. Italy’s data start after most of the 1970s Red Brigade terror ended. These data too comprise a great many labor disputes, some social and regional problems, but no large-scale or overriding issues, save the diffuse response to government corruption. In Italy’s case, only parameter b in Table 2.9 is statistically significant. Repression rises in the absence of protest, but neither of the interactive parameters Table 2.7 Iceland Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.337∗ 0.0000000000000000031 0.0564 0.000000000000000000542 λ1 = 0.337

26.49 0 0 1.61 λ2 = 0.000000000000000000031869

0.0001 1.0 0.9992 1.0

∗


22

2 Dynamic Relationship Between Protest and Repression Table 2.8 Ireland Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 5, 884 ∗

Estimate ∗

0.03397 0.0025 0.00375 −0.000000133 λ1 = 0.0342140095

t-value

p(t)

2.47 0.15 0.14 0.05 λ2 = −0.0002741425

0.0137 0.8825 0.8901 0.9633

indicates statistical significance Table 2.9 Italian Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 )

Parameter

Estimate

t-value

p(t)


0.00844 0.10585∗ 0.000107 −0.000000021 λ1 = 0.009617622

0.61 7.61 0.08 1.35 λ2 = −0.001177643

0.5424 0.0001 0.9354 0.1772

∗


approaches statistical significance, implying that Italian police stand ready, but do little repression in general. Luxembourg has almost no repression or police action. In 16 years, there were only three incidents of arrest or injury. From 1980 until the early 1990s, only two people had been arrested by police for political or labor actions. A total of 42 arrests and 2 injuries in more than one and one-half decades marks Luxembourg as a peaceful democracy on the street. Because there is so little repression, the parameter estimates for the second (repression) equation are zero in Table 2.10. We thus have a second case of the special situation of a linear algebra lower triangle matrix, which mean the eigenvalues match the parameter estimates on the main diagonal of the Jacobian matrix. Table 2.10 Luxembourg Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 5, 493 ∗

Estimate ∗

0.2135 0 −0.0671 0 λ1 = 0.2135


t-value

p(t)

16.16 0 0.99 0 λ2 = 0.0

0.0001 1.0000 0.3223 1.0000


23

Table 2.11 Lotka–Volterra results, Netherlands, 1980–1995, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


−0.0218 −0.01534 0.01339 −0.000000692 λ1 = 0.028940685

1.57 0.84 0.36 0.16 λ2 = 0.00714377

0.1169 0.3988 0.7155 0.8757

The Netherlands has a high probability of being a country with little or no repression. While riots occur even in the Netherlands, they are infrequent. Most of the strife is typical labor relations, that is, strikes, occupations, and lockouts, little of which is ever violent. Table 2.11 shows the results of the predator–prey model for the Netherlands. The Netherlands clearly represents a different case. Protest and repression do not arise easily or significantly (a and b), and there is no significant interaction between dissidents and the state. The Netherlands’s results are a model what we would expect to find in a democratic, nonviolent country with little or no repression. The largest t-value in Table 2.11 is just 1.57 and it is where we would expect it, the parameter (a) that represents protest rising in the absence of repression. Norway is quite similar to the Netherlands as well as very much like its Scandinavian neighbor Iceland. As Table 2.12 shows, only the parameter that represents protest rising in the absence of repression is statistically significant. The t-values on the other parameters are quite low, in particular given such a large sample size. This indicates that police are inactive in Norway and that there is little interaction between police and dissidents. The Portuguese results show more interaction between state agents and dissidents. Both the parameters in the protest equation are statistically significant (a and g) in Table 2.13, while in the repression equation both the parameter estimates and their respective t-scores are also zero or near zero. Therefore, beginning in 1980, with the center-coalition of Francisco Balsemao, repression in Portugal almost Table 2.12 Norway Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.2384∗ 0.01095 0.001438 −0.00000851 λ1 = 0.238466

18.09 0.58 0.56 0.40 λ2 = −0.00006688155

0.0001 0.5602 0.5727 0.6870

∗


24

2 Dynamic Relationship Between Protest and Repression Table 2.13 Portugal Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

∗

a 0.2094 13.39 b 0.000122 0.01 g 0.0414∗ 6.88 h −0.00000000000148 0.0 Eigenvalues λ1 = 0.2094241 λ2 = −0.00002411757 N = 5, 406 ∗

p(t) 0.0001 0.9929 0.0001 0.9996


vanished. This result stems from the constitutional reform in 1982 in Lisbon that severely reduced the power of the president. Spain’s results are similar to Portugal’s, and for similar reasons. After the death of dictator Franco in 1975, Spain entered a transition phase with Juan Carlos as king. In 1978, several opposing parties came together to write a new, democratic constitution. The most different aspect to the Spanish data is the presence of Basquebased terror by ETA. Interaction with terrorists and suspected terrorists accounts for a great deal of the repression. Nonetheless, it is only the parameters in the protest equation that are statistically significant wherein the interactive parameter shows dampening of protest and repression in Table 2.14. Sweden’s results in Table 2.15 almost conform to the stereotype of a Scandinavian country with human rights and social democracy. The only positive figure of note in the Swedish results is a statistically significant parameter (a), protest rising in the absence of repression. Thus, a good deal of dissent emerges, but there is almost no interaction with state agents and little evidence of repression. Switzerland presents us with a unique form of government, one anchored at the local level and governed largely by referendum voting. It is interesting that the only significant parameter estimate in Switzerland’s Table 2.16 represents repression rising in the absence of protest (b). This is a hallmark, it appears, of countries with high degrees of local control, mostly our federalist states Austria, Germany, and Illinois in the United States as well as Switzerland with a kind of hyper-federalism

Table 2.14 Spain Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.3192∗ 0.00003 0.003027∗ −0.000000005354 λ1 = 0.3192003

17.15 0.0 11.07 0.01 λ2 = −0.000002791382

0.0001 0.9984 0.0001 0.9928

∗



25

Table 2.15 Sweden Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 5, 439 ∗

Estimate ∗

0.5252 0.012326 0.002979 0.000003153 λ1 = 0.5252699

t-value

p(t)

45.51 0.62 0.04 0.19 λ2 = −0.00006675273

0.0001 0.5325 0.9669 0.8498

indicates statistical significance Table 2.16 Switzerland Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 )

Parameter

Estimate

t-value

p(t)


0.003028 0.1279∗ 0.00012 −0.000004638 λ1 = 0.005712541

0.22 7.31 0.05 0.65 λ2 = −0.0002689179

0.8248 0.0001 0.9613 0.5172

∗


(see King, 1982). When local control is autonomous, one might expect that at least in some locales, police would be more active than the norm. This is true of a sufficiently large number of cities in Switzerland, but there is no interaction evident between state agents and dissidents. The Swiss data contain a good deal of youth protest, some of it violent, in the early 1980s, but even this did not affect the results in the 16-year time series. The United Kingdom has a more repressive profile than most other countries, but its repression is even less likely to involve injuries and deaths (93.17% of its repression is arrests). The UK has a higher death mean than other countries, but that is almost wholly attributable to urban race riots and PIRA versus UFF and UVF terror.1 The information we seek lies in the parameter estimates. Note the signs on the coefficients in Table 2.17. In the United Kingdom, a raises protest without repression. UK constables accelerate their repression (b) in the absence of protest. We see differences in Margaret Thatcher’s and John Major’s UK in the effect of interaction of protest and repression. Interaction accelerates protest and repression. In the United Kingdom, g increases protest and h increases repression when there is interaction between dissidents and police. The period covered by the UK data includes the violence of a year-long coal strike, a newspaper printer strike and many summer 1

PIRA is the Provisional Irish Republican Army; UFF is the Ulster Freedom Fighters and UVF is the Ulster Volunteer Force; the latter two organizations are Protestant.

26

2 Dynamic Relationship Between Protest and Repression Table 2.17 United Kingdom Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 5, 301 ∗

Estimate ∗

0.2471 0.0769∗ −0.00494∗ 0.00000063575∗ λ1 = 0.248628

t-value

p(t)

18.54 4.66 4.67 2.23 λ2 = −0.0015273

0.0001 0.0001 0.0001 0.026


youth and race riots in London, Liverpool, Manchester and other cities. Also important is that Northern Ireland is part of the United Kingdom. The United Kingdom has no written constitution and therefore no true foundation of human rights. Perhaps we should expect that the interactive parameters might be statistically significant in the United Kingdom. We take up the cases of France and UK below. In the mean time, though, we move eastward to the United States, where we can test the state of Illinois from 1985 to 2005, 21 years of daily data. It is clear from Table 2.18 that Illinois as a representative of a federal, presidential democracy is at least as repressive as are the United Kingdom and France. The t-values show much more interaction and activity in the United States than in our most repressive European contexts. The system of interaction in Illinois is nonetheless stable. The eigenvalues here too are real numbers within the bounds of −1 and 1. One puzzle is whether the city of Chicago, by far the largest city in the state, is responsible for these results alone. The city has a history of racial segregation and of police brutality. One way to determine this is to test the Chicago events versus the rest of the state. Tables 2.19 and 2.20 tell this story. It is clear from these results that not Chicago, but the rest of the state, usually termed “downstate Illinois” is responsible for the repression and for all parameters rendered statistically significant. These are counter-intuitive findings, but they are empirically based and show that smaller cities and towns are much more likely to interact with dissidents and to repress them than does a large metropolitan urban

Table 2.18 Illinois Lotka–Volterra results, 1985–2005 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.6769∗ 0.6652∗ 0.000399∗ 0.000225∗ λ1 = 0.071395832

80.01 330.76 3.43 4.0 λ2 = −0.003930832

0.0001 0.0001 0.0006 0.0001

∗


2.7 Survey of the West European Democracies and Illinois

27

Table 2.19 Chicago Lotka–Volterra results, 1985–2005 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 7, 615 ∗

Estimate ∗

0.4797 0.001424 −0.000034 0.00002∗ λ1 = 0.479699

t-value

p(t)

33.65 0.05 0.06 5.15 λ2 = 0.000201009

0.0001 0.964 0.9539 0.0001

indicates statistical significance Table 2.20 Non-Chicago Illinois Lotka–Volterra results, 1985–2005 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 )

Parameter

Estimate

t-value

p(t)


0.9456∗ 0.9949∗ −0.000543∗ −0.00000195∗ λ1 = 0.946170964

204.68 283.7 3.21 2.74 λ2 = −0.00057914

0.0001 0.0001 0.0013 0.0062

∗


area. In the Austrian results, three of the four parameter estimates were statistically significant. Germany showed a different pattern with little interaction. Only the a and b parameters were statistically meaningful in Germany. In Switzerland, with much more local control, only one parameter (b) was statistically significant. So, there is some evidence that allowing local control leads to more interaction between state agents and dissidents, as well as more repression. This might simply be a cost of local autonomy, which on the whole is desirable in most contexts. It is most likely caused by individual units with more energy and harsher attitudes toward dissent. While these tendencies might be controlled in a unitary state, they remain unfettered in a federal one (see King, 1982).

2.7 Survey of the West European Democracies and Illinois What do we know from our results summary of West European countries and the state of Illinois? Notable from our findings are the facts that only France and the United Kingdom in Europe, but Illinois as a whole in the United States also meets all the parameters of the Lotka–Volterra model with statistical significance. It is especially likely that each of these two European countries would meet such an interactive profile. The United Kingdom has no constitution, unless one regards all (contradicting) bills of the House of Commons and House of Lords a constitution, but they

28


clearly are not. So, the United Kingdom lacks what all other countries have: a bill of human rights at the level of an organic law. France is in a different position, with both a constitution and a bill of human rights. In 1958, Charles DeGaulle and Michelle Debre invented a new form of government: semi-presidential government. It is notable because it generates only one complete veto player: the president. The French president appoints and dismisses the prime minister (a fractional veto player) and cabinet and cannot be impeached. The French executive has more control of the legislature than in any other West European country save the United Kingdom, where the prime minister is the only veto player. From the street perspective, at least, these two countries are the most repressive in Western Europe, largely because they face more protest. Let us leave our street-view and look at this situation from an institutional perspective. The most applicable institutional perspective on democratic countries is George Tsebelis’s work on veto players Tsebelis (2002). From Tsebelis’s analysis of Doering’s (1995b) investigations, the countries in Europe with the fewest veto players (i.e., higher levels of policy instability and institutional control) are Greece and the United Kingdom, with one veto player each, and France with 1.57 (some power is imputed to the premier) veto players. Fewer than two veto players means that decisions can be made by one leader, as long as there is no level of opposition that would lead to ouster or nonre-election. A second way to think about this problem is in terms of agenda control. As we know from Richard McCelvey’s “chaos” theorem, whoever controls an agenda in institutional terms controls outcomes, unless there is policy symmetry, which is rare. The highest level of agenda control exists in the United Kingdom (0.69), followed by Ireland (0.519) and France (0.333). Ireland still suffers from the split engendered in its civil war (see Chapter 5). It has two veto players, to UK’s one and France’s 1.57. So, agenda control peaks again in our two countries, one with no constitution and one with the original semi-presidential form of government. The final institutional perspective for our inference comprises executive (vs. legislative) control (see Tsebelis, 2002). If we leave aside Switzerland, which has a locally and referendum-dominated government, the highest scores in Europe on executive control are for France and the United Kingdom (both 5.52). In other words, from an institutional perspective, France and the United Kingdom are the only recurring countries with the fewest veto players, the highest agenda control, and the highest executive control. Indeed, another measure of central control is legislative control of plenary agenda and legislative output. On these measures, UK and Ireland score highest, with France and Greece the runners-up (Doering, 1995a). These are the only countries in the “high” measure of political control. But Ireland’s empirical results demonstrate that on the street, it is peaceful and does not interfere much with standard citizen rights (abortion is an entirely different matter). Greece’s parameter estimates are all statistically significant except for h, the repression interactive parameter. Once more, France and the United Kingdom stand out as democracies with repression on the street and with centralized executive control of politics. Our point here is that our street-level results correlate with institutional investigations that show high levels of control of a single leader in the UK and France. In other places with high control, notably Greece and Ireland, there is much less

2.7 Survey of the West European Democracies and Illinois

29

street-level evidence of state control and limited human rights. In Greece, three parameter estimates are statistically significant, but t-values are relatively low compared with France. In Ireland, we see no evidence of street-level interaction. Since the United Kingdom’s model parameters are statistically significant, we need to consider that Northern Ireland was part of its data mix for the 16 years, 14 of which were characterized by Catholic vs. Protestant terror and state attempts to prevent and reduce it. Obviously, when Northern Ireland terror is separated from the United Kingdom data, we would expect differences, but not necessarily what we see in Table 2.21. Here, only the repression parameter estimates are statistically significant, but the eigenvalues indicate instability. Figure 2.4 depicts the oscillatory pattern of terror and repression in Northern Ireland. This is a graph simulated from the parameter estimates for Northern Ireland.2 It is apparent from the figure that Northern Ireland terror oscillates in a tight periodicity. This makes sense in a local tit-for-tat terror exchange. We would not expect the system to diverge, but to oscillate according to the high frequency of bombings, shootings, and kidnappings (see Francisco, 1996). The reason for the tight oscillation is apparent from the eigenvalues. Although they are complex conjugate eigenvalues, both parts are small fractions, which when taken to a 5,884 exponent will go to zero as a limit. We know that Northern Ireland has damped oscillation, indicating terror. But what is the explanation for small towns and rural areas in Illinois displaying a repressive form? The United States is a federal country, unlike most European countries except for Germany and Austria? Police functions are delegated to the local level. It is apparent from our results that this delegation leads to a large amount of police–dissident interaction and repression. We can disaggregate repression among the democracies we cover in this chapter. Table 2.22 shows the daily means of arrests, injuries, and deaths for each country and/or region. It is clear that most of the smaller countries have the least repression, especially Iceland, Luxembourg, Norway, and Portugal. France, Germany, Spain, and UK are larger countries with the most repression. Other countries, both larger and smaller, fall in between these two poles. The Illinois data show that Chicago Table 2.21 Northern Ireland Lotka–Volterra results, 1980–1995 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.0072 0.1356∗ −0.000102 −0.000001302∗ λ1 = 0.003599935 + 0.00093313i

0.5 9.3 0.11 2.96 λ2 = 0.003599935 − 00093313i

0.6184 0.0001 0.9150 0.0031

∗


2 Thanks to Paul Johnson, Department of Political Science, University of Kansas for programming R to simulate the parameter estimates of the Lotka–Volterra model.


600 400

0

200

cbind(RR, CC)

800

1000

30

0

200

400

600

800

1000

Fig. 2.4 Simulated northern Ireland terror & repression, 1980–1995

arrests dissidents at a far lower rate than downstate Illinois, but that dissidents in Chicago are slightly more probable to be injured or killed during interaction with police.

2.8 Discussion We have tested 21 time-series data with the predator–prey model to investigate the relationship between protest and repression in Europe. What have we found? First, that in almost all cases, protest arises in the absence of repression. We would expect this in a democratic setting. In most cases, the b parameter is statistically significant, indicating repression arising in the absence of protest. Second, we saw that UK in the 16 years of Margaret Thatcher’s and John Major’s terms as prime minister, the model’s parameters fit the predator–prey model well and completely. In France this is the case as well, but in no other countries of democratic Europe. If nothing else, this should remind us that free and democratic states with a high degree of central control and limited chance for citizens to talk to powerful legislators have more repression. It is striking, however, how much more repressive (in the sense of arrests only) downstate Illinois is compared either with Chicago or with Europe as a whole. Third, in seven of our European cases, at least one of the interactive parameters was statistically significant. For the most part this is g, the interactive parameter in

Bibliography

31 Table 2.22 Means of arrests, injuries and deaths per day in all cases Country/Case

Arrests

Injuries

Deaths

Austria Belgium Denmark France Germany Greece Iceland Ireland Italy Luxembourg Netherlands Norway Portugal Spain Sweden Switzerland United Kingdom Northern Ireland Illinois Chicago Non-Chicago Illinois

0.112127 0.2447 0.1261 5.0118985 5.238697 0.215 0.0012771 0.11795 0.41231 0.0076461 0.22628 0.098952 0.0216386 6.974475 0.3275735 0.6930913 3.2749 0.4782364 44.5414314 0.6787023 65.5129453

0.0082312 0.067724 0.0161 1.421 0.6562763 0.1433 0 0.041651 0.3545352 0.0003641 0.1176812 0.0424867 0.0503052 0.9882 0.0023897 0.29696068 0.20166 0.0801126 0.0059094 0.0112507 0.000389332

0.000183 0.000178 0.0001846 0.011374 0.0033698 0.0022692 0 0.000189322 0.0028227 0 0 0 0.0018495 0.020229 0.000183824 0.000555568 0.02452 0.021576 0.021576 0.0010466 0.000583998

the protest equation. Belgium, Denmark, Greece, and Spain have this profile. Only France and the United Kingdom generate statistical significance in both interactive parameters, and Austria is the only European country in which the h parameter in the repression equation met statistical significance. Northern Ireland, the terror-plagued region of the UK and Chicago, also met this profile. Finally, all the cases we test in this chapter are stable except Northern Ireland— all others remain in equilibrium. In tests from data over most of the world (not only democracies), only Northern Ireland during its terror period generated two complex conjugate eigenvalues, essentially close to a perilously unstable equilibrium. All other cases have been stable in the difference equations sense, that is, all eigenvalues are real numbers and all occur between –1 and 1 (Goldberg, 1986; Elaydi, 1996). If we assume that both the regime and its dissidents think and that both are averse to injury and death, we can understand why protest and revolution might appear to be disorderly and unstable. However, research in this chapter and elsewhere show that order prevails.

Bibliography DeNardo, James. 1985. Power In Numbers: The Political Strategy of Protest and Rebellion. Princeton, NJ: Princeton University Press. Doering, Herbert. 1995a. “Fewer Though Presumably more Conflictual Bills: Parliamentary Government Acting as a Monopolist.” In Parliaments and Majority Rule in Western Europe, New York: St. Martin’s Press. Doering, Herbert. 1995b. Parliaments and Majority Rule in Western Europe. New York: St. Martin’s Press.

32


Elaydi, Saber N. 1996. An Introduction to Difference Equations. New York: Springer Verlag. Francisco, Ronald A. 1996. “Coercion and Protest: An Empirical Test in Two Democratic States.” American Journal of Political Science 40(4):1179–1204. Goldberg, Samuel. 1986. Introduction to Difference Equations. New York: Dover Publications. King, Gary. 1989. Unifying Political Methodology. New York: Cambridge University Press. King, Preston. 1982. Federalism and Federation. Baltimore, MD: Johns Hopkins University Press. Lichbach, Mark I. 1987. “Deterrence or Escalation? The Puzzle of Aggregate Studies of Repression and Dissent.” Journal of Conflict Resolution 31:266–297. Lichbach, Mark I. 1995. The Rebel’s Dilemma. Ann Arbor: University of Michigan Press. Lichbach, Mark I. 1996. The Cooperator’s Dilemma. Ann Arbor: University of Michigan Press. Oliver Pamela E. and Daniel J. Myers. 2003. “The Coevolution of Social Movements.” Mobilization 8(1):1–24. Tsebelis, George. 2002. Veto Players: How Political Institutions Work. Princeton: Princeton University Press.

Chapter 3

The Dynamics of Protest and Repression in Dictatorships and Democratic Transitions

We have lost the measure of freedom. We have no means of determining where it begins and where it ends. Aleksander Solzhenitsyn, The Gulag Archipelago

3.1 Introduction We investigate nondemocratic governments in this chapter. All are dictatorships in some form, but we do not expect that all will be equally repressive. One purpose of this chapter is to illustrate how different dictatorships repress at quite varied levels. Repression imposes costs for regimes. Those who can rule without a great deal of repression lower costs. For the most part we have communist dictatorships in Eastern Europe, plus one of the former USSR republics and also data from Burma for the 1988–1989 democracy movement and its subsequent harsh repression. Because the communist governments in East Central Europe stopped being so in 1989 and 1990, we also have data on the varied transition to democracy. This chapter will devote a section to the dynamics of such democratic transitions. The governments in this chapter do not tolerate dissent. That does not mean that they commit regular repression: most do not. Rather, they discourage dissent openly. Signals are sent by mass media that everyone should be pleased and quiescent. When someone does challenge the regime, especially a group not representative of workers or farmers, repression is quick, harsh, and widely known, if not widely reported. That is the typical situation in the cases we have in our sample for this chapter. Of course, there are instances in which dissent is not tolerated at all. The closest we come to this situation is in some of the harshest East Central European communist countries (e.g., Romania) as well as Burma in Asia and Belarus, a former republic of the Soviet Union. Two theorists are most relevant for dictatorships: James DeNardo and Ronald Wintrobe. DeNardo used formal models to discover what citizens might do if there were harsh repression (DeNardo, 1985). DeNardo shows how a resolute strategy against a dictatorship can force concessions and create greater peril for the dictator. One empirical difficulty with DeNardo is that he assumes people are mobilizable with ideology and that there is no limit on mobilization. We now know that five percent of the local population is the maximal limit of mobilization (Lichbach, 1995). Wintrobe takes the perspective of the dictator himself (there has never been a female R.A. Francisco, Dynamics of Conflict, DOI 10.1007/978-0-387-75242-6 3, C Springer Science+Business Media, LLC 2009

33

34

3 Dynamics of Protest and Repression in Dictatorships and Democratic Transitions

dictator) (Wintrobe, 1998, 106). Dictators face an array of dangers: (1) a cabal of associates; (2) loss of support by bureaucrats and the army; (3) a revolt of the mob; and (4) foreign intervention. Of these perils, the fourth is irrelevant because of the strength of the USSR in our sample cases. The secret police in each country sought to diminish the third danger and the communist bosses used the party apparatus to control the military and much of the bureaucracy. That left only a coup danger, and there were few of those in the life of the Soviet Union and its allies.

3.2 Cases and the Context of Dictatorship We have a reasonably wide distribution of authoritarian countries to test. Almost all of them, though, occur in Europe. Our cases of dictatorships are Albania, Belarus, Bulgaria, Burma, Czechoslovakia, German Democratic Republic, Hungary, Poland, and Romania. For democratic transitions to democracy we have Albania, Bulgaria, the Czech Republic, Hungary, Poland, and Slovakia.

3.2.1 Mobilization Under Dictatorship and Harsh Repression We know that mobilization happens even in the harshest environments. Little actual protest occurs, but there will be many attempts of clandestine mobilization. Tilly (1986, 332) describes the pattern of mobilization under serious repression: the Nazi occupation of Paris. Open defiance was dangerous and difficult. Those Parisian energies that were not commandeered or snuffed out by the occupying power and its French collaborators flowed mainly into survival, individual, and collective: the creation of escape routes, information channels, black markets, and networks of mutual aid. Slowly and later, however, a few of these half-hidden structures became means of collective resistance.

Many groups fought against Nazism during Hitler’s regime in Germany (Hoffman, 1977; Gill, 1994). But it was in Nazi-occupied Austria that clandestine mobilization found its most articulate expression. Bauer (1939), an Austrian social democrat, published in Paris a manual for dissidents who sought to maintain their organizations during harsh repression under a dictatorship. Bauer’s (1939) key points were (1) a cadre (cell-based) structure; (2) anonymous leaders; and (3) exclusion of probable spies, people who have families or other affective connections and those who are not fully committed to the cause. In other words, a clandestine organization survives only by becoming invisible both to the public and to the state. Recruited members in such associations tend to be loners; they are unattached people, have few family members, and they substitute the group for true family. These predictions are consistent in history (Almond, 1954; Kornhauser, 1959) and in the late 20th century, for example, the Provisional Irish Republican Army (Toolis, 1995). The membership of a clandestine cell-based organization cannot mobilize many people and cannot act publicly–terror is possible, but members cannot otherwise reveal themselves. Organizations of this type are first and foremost maintenance

3.2 Cases and the Context of Dictatorship

35

groups—they attempt to preserve themselves through the dictatorship. DeNardo (1985, 43) argues that the style of mobilization determines the strategy of action and the number of people acting. Certainly, that is correct in clandestine recruitment. The nature of protected and exclusion mobilization requires few people and nonpublic acts. The implication of these facts for our analysis is that a great deal of the dissent in dictatorships is hidden from view. Mobilization is not publicized, and therefore not including in the data we coded. In the sample of countries we have, the samizdat (clandestine newspapers) movements would not be in the event data. Thus, we would not necessarily expect results from the model to appear as harsh repression in all countries. In many countries, the barriers to action were well known and respected; street dissent was considered insane, not necessarily because of direct repression, but because of the loss of education opportunities, employment, or state health benefits. With a protected leadership and membership of unattached people, the group can be stable during a dictatorship. What happens, however, when the regime falls— especially if it does not fall to the clandestine organization? History reveals many examples of this sort. The clearest point is an exile leadership group that equals the clandestine organization in the home country. Almost always the exile and internal leaders contest each other’s right to legitimate power. The internal leaders seek to lead since they survived repression. If they win, they face challenging mobilization problems, since their organization has nonfamily oriented, unattached members. Who wants to listen to or follow a lonely and alienated person? Such was the problem that confronted the German Social Democratic party faced after World War II. Although the internal leaders won, they lost every election thereafter until the exile leaders (e.g., Willy Brandt) captured the party in 1959. Mobilization is easier when the exiled leaders ascend to power almost immediately. For example, when the exiled Bolshevik command took over after the Russian revolution in 1917 (Lenin, Trotsky, and Stalin) mobilization faced little difficulty. The situation above exists in constant repression under a dictatorship. But what about other sorts of states that do not repress consistently–those that are autocratic and occasionally massacre their citizens? We have tested scenarios like this when occupations of other countries occurred (e.g., UK in Ireland, UK in India, France in Germany, or Israel in Palestine) and also in highly repressive, if inconsistent, regimes, for example, Burma. In the wake of 31 urban twentieth-century massacres across the world, we found impressive backlash mobilization. In almost all cases, backlash mobilization after a massacre dwarfed the original event massacre. Yet, Bayesian updating test shows that when backlash was repressed, mobilization was highly and quickly damped (Francisco, 2004).

3.2.2 What Happens to Repression When Mobilization Grows to a High Magnitude? It depends. If the mobilization grows beyond the regime’s capacity to repress it, then there is low probability of repression. The most obvious example of this

36


phenomenon is what happened in the summer and fall of 1989 to most of the countries in our sample. Especially in east central Europe, large-scale mobilization essentially precluded repressive deployment. Too many people came to the street to argue that this was a peculiar minority. We saw this as well in 2004/2005 in Kyiv during the Orange revolution of Ukraine. If the regime is able to repress, it enjoys the choice of acceding to or denying the dissidents’ demands. Many states find serious indecision in this case. For example, the fraudulent election in Kyrgyzstan led to huge mobilization, especially in the south of the country. The regime sought to repress dissidents, but was unable to mobilize the forces to contain the protest, resulting in a successful nonviolent revolution.

3.3 Empirical Results on Dictatorship Periods We turn now to the results of our tests, first on dictatorships, and then on transition regimes. Once again our model of choice is Lotka–Volterra. As with democracies, we will tour the dictatorships alphabetically and attempt to discern patterns among them. We begin with Albania during the latter period dominated by communist strongman Enver Hoxha, who died in 1985. Table 3.1 contains the results from our tests. Albania and Romania were two communist states that defected from the Soviet bloc and nonetheless maintained communism as their governing ideology. As the results in Table 3.1 indicate, throughout the decade of the 1990s, Albania was a repressive autonomous communist dictatorship. It generates extremely high t-values. In spite of its high level of repressive interaction, as a system it remained stable. Both eigenvalues are bounded by −1 and 1, in fact by an even narrow range of zero to one. While the Albanian authorities attempted to seal hermetically the country from all others, by the late 1980s, Albanian citizens were tuning into radio broadcasts from Yugoslavia and Italy. What they heard about Mikhail Gorbachev and the liberalization of the USSR generated a great deal of interest. In May 1989, a conference of writers and artists discussed the need for open dialog and criticism of regime policies. A few months thereafter, a novel was published that directly attacked the Sigurimi, the Albanian secret police (Ramet, 1991). Even in this most-closed society, the ideas of freedom could no longer be suppressed. Table 3.1 Lotka–Volterra results, Albania 1980–1989, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.9996∗ 0.877386∗ −0.0476∗ 0.04182∗ λ1 = 0.9538059

492.39 13.8 52.67 13.3 λ2 = 0.0876141

0.0001 0.0001 0.0001 0.0001

∗


3.3 Empirical Results on Dictatorship Periods

37

Table 3.2 Lotka–Volterra results, Belarus, 1995–1997, weekly aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 156 ∗

Estimate ∗

0.13155 0.16941∗ 0.00001 −0.00000431 λ1 = 0.1315629

t-value

p(t)

4.54 1.99 0.85 1.75 λ2 = −0.00001718631

0.0001 0.0464 0.3935 0.0810


The parameter estimates in Table 3.2 show eagerness on the part of dissidents and police, but relatively little interaction with one another. These results represent university students clashing with police each week during the three years of 1995–1997. Standard arrests, common injuries, especially bruises and sometimes broken arms or legs occurred, but for the most part the regular repression stayed within these bounds. Students, dissidents with unstructured time, continued to act. But most citizens stayed home, eschewing interaction with law enforcement officers. Note that interaction with the police dampens repression slightly, and not quite statistical significantly (parameter h). These data were coded by Marc Nordberg and differ from the rest of our data in the sense that they are aggregated not daily but weekly. Belarus was not destined to be this way. In the early collapse of the USSR and independence for individual republics, Belarus was a reasonably democratic entity. It only devolved to repression when Aleksander Lukashenko was elected president. Belarus is controlled by Lukashenko, a hardline unreformed-communist dictator with persistent hopes to merge his country with mother Russia. So far at least, this is unrequited love. The waiting has led to a sustained status-quo in Belarus, including regular repression of unhappy citizens. Hope for meaningful reform in Belarus disappeared when Vladimir Putin became President and then Premier of Russia. Without a push from Russia, not much will happen in Belarus until Lukashenko is gone. Bulgaria, our next case, has many similarities to Belarus, despite the distance between them. Like Belarus, Bulgaria was a loyal, unquestioning communist ally to the USSR. This quiescence was accomplished by keeping the population ignorant: (1) the press and other media were barren and without information value; (2) statistical information was unreliable at best; (3) censorship was virtually total, even on books or articles that carried nuanced criticism of the regime; and (4) the state publishing firm ignored demand and published only what the leadership wanted (Ramet, 1991). All this kept the Bulgarian populace focused on family, work, and leisure, largely ignoring politics and newspapers. The secret police ferreted out any dissent and punished it ruthlessly. The first human rights organization in Bulgaria emerged only in 1988 (Ramet, 1991). Bulgarian results (Table 3.3) show a country without a great deal of regular repression. Both dissidents and police rise on their own (parameters a and b), but the interaction of police and dissidents in both equations is negative and only

38

3 Dynamics of Protest and Repression in Dictatorships and Democratic Transitions Table 3.3 Lotka–Volterra results, Bulgaria, 1980–1989, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 3, 652 ∗

Estimate ∗

0.1723 0.7756∗ −0.000031 −0.00006∗ λ1 = 0.72604

t-value

p(t)

9.84 17.1 0.39 10.9 λ2 = 0.0000796048

0.0001 0.0001 0.6957 0.0001


statistically significant in the repression equation (parameter h). Todor Zhivkov was the loyal-to-USSR communist dictator from 1956 until his ouster in 1989. So loyal was Zhivkov that many believe he attempted to make Bulgaria the sixteenth republic of the USSR. During early 1989, the renewed campaign of Bulgarian nationalization (mainly changing Turkish names to Bulgarian) generated substantial ill-will and a mass exodus of Turks to Turkey. In July 1989, the Discussion Club for the Support of Glasnost and Perestroika summoned up the courage to protest the treatment of Turks in Bulgaria (Ramet, 1991). August in the fateful year 1989 brought clashes in ethnic-Turkish villages in Bulgaria with state troops (Ramet, 1991). All this was handled badly and resulted in large-scale international condemnation of Zhivkov. In November, Zhivkov was removed and replaced by Petar Mladenov, who actually met with and encouraged dissident organizations in mid-November 1989. Subsequently, he allowed the formation of opposition political parties. We have one case wholly outside the communist world. That case is Burma, or as its military dictators call it, Myanmar. The Burma data were coded by Federico Ferrara and are aggregated daily, but unlike the rest of the sample, they only cover 244 days of the democratic rising in 1988 (Ferrara, 2003). The rising was repressed brutally and creatively by seeming to accede to the dissidents’ request for the release of political prisoners. The prisoners released, though, were rapists, murderers, and armed robbers. The state suspended social order for a few days to get everyone off the streets. Then a military coup by General Saw Maung sealed the fate of the democratic movement (Ferrara, 2003). The communist regime in Czechoslovakia was headed by Slovak hardliner Gustáv Husák, the leader chosen by the Soviet Union after the Alexander Dub˘cekled Czechoslovakia Prague Spring in 1968 was repressed and shut down by Warsaw Pact troops invading the country. Husák was “elected” by the Czechoslovak communist party in 1969 and served until his own ouster by the new Soviet government. Milo˘s Jake˘s took over as leader until he resigned during regime transition in December 1989. The dissident movement Charter 77 began in 1977 after the signing of the Helsinki accords and the repression of the rock group the Plastic People of the Universe. Charter 77 was immediately repressed. But because it sent signature names put on its charter into diplomatic bags that were sent to the West European press, the regime could not repress the movement as it otherwise might


39

Table 3.4 Lotka–Volterra results for Burma, 1988, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 244 ∗

Estimate ∗

0.537906 0.7655∗ 0.00052∗ −0.000000199∗ λ1 = 0.5386450028

t-value

p(t)

9.28 13.25 3.09 5.93 λ2 = −0.0007388038

0.0001 0.0002 0.0022 0.0001


have without stirring up trouble across the barbed wire. The group operated in the open, albeit under the intense scrutiny of the secret police (Ramet, 1991). By 1987, 1,300 Czechoslovak citizens had signed the charter; despite the intense pressure, only 15 signatories withdrew their signatures (Ramet, 1991). Charter 77’s leader was Václav Havel, who later emerged as a transition political leader. Much of the mobilization in Czechoslovakia was in clandestine form. Since the regime precluded jazz music, there were many jazz organizations that met in family flats (usually rotating) each week or month. This sort of mobilization was hidden from the authorities. The dissent was listening to prohibited music. But because this activity was so widespread, when the regime weakened in late 1989, there was already a good deal of latent mobilization. After all, these jazz clubs were a great venue for more generalized political complaining (Ramet, 1991). This erupted into almost a nation-wide school and university boycott in November 1989. Yet at this time the regime had no way to repress such broad and dispersed mobilization. This phenomenon is part of a larger set of dictatorship problems; see Timur Kuran (1995). In the end it took only 21 days to remove the entrenched communist dictatorship and enter the new era of free elections by June 1990. Czechoslovakia’s results (Table 3.5) show only the repression equation parameter estimates to be statistically significant. Nonetheless, parameter h has a negative sign, so interaction between protesters and police dampened repression. Even a (the parameter representing the rise of protest in the absence of repression) has

Table 3.5 Lotka–Volterra Results, Czechoslovakia, 1980–1989, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.000214 0.0854∗ 0.00014 −0.000016∗ λ1 = 0.001198466

0.01 3.66 0.02 2.25 λ2 = −0.001000466

0.9900 0.0003 0.9813 0.0246

∗


40


an extremely low value, probably because of the prominence and visibility of the Charter 77 movement. The German Democratic Republic (GDR) was carved out of Germany after World War II. It was the Soviet Union’s occupation sector of Germany and became the GDR after West Germany (Federal Republic of Germany) was created in 1948 and 1949. The GDR was the front-line cold war state and was a loyal member of the Soviet bloc. During the period from 1945 to August 12, 1961 millions of East Germans voted with their feet and fled to the West, especially through West Berlin, a capitalist obstacle island within the territory of the GDR. This population loss was significant, in particular, because those who left were generally higher skilled than those who stayed behind. The Berlin wall was built on August 13, 1961 and thereafter leaving the republic without permission became a capital crime. The GDR was a repressive state, but its reputation for intolerance of protest prevented most large-scale mobilization. We see in Table 3.6 that the parameter estimates of a and b are statistically significant, but none of the interactive variables is significant. So protest mobilization arose without repression and repression arose without protest mobilization, but there was relatively little interaction between protest and repression on the streets of the GDR. For the most part, the Stasi, East German secret police, operated in clandestine fashion and made massive arrests at various times to intimidate dissent. In January 1988, the Stasi arrested more than 70 of the most prominent democratic activists and deported them to the West (Ramet, 1991). But by this time Gorbachev had set a liberal tone in the Soviet bloc. Remaining dissidents, led by Bärbel Bohley and Sebastian Pflugbeil, reorganized in the wake of the deportations and formed a new organization, symbolically announcing it on August 13, 1989: Neues Forum, or New Forum, originally designed as a pressure group for dialog with the regime (Ramet, 1991). This group was the rallying point for the dissent that exploded throughout the GDR in the fall of 1989 and eventually brought down the regime. The circumstances of Hungary’s role in the Warsaw Pact and the Soviet sphere altered in the wake of its 1956 rising against USSR-imposed communism. In contrast to the USSR’s reaction to the Prague Spring in 1968, Soviet leaders in the 1950s insisted on no departure from the base Soviet model, but gave Hungary

Table 3.6 Lotka–Volterra results, German Democratic Republic, 1980–1990, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.6842∗ 0.041508∗ −0.001417 0.0000007855 λ1 = 0.684114

53.94 2.28 0.74 0.03 λ2 = 0.000087608

0.0001 0.0226 0.4593 0.9728

∗



41

a wide latitude to attempt new socialist policies, including private property at a micro-level. The intellectual dissent in Hungary was better tolerated as well than in Czechoslovakia. Consumer goods were more available, living standards higher, and communism not so draconian as in other parts of the bloc. Of course the Hungarians had useful experience in trying wrench autonomy from the Hapsburg empire a halfcentury earlier. Nikita Khrushchev in the 1950s campaigned against the “cult of personality” that had grown up around Stalin. Khrushchev was quick to put down the 1956 rising, but then tolerated reform in Hungary after the “troublemakers” (those who rejected Soviet communism) had been taken care of. Foremost among the latter was Imre Nagy. János Kádár cooperated with the USSR and put the unrepentant Nagy on trial with other leaders. They were found guilty of treason and put to death. Buried in unmarked graves, these dissident leaders led to a smoldering of dissent that only emerged in 1989 when Nagy and his friends were again unearthed and reburied with dignity. From the mid 1960s, Hungary was a substantially more affluent and less repressive member of the Soviet bloc. West Europeans were allowed to set up factories in Hungary led to better wages and a higher-level consumer culture. I was a student at the University of Vienna in 1968–1969; it was literally cheaper and more pleasant for us to spend weekends in Budapest, just 40 miles down the road. Austrian tourism was encouraged (since Austria was a neutral country in the cold war) and developed the economy still further. But in the 1980s, the economic growth rate in Hungary plummeted from 4.8% annually to just 1.8%. Living standards stagnated. By the time of Mikhail Gorbachev’s accession to the General Secretary’s position in the USSR communist party, Hungarians were ready to press for more gains. A meeting of about 155 Hungarians in Latitekek resulted in a new organization, the Democratic Forum. These high-placed dissidents recognized the opportunity afforded them by the liberalization in the USSR and the malaise in the Hungarian communist party (Ramet, 1991). With Hungary’s intense indebtedness to the International Monetary Fund and its signature on the Helskini Accords it was unlikely to repress small episodes of dissent. Kádár was forced to step down the next year, in 1988. After that, reformist elements in the communist party and in intellectual circles seriously (Democratic Forum) negotiated a new path for the country. They met in roundtable fashion in early to mid 1989. Almost 300,000 attended the reburial of Imre Nagy and his cohorts. In October 1989, Hungary ceased being a communist country and began its foray into Western democracy. All this is reflected in Table 3.7, in which only the parameter representing the rise of protest in absence of repression is statistically significant. Table 3.8 shows results for Poland in the 10 years of 1980–1989, that is, from the rise of Solidarity in August 1980 to the regime transition. Here, a, b, and g are statistically significant, signaling the probability that repression was inconsistent. Note that both protest and repression rise in the absence of each other. Interactive repression between state agents and dissidents lowers repression levels. So, we see a profile of a communist dictatorship that was in control much of this time period, but that even when it lost control, for example, August 1980–December 13, 1981 and from the mid-1980s until 1989 it was not a particularly repressive country. That is

42

3 Dynamics of Protest and Repression in Dictatorships and Democratic Transitions Table 3.7 Lotka–Volterra Results, Hungary 1980–1989, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 3, 652 ∗

Estimate ∗

0.171102 0.001825 0.003806 −0.0000000536 λ1 = 0.01711426

t-value

p(t)

10.18 0.08 0.15 0.05 λ2 = −0.00004063934

0.0000 0.9337 0.8826 0.9583


Table 3.8 Lotka–Volterra results, Poland, 1980–1989, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.193008∗ 0.366∗ 0.000001857 −0.000038∗ λ1 = 0.1960115

6.93 14.09 1.96 2.73 λ2 = −0.00004152066

0.0001 0.0001 0.0501 0.0065

∗


not to say that Solidarity and the human rights organization KOR were not repressed, only that the street interaction between dissidents and regime was not as active and extreme as we see in other east central European states. The problem that military and communist authorities faced in Poland was a poor and deteriorating economy and proven widespread public support for a movement it continued to repress. When Mikhail Gorbachev assumed the summit of the communist party in the Soviet Union, the harsh repressive tactics of General Jaruzelski appeared increasingly out of place. Strikes were rampant, the regime could not stop them, and could not supply the population’s needs. Eventually, the regime was forced to make the humbling gesture to Lech Wales¸a to curtail labor strife. Our next case is wholly different in character. Romania was a communist dictatorship that shunned the leading role of the USSR. The principal dissident organization was “Free Romania”, a worker-based antiregime group that nonetheless supported Gorbachev’s reform communism. But this was necessarily a clandestine movement (Ramet, 1991). It found a natural alliance with the Hungarian minority in Romania. Free Romania used dissident Hungarian printing facilities and these two groups formally protested the regime’s state intent to destroy seven to eight thousand villages in Romania (Ramet, 1991). It is clear that Romania was a repressive state with a good deal of interaction between state agents and dissidents. The data underlying Table 3.9 run from 1980 to 1995, in other words, through the transition time that occurred elsewhere in the region. Regime transition in Romania was intensely violent, far more so than for any

3.4 Empirical Results from Transition Periods

43

Table 3.9 Lotka–Volterra results, Romania, 1980–1995, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 5, 844 ∗

Estimate ∗

0.0934 0.4537∗ 0.00072∗ −0.0000030293∗ λ1 = 0.096775585

t-value

p(t)

7.03 30.32 3.58 10.81 λ2 = −0.003372556

0.0001 0.0001 0.0003 0.0001


other east European country. Romania defected from the Soviet bloc in 1958, after the first decade of the cold war. As such, it was an autonomous communist dictatorship run by Nicolae Ceaus¸escu. In the mid-1980s, the economy faltered badly and led to numerous power outages and brownouts. In 1987, workers were emboldened to go out on the street to protest the austerity program implemented by the state. During the December holiday period of 1989, elements of the military joined with dissidents to capture, try, and execute Ceaus¸escu and his wife. An estimated 10,000 total of Romanians died in the subsequent civil war.

3.4 Empirical Results from Transition Periods We now investigate how the results change when communism gave way and democracy grew from 1990 to 1995. For most of our cases such a change did take place, save for Belarus, Burma, and Romania. Belarus and Burma experienced no transition, while Romania’s came somewhat later than other countries in the region. We would expect that there would be less repressive interaction in the transition period, although mobilization increases in such a context, so more repression might occur. Our first transition case is Albania. Because Albania was not part of the Soviet bloc, its transition occurred differently and was not part of the contagion that spread from Poland and Hungary through the GDR to Czechoslovakia and finally Bulgaria. Unrest broke out in northern Albania in January 1990. Within five months, there were clear signs of liberalization. A look at the data underlying Table 3.10 will show how much dissent emerged before the entrenched communist elites decided to grant concessions. It is clear from Table 3.10 that a good deal of interaction took place after regime transition between dissidents and police. The only parameter estimate that is not statistically significant (b) is repression rising in the absence of protest. This is a street-level indication of democratic transition, but the negative sign of the iteration of protesters and police in the protest equation (g) shows that protest grows under interaction. The near-zero parameter estimate of interaction in the repression equation (h) indicates the small rise of repression under dissident-police interaction. The change from the dictatorship era is that the parameter estimate (b) represent-

44

3 Dynamics of Protest and Repression in Dictatorships and Democratic Transitions Table 3.10 Lotka–Volterra results for Albania, 1990–1995, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 2, 191 ∗

Estimate ∗

0.9086 −0.0101 −0.003838∗ 0.000008564∗ λ1 = 0.9086427

t-value

p(t)

94.4 0.4 2.97 5.46 λ2 = −0.00004180486

0.0001 0.6874 0.0030 0.0001


ing repression arising without protest loses statistical significance. Also, t-values decline greatly in Albania between Enver Hoxha-dominated dictatorship to the 1990–1995 transition, marked by the beginning event of the first formation of an opposition political party in 1990. The fall of Todor Zhivkov in November 1989 inaugurated the Bulgarian political transition. On the 17th of November, 100,000 Bulgarians massed in front of the National Assembly in Sofia to protest corruption and dictatorship and support Gorbachev’s reform agenda (Ramet, 1991). The new regime, clearly faced with a more active population than ever before, began to see the light and mend its ways. Meetings with newly mobilized opposition parties were held and brought the promise of free elections. The order for the Turkish to change their names to Bulgarian was rescinded on 29 December (Ramet, 1991). Bulgarian nationalists were not pleased by the recension, so the new regime changed its national day from the day communism began to the day it overthrew Turkish rule for independence (Ramet, 1991). As Table 3.11 demonstrates, only the parameters in the protest equation are statistically significant. It is a mark of some progress that the parameters in the repression equation have t-values near zero. Our Czechoslovak and Czech Republic tables (Tables 3.12 and 3.13) tell a peaceful story about regime transition under the presidency of former dissidentplaywright Václav Havel. In the 1980–1989 era only the repression parameter estimates achieved statistical significance. In the transition period, those parameter

Table 3.11 Lotka–Volterra results for Bulgaria, 1990–1995, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.6159∗ 0.00327 0.0184∗ 0.0000001006 λ1 = 0.615977

34.85 0.14 5.71 0.76 λ2 = −0.00009666985

0.0001 0.8899 0.0001 0.4502

∗



45

Table 3.12 Lotka–Volterra results for Czechoslovakia, 1990–1992, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)

a b g h Eigenvalues N = 665

0.002036 −0.0000000000000000167 −0.07814 0.0000000000000000000136 λ1 = 0.002036

0.05 0 0.12 0 λ2 = −0.0000000000000006409186

0.9573 1.000 0.9067 1.000

Table 3.13 Lotka–Volterra results for the Czech Republic, 1993–1995, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.063254 −0.0000000000000000214 0.003576 0.000000000000000000108 λ1 = 0.063254

1.67 0 0 0 λ2 = −0.00000000000000000131827

0.0945 1.000 0.9979 1.000

estimates lose statistical significance and result in a table with no significance. This case is different from all of our others in the sense that Czechoslovakia stayed unified from 1990 through 1992, but split apart in 1993. Václav Klaus and Vladim´ır Meˇciar became respective presidents of the Czech Republic and Slovakia. As a consequence, Slovakia and the Czech Republic are together until 1992, but have separate three-year samples from 1993 to 1995. We have so little repression or state-dissident interaction in Czechoslovakia, the Czech Republic, and Slovakia (Table 3.16) that we once again confront the linear algebra lower triangle context, where the eigenvalues are close to the main diagonal parameter estimates because of zero or near-zero parameter estimates in the Jacobian matrix (see Chapter 1). The portrait of the street emerging in Prague and its surroundings is a peaceful one. It is noteworthy that the impetus for reform came from the art and cultural world, not the hard-core human rights community. In January 1989, the same month in which Charter 77 leader and playwright Válav Havel was imprisoned, the cultural ministry replaced on library shelves titles of Czech authors that had been banned for decades (Ramet, 1991). New, critical films were released. All this liberalized cultural activity was probably designed to serve as minimal accession to Gorbachev’s reform policies. After all, how many people read library books or attend art films? The really important repression, of active dissidents, did not retreat. In fact, we now know that Gorbachev was unhappy with the Czechoslovak leadership and sought to depose the dictator Miloˇs Jakeˇs (Ramet, 1991). Once mobilization reached a mass level, the communist regime was doomed. There would be no help from Moscow in 1989. The dictatorship surrendered timidly. Within weeks, a broad spectrum of

46


political parties had emerged and Havel stood with Dubˇcek in front of hundreds of thousands of cheering citizens. From this point, repression essentially vanished. The Hungarian transition was perhaps the easiest in all the countries we investigate. The old regime was ossified, the communist party increasingly a joke to most informed citizens; a young cadre of leadership emerged in this context and simply shifted from communism to democracy, without muss or fuss. As in the Czech Republic, repression, never strong in Hungary, also vanished. It is interesting that in many transition cases we find a shift from major to minor repression, while in others, especially the Czech Republic and Hungary, shift from minor to none. As noted earlier, the communist party essentially admitted its inability to manage the economy. So from November 1988 to April 1989, the process of transformation was complete, and Hungary was the first to demolish the “iron curtain” of barbed wire that separated it from Austria. This act, in turn, led the flight of hundreds of East Germans vacationing in socialist Hungary, which eventually led to the fall of the entire Soviet bloc in east central Europe. It is clear that Hungary, which had the least repression during the communist period, has almost none during the transition to democratic rule. As in Chapter 2, we have a lower triangle linear algebraic situation with near-zero parameters in Table 3.14. The first eigenvalue is the same as the parameter estimate a and the second eigenvalue is close to the value of the parameter estimate h, just a bit farther away from zero. The pattern for Hungary is much like the least repressive West European democracies. Only the parameter denoted the rise of protest in the absence of repression (a) is statistically significant. All communist dictatorships were not equally repressive. Nor were repressive communist dictatorships necessarily repressive in political transitions. Nonetheless, the less repressive the country in dictatorship, it is generally true that it will also be less repressive in the movement toward a new political system. Poland continues our parade of impressive declines of repression. Unlike Hungary, Poland (Table 3.15) has no parameter estimate achieving statistical significance. Like Hungary, Poland has one of the unusual (but common in the transition sample) lower triangle matrix contexts in linear algebra. Once again we have two zero t-values and two fractional ones. After Solidarity took over the government, repression more or less ceased. Once again we confront the lower triangle phenomenon. Even the police and security Table 3.14 Lotka–Volterra results for Hungary, 1990–1995, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.0792∗ 0.0000000000000000202 0.001888 0.000000000000000000163 λ1 = 0.0792

3.61 0.0 0.0 0.0 λ2 = −0.0000000000000000006445354

0.0003 1.0000 0.9991 1.0000

∗



47

Table 3.15 Lotka–Volterra results for Poland, 1990–1995, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)

a 0.000000000000000000445 0 b 0.026688 0.85 g −0.0000000000000000000000813 0 h −0.00023 0.15 Eigenvalues λ1 = 0.00023 λ2 = −0.00000000000000000000943383 N = 2, 191

1.0000 0.3928 1.000 0.8814

forces supported the popular legitimacy of Solidarity, something palpably obvious in Poland during the 1980s. The Poles selected a semipresidential government, with Lech Walesa as the first president and Tadeusz Mazowiecki as prime minister. Mazowiecki worked fast to dismantle a state-run economy and convert it to free-enterprise. Poland had no major social divisions. Most Jews had been killed in the Holocaust; those who left largely did not return. Everyone was either nominally Roman Catholic or not, but there were not so many other choices. So from the outset of the transition, there was no reason for security beyond normal police criminal functions. If the Czechoslovak revolution were the velvet one, then the split between the Czech Republic and the Slovak Republic is called the velvet divorce. Once the Soviet yoke was removed from east central Europe, longstanding autonomy movements emerged or shifted from western to east central Europe. The Slovak Independence Movement, an emigre organization based in Munich, moved to Bratislava in the spring of 1990. Part and parcel of the removal of communism was the related idea of independence for the Czech Republic and Slovakia. Manifold Slovak independence parties formed, all with the fervent wish to finally, after centuries of Hapsburg and Soviet rule, to have independence (Ramet, 1991). Vladimir Meˇciar, an aspiring prime minister without a country, was happy to accommodate all these wishes. Although the Slovak Republic started out under a former communist ambitious politician (Meˇciar), it quickly met the standard criteria for democratic consolidation. Independent political parties formed and contested free elections. From 1994 they have cooperated in coalition governments that have exhibited stability (Wolchick, 2008). Although parties remain organizationally weak, they are working Table 3.16 Lotka–Volterra Results for Slovakia, 1993–1995, daily aggregated data Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-value

p(t)


0.000015 0.00000000000837 0.000015 0.00000000000000199 λ1 = 0.00001499999

0 0 0 0 λ2 = 0.0000000000199

0.9997 1.0000 1.000 1.0000

48


through the consolidation that is necessary. The electoral law, of course, determines how many parties will eventually remain. Slovakia uses the largest remainder proportional representation formula, so parties have the incentive to split to gain more seats in the legislature. Both the Czech and Slovak Republics maintain a five percent threshold to keep extremist or frivolous parties at bay. One of the reasons that the Slovak Republic’s transition was so peaceful and devoid of coercion (Table 3.16) was the decision taken in 1990 to dissolve the political function secret police in all of Czechoslovakia and to deprive it of any repressive resources. The unified general transition under Havel also expanded the foreign investment by allowing greater that 49% ownership. Both these measures led to a benign Slovak Republic after 1993 (Ramet, 1991).

3.5 Conclusion Dictatorships are rarely benign and usually terrible, but we have seen many of them transform into peaceful and nonrepressive countries in this chapter. In general, dictatorships are not uniformly repressive. The worst countries in our dictatorship sample were Albania, Belarus, Bulgaria, Burma, Czechoslovakia, Poland, and Romania. In transition, Albania remained repressive, Bulgaria slightly so, and in Czechoslovakia, the Czech Republic and Slovakia, it disappeared. In Hungary there was little repression under communism and none thereafter. Belarus, Burma, and Romania remained repressive and for the most part the former two have not reformed at all. Repression in an authoritarian state that has consistent yet reasonably mild repression can either see very little mobilization or alternatively small and consistent protest. Also observe that in all these results in tables, the eigenvalues are members of the real number system (as opposed to complex conjugate numbers); thus, they signify that the interaction of dissidents and the police stays in equilibrium (Elaydi, 1996; Goldberg, 1986; Merkin, 1997; Francisco, 2000). As Tilly (1986, 4) puts it, “conflict, not disorder.” All these estimations were calculated with SAS Model using seemingly unrelated regression with correction for serial correlation and heteroscedicity when necessary. Do these parameter estimates make a difference for the mechanism between protest and repression? Which of our dictatorships is most repressive? The contestants are for this distinction are only Albania, Burma, and Romania, the only countries for whom all parameters are statistically significant. Judging by t-values, Albania wins hands down (see Table 3.1). But while repression was almost consistent, it was not terribly violent. The maximum arrested at any given time was two, and the maximum killed was 21. No one was injured by police in Albania, or at least none was reported. In the Burma case we have only deaths, since the regime mostly shot people dead. The mean of deaths per day in Burma is 26.3443, a large number of people killed per day. From this perspective, Burma is clearly the most repressive. What about Romania? Its mean arrest number per day is 0.725; for injuries, it is 0.491; and for deaths, 0.1539. This is a logical decline from arrests, injuries to deaths. The maximum arrests in Romania were 1,049; maximal injuries

Bibliography

49

765; and highest incident of deaths, 610. Based on the foregoing evidence, Burma would have to be judged the worst violator of human rights in our rouge gallery of dictatorships. Of the countries in the transition sample, only Albania and Bulgaria show clear signs of continued repressive interaction. The unexpected and truly felicitous news is that by far the majority of the east central European cases are not only not repressive, but most actually meet the requirements of the lower triangle linear algebra context, which are rarely met (see Tables 3.12, 3.13, 3.14, 3.15 and 3.16). To have an entire class of countries meet, these requirements in the same time period is remarkable. Czechoslovakia, the Czech Republic, Hungary, Poland, and the Slovak Republic all have near-zero parameter estimates and at least some zero-level t-values. In this sense, these are cases devoid of repression, less intrusive on the street than are even most democratic west European countries in Chapter 2. They say that old habits die hard; not so in these cases. It is a remarkable turnaround from communist dictatorship to total noninterference at the street level. The forces who took over in the wake of the communists, in some cases communists themselves, created policies totally different from the palpably recent past. Their citizens enjoy freedom denied to them for nearly a half-century.

Bibliography Almond, Gabriel. 1954. The Appeals of Communism. Princeton: Princeton University Press. Bauer, Otto. 1939. Die illegale Partei. Paris: Editions “La Lutte Socialiste”. DeNardo, James. 1985. Power In Numbers: The Political Strategy of Protest and Rebellion. Princeton, NJ: Princeton University Press. Elaydi, Saber N. 1996. An Introduction to Difference Equations. New York: Springer Verlag. Ferrara, Federico. 2003. “Why Regimes Create Disorder: Hobbe’s Dilemma during a Rangoon Summer.” Journal of Conflict Resolution 47(3):302–325. Francisco, Ronald A. 2000. “Paths to State Repression.” chap. “Why are Collective Conflicts ’Stable’?”, pages 149–172, Lanham, MD: Rowman and Littlefield. Francisco, Ronald A. 2004. “After the Massacre: Mobilization in the Wake of Harsh Repression.” Mobilization 9(2):107–126. Gill, Anton. 1994. An Honorable Defeat: A History of German Resistance to Hitler. New York: Henry Holt. Goldberg, Samuel. 1986. Introduction to Difference Equations. New York: Dover Publications. Hoffman, Peter. 1977. The History of German Resistance, 1933–1945. Cambridge, MA: The MIT Press. Kornhauser, William. 1959. Politics of Mass Society. New York: The Free Press. Lichbach, Mark I. 1995. The Rebel’s Dilemma. Ann Arbor: University of Michigan Press. Merkin, David R. 1997. Introduction to the Theory of Stability. New York: Springer Verlag. Ramet, Sabrina P. 1991. Social Currents in Eastern Europe. Durham, NC: Duke University Press. Tilly, Charles. 1986. The Contentious French. Cambridge, MA: Harvard University Press,. Timur Kuran. 1995. Private Truths, Public Lies: The Social Consequences of Preference Falsification. Cambridge, MA: Harvard University Press. Toolis, Kevin. 1995. Rebel Hearts: Journeys Within the IRA’s Soul. New York: St. Martin’s Press. Wintrobe, Ronald. 1998. The Political Economy of Dictatorship. New York: Cambridge University Press. Wolchick, Sharon L. 2008. “The Czech and Slovak Republics.” In Central and East European Politics: From Communism to Democracy, Lanham, MD: Rowman and Littlefield.

Chapter 4

Varied Dynamics of Bandwagon Mobilization

With every step, the crowd swelled like a river being fed by tributaries. At first it was only students, but then more and more adults joined in. The crowd thickened, occupying entire avenues. The trams were forced to stop running, because people overflowed onto the tracks in the street. Everyone was hollering and shouting happily as more and more people shared in the increasing excitement. Every window facing the street had someone hanging out of it, waving madly. After all the years of sullen, silent May Day marches, there was something magical about a large spontaneous demonstration. I kept looking around, soaking it all in, feeling that I was in a dream. Andrew S. Grove1

4.1 Introduction What happens when dissident mobilization accelerates so rapidly that it overwhelms the state’s coercive or control capacity? This turn of events is called bandwagon mobilization. We know almost nothing about its dynamics, other than the fact that it shows high levels of acceleration. Bandwagon mobilization occurs when the Rebel’s dilemma is solved and the State’s dilemma (mobilizing sufficient control forces) remains confounding (Lichbach, 1995, 1996). If we assume that in many states there is large-scale latent discontent, an event that shows that the state’s repressive capacity is constrained can accelerate mobilization. Such an event focuses the attentive population on the state’s coercive ability. When the minimum-winning mass of dissidents are able to shut down the repression or overwhelm the state’s coercive capacity, the door is open to bandwagon participants. The process of bandwagon mobilization usually begins when a typical mobilization grows larger and is either not repressed, or the state struggles to repress or contain it. When the population as a whole sees this process unfold, they are likely to pay closer heed, since often bandwagon mobilizations have the potential power to change a regime. Bandwagon mobilization is only instance when we expect the total mobilization in local areas to exceed five percent of the population (Lichbach, 1995).

1 Swimming Across: A Memoir. New York: Warner Books, 2001. Describing a scene in the 1956 Hungarian rising.

R.A. Francisco, Dynamics of Conflict, DOI 10.1007/978-0-387-75242-6 4, C Springer Science+Business Media, LLC 2009

51

52

4 Varied Dynamics of Bandwagon Mobilization

There is little true bandwagon theory in the field of protest and repression. Certainly Granovetter’s (Granovetter and Soong, 1983; Granovetter, 1978) work is relevant, because it concerns how mobilization occurs when dissidents have preference thresholds. Nonetheless, those who have attempted to model the dynamic bandwagon behavior are mostly economists. The work began with telephone demand (Artle and Avernous, 1973; Rohlfs, 1974). These pioneers essentially maximized social welfare functions to model the growth of clamor for telephones. They found that bandwagon demand is always logistic in form. This is encouraging, because we know that mobilization must almost always assume a logistic form. The economists work wholly within a non-conflict (except for price) environment. We at least have dissidents demanding a public good that the state rejects initially. The state, for its part, would seek to minimize a social welfare function representing the dissidents’ public good. Even Harvey (1996), who writes about bandwagon mobilization after women’s suffrage does not have to consider conflict. Also relevant are so-called iceberg models, e.g., Brookmeyer (2006). The idea here is that the state knows about the consistently active dissidents, but not how many citizens support them. Once mobilization accelerates logistically to a high slope, the state can do nothing because its coercive capacity is overwhelmed. Timar Kuran offers propagation models to consider the changes of public opinion Kuran (1995). Kuran considers bandwagon mobilization in a manner close to Granovetter (1978): “Each new person on this upward bandwagon induces additional people to climb on, until the entire population is on board (Kuran, 1995, 71).” Kuran considers only public opinion, not mobilization, which means we need to scale back the idea of the whole population. So models and theories are available to us in this realm, but they require evaluation for their appropriateness for the dynamics of bandwagon mobilization. The iceberg and propagation models are both univariate procedures. They work well to show the speed of mobilization to simulate Granovetter (1978) threshold model of preferences, but they are not interactive. Perhaps with bandwagon mobilization an interactive model is less important, but the inherent risk in even bandwagon mobilization is repression. As we will see, repression is most likely at the beginning or end of the bandwagon period, but not in the middle. Below we present the raw graphs of daily aggregated bandwagon mobilization. There is a significant amount of graphical variance in our sample cases. The model we use instead is the standard Lotka–Volterra (predator-prey) simultaneous equation model (see Chapter 1). This is a widely used model in ecology and has also been used for protest and repression Tsebelis and Sprague (1989). We do not expect the interactive parameters g and h to be statistically significant, because bandwagon mobilization only occurs in absence of risk. The standard model is depicted in the tables below for ease in deciphering the meaning of parameter estimates.

4.2 Cases The cases we investigate comprise: (1) the events of May, 1968 in France; (2) the rise of Solidarity and its initial repression in Poland (1980–1981); (3) the fall of

4.3 Results

53

1989 in the German Democratic Republic; (4) the fall of 1989 in Czechoslovakia; (5) the fall of 1989 in Bulgaria; and (6) the fall of 1989 to January 1991 in Romania. We range in time from 1968 until 1991, a relatively short interval, but the one for which we have data. We have no reason to believe that the dynamics of previous eras would differ substantially. While the east European cases were part of a system and obviously interrelated, neither the events of May, 1968 in France nor the rise of Solidarity in Poland was involved in the 1989 contagion. Everyone who has researched bandwagon mobilization has always found a logistic curve, whether in conflict or economics. That is one thing that we know must be true, for whatever its detailed shape, all mobilization has the distinctive, yet continuous, three-phase pattern.

4.3 Results The more one looks at French long time series of strike data, the more one is frustrated by an exasperating gap. That gap represents the 1968 extraordinary events of May (Mathieu, 2008). The French government records only blanks for this period, for they never truly measured the massive mobilization that took place. The bandwagon mobilization in France accelerated rapidly after university students began to strike. They demanded a more modern curriculum and more power for students and lower-ranking faculty. Young factory workers rejected the sloth of their trade unions and demanded reform. As is typical in bandwagon mobilization in democratic contexts, public opinion shifted quickly to support of the movement. On 18 May 1968, 55 percent of French citizens supported the student and young worker claims, 60 percent wanted a new form of society, and 50 percent approved the student and worker strikes. It is this sort of community support that helps to sustain bandwagon movements. Table 4.1 shows the statistical results from the Lotka–Volterra model. Note that the only parameters to achieve statistical significance represent dissident and state mobilization arising without interaction. The interaction parameters, especially g, which is at the Verge of statistical significance. We will see this pattern in most cases of bandwagon mobilization: there is almost no statistical indication of interaction between protesters and the state. In a way, this is part of the definition of bandwagons. They cannot arise under substantial risk. Also note that the eigenvalues Table 4.1 France’s Lotka–Volterra results, May 1968 Pt = a Pt−1 + g(Pt−1 × Rt−1 ) Rt = b Rt−1 − h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 103 ∗

Estimate ∗

0.8669 0.2783∗ −0.000000021527 0.000000737 λ1 = 0.8694


t-score

p(t)

12.66 2.05 1.97 1.35 λ2 = 0.0000004739

0.0001 0.0429 0.0512 0.1818

54


14000000 12000000 10000000 8000000 6000000 4000000 2000000 0 6

12

18

24

30

36

42

48

54

60

66

72

78

84

90

96 102

Fig. 4.1 French March–June 1968 mobilization

of the Jacobian matrix indicate complete stability. Both are positive and within the required bounds of −1 to 1. The mobilization figure begins in March when the incipient events began. The acceleration begins in May and continues through most of the month. The figure ends in mid-June when repression reappeared. Once DeGaulle restored gasoline to service stations at the end of May, most of the bandwagon mobilization halted. The logistic form is evident in the upswing of mobilization during May. The movement in France was sustained throughout the month of May by adaptation of the students and young workers. They made bargains with nearby farmers to buy food directly and therefore more cheaply, and they also made makeshift living accommodations on the left bank of the Seine. The movement ended when Charles DeGaulle, out of the country visiting Germany, was advised to put gasoline back into gas stations. He did, and that ended the movement, as French people could once again drive. There was some clearing out of dissident concentrations by police, but there were few injuries. Most of the rest of our cases are related. They occurred because in 1985, Mikhail Gorbachev became General Secretary of the Communist Party in the USSR. After Hungary and Poland had reformed their governments, Gorbachev did not act. In fact, in the early fall of 1989, the Warsaw Treaty Organization met and formally dissolved the Brezhnev Doctrine, which held that the WTO had the right to intervene militarily when any member faced a challenge to maintain socialism. The fact that the USSR had neither intervened in Hungary nor Poland and that Gorbachev scolded the GDR’s hard-line leader Erich Honecker on its 40th anniversary was a signal to dissidents in the entire bloc that the constraints of mobilization had lowered a great deal. Timur Kuran has noted that dictatorship creates false public preferences. It is

4.3 Results

55

only when the dictatorship weakens that it discovers how weak its political support has been Kuran (1991, 1995). Poland’s case is different from the rest of our East European sample. It models the rise of Solidarity, the only free trade union in the Soviet bloc that was approved on August 1, 1980 and led to rapidly accelerating mobilization. Within a month, Solidarity had Fashion Solidarity, Steel Solidarity, Rural Solidarity, Mining Solidarity, and countless other branches. It grew from the workers in the Gdansk shipyard on the first day to 20 million members after two months. By the end of the fall, the state began to signal opposition to Solidarity. The mobilization was stunning and surprised the regime in Warsaw as much as anyone else. In 1981 strikes increased, as did the demands of strikers for still more freedom. When the firefighters’ academy struck in Warsaw in December, the regime appeared powerless. In its stead came a communist military coup led by General Jaruselski. The military leaders imposed martial law, proscribed Solidarity and its human rights affiliate KOR on December 13, 1981. It took a full two weeks of severe repression to stop mobilization and dissident reaction. The parameter estimates in Table 4.2 are typical for bandwagon mobilization. Since bandwagons form unexpectedly, the state is rarely prepared to deal with them, even if it possessed sufficient coercive capacity to oppose dissent. It is not surprising then that a is the only parameter estimate that achieves statistical significance. After all, it represents the rise of dissent in the absence of repression, and that is what bandwagon mobilization represents. There is likely to be neither much repression nor interaction with state forces that generally watch in amazement as mobilization rises exponentially. The Polish case is unusual in the sense that initial mobilization accelerated exponentially and then at its conclusion was crushed with violence, but took a long time to subside completely. Nonetheless, we do not see these dynamics reflected in the parameter estimates. As a consequence, the interactive parameter estimates reflect nothing happening. Repression in these cases almost always appears at the beginning of the bandwagon briefly or at its conclusion (as in Poland) in an attempt to regain state control. For the most part, these front and end repressions do not affect parameters for the whole bandwagon mobilization. Figure 4.2 shows the shape of mobilization in Poland. Note that the logistic form begins early on, in August 1980 when Solidarity was approved as a free trade union Table 4.2 Poland’s Lotka–Volterra results, August 1, 1980–December 31, 1981 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter

Estimate

t-score

p(t)


0.7974∗ −0.18202 0.000158 0.0000003636 λ1 = 0.7973639

15.56 0.11 0.43 0.11 λ2 = 0.00003643141

0.0001 0.9149 0.6665 0.915

∗


56


800000 700000 600000 500000 400000 300000 200000 100000 0 8

16

24

32

40

48

56

64

72

80

88

96 104 112 120 128 136 144 152

Fig. 4.2 Solidarity mobilization in Poland, 1980 700

600

500

400

300

200

100

0 8

16

24

32

40

48

56

64

72

80

88

96 104 112 120 128 136 144 152

Fig. 4.3 Polish repression of solidarity, 1980

by the government. The spikes toward the middle and end reflect the attempt of the Polish government to stop Solidarity and preclude it. For this reason, it is helpful in the Polish case also to look at repression graphically (Fig. 4.3). The repression in 1980 was done early after the decision to try to halt a free trade union. It did not succeed, as Fig. 4.2 shows. Table 4.3 depicts the German Democratic Republic bandwagon mobilization. Astute readers might recognize that the eigenvalues are the same numbers as the

4.3 Results

57 Table 4.3 German Democratic Republic Lotka–Volterra results Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 122 ∗

Estimate ∗

0.2233 0.412474 0.0 0.000000792 λ1 = 0.2233

t-score

p(t)

2.5 1.51 0 0.89 λ2 = 0.000000792

0.0139 0.1341 0.9974 0.3771


a and h parameter estimates. This is because, as we saw in Chapters 2 and 3, the upper right cell of the Jacobian matrix (g) is zero, we have another linear algebra context of a lower triangle matrix, rendering the eigenvalues equal to the value of the main diagonal cells. Mostly, this simply confirms that there is little interaction between state forces and dissidents in bandwagon mobilizations. In the GDR it was the combined action of an orchestra conductor in Leipzig and the supposed last line of protection in the Socialist Unity Party, the ruling communist part of East Germany. There is no great puzzle about the East German case and the cessation of repression in October 1989. Hundreds of thousands massed in Leipzig, facing lines of police. Kurt Masur, director of the Leipzig Gewandhaus Orchestra, phoned the Politburo in East Berlin and convinced the authorities to withdraw forces. One of the most persuasive reasons Masur cited was the surrender of the Kampftruppen (battle troops) of the Socialist Unity Party (the communist party in the GDR). Battle troops were young men who served as the last bastion in the defense of socialism with weapons issued by the party. These young men wanted nothing to do with the repression of masses of demonstrating citizens. After Poland and the GDR the rest of the bandwagon mobilizations in Eastern Europe were part of a system of the Warsaw Treaty Organization. The USSR had kept these countries on a short leash and had demanded of all that they not waver from socialism. But when USSR General Secretary Gorbachev agreed to nonintervention, there was little that the dictators in each country could do about mass insurrection. In Czechoslovakia even the rumor that a student had been shot dead generated a massive general strike of secondary and university students. Dispersed in their individual homes, they could not be repressed effectively. Finally, the regime of General Secretary Husak was forced to accede to dissident demands, and that led to the release of prisoner Vaclav Havel and a reuniting of Alexander Dubcek and Havel in front of 300,000 Czechoslovakian citizens in November 1989. As Table 4.4 shows, we have once again only parameter a statistically significant. Bulgaria was the most loyal member of the Warsaw Pact and a fairly typical communist dictatorship. There was strong repression of the Turkish minority in Bulgaria and quite slavish following of the USSR’s orders of any day. In the beginning of 1989 mass numbers of ethnic Turks fled into Turkey, but by the late autumn the

58

4 Varied Dynamics of Bandwagon Mobilization Table 4.4 Czechoslovakia’s Lotka–Volterra results, September 1 – December 31, 1989 Pt = a Pt−1 − g(Pt−1 × Rt−1 ) Rt = b Rt−1 + h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 122 ∗

Estimate ∗

0.325 −0.1414 −0.000146 −0.00000045 λ1 = 0.3250635

t-score

p(t)

3.75 0.09 0.01 0.07 λ2 = −0.000064

0.0003 0.9307 0.9969 0.9428


Table 4.5 Bulgaria’s Lotka–Volterra results, October 15, 1989–December 31, 1989 Pt = a Pt−1 + g(Pt−1 × Rt−1 ) Rt = b Rt−1 − h(Pt−1 × Rt−1 ) Parameter

Estimate

t-score

p(t)


−0.09421 0.995774∗ 0.433663 0.000022 λ1 = −0.7059189

0.15 3.69 0.94 0.46 λ2 = −0.52343

0.8788 0.0004 0.3515 0.6443

∗


dictator Zhvivkov was ousted and replaced by Petar Mladenov. Opposition political parties were authorized in December 1989. This was a nonviolent transition that led to a long period of political conflict over the level of democracy. But we ended our data series after the Zhvivkov regime resigned. It was only then that Bulgaria began its torturous journey toward democratic government. The Bulgarian results in Table 4.5 measure a shift from parameter a to parameter b as the only statistically significant coefficient. As Fig. 4.6 shows, the mobilization in Bulgaria looks a good deal more episodic, but even so it is in logistic form. The major government concessions came in the wake of the spiked mobilization in center of the figure (see the number of 36 on x-axis). The Zhvivkov regime did not so readily surrender, and even after it did, the transition leaders were from the communist party, which maintained dissident mobilization through December 1989. In fact, if one looks closely at Fig. 4.6, it is apparent that another logistic bandwagon mobilization began (see after number 52 on x-axis). This one was smaller, which we would expect from existing theory, for example, Granovetter (1978 and DeNardo (1985). Both Granovetter and DeNardo claim that after a state makes policy concessions, mobilization should dampen, since many more people’s preference thresholds have been satisfied. Throughout, the Bulgarian mobilization, there were only two days where any repression occurred. This is important, because it is difficult to generate bandwagon when dissidents are falling from state repression.

4.3 Results

59

1000000

800000

600000

400000

200000

0 7

14

21

28

35

42

49

56

63

70

77

84

91

98

105 112 119

70

77

84

91

98 105 112 119

Fig. 4.4 GDR mobilization, September–December 1989 5000000

4000000

3000000

2000000

1000000

0 7

14

21

28

35

42

49

56

63

Fig. 4.5 Czechoslovak mobilization, October–December 1989

It is apparent in Fig. 4.7 that there was only one instance of repression in Bulgaria, and that came near the beginning of the bandwagon time series. So, there was no repression to impede the second and subsequent bandwagon mobilizations. Romania’s case is clearly different from the previous others. While it too remains in stable equilibrium, all its parameter estimates in Table 4.6 achieve statistical significance. Obviously, there was a great deal of interaction in the Romanian case. We also allowed a longer time series in this case because it took a long time for the

60


120000

100000

80000

60000

40000

20000

0 4

8

12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76

Fig. 4.6 Bulgarian mobilization, 1989 40

30

20

10

0 4

8

12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76

Fig. 4.7 Bulgarian repression, 1989

interaction to settle down. This was the only East European revolution that included violence between the state and the dissidents. This violence remained active after the surrender of the previous regime. In fact, this was a case of regicide, where the old dictator Ceaus¸escu and his spouse were executed after a revolutionary coup. Romania stayed outside of the Warsaw Pact and was not a client state of the USSR. Nonetheless, it was a communist dictatorship of a harsh character that tolerated no dissent. The dissidents had to rely on disaffected members of the regime

4.3 Results

61 Table 4.6 Romania’s Lotka–Volterra results Pt = a Pt−1 + g(Pt−1 × Rt−1 ) Rt = b Rt−1 − h(Pt−1 × Rt−1 ) Parameter a b g h Eigenvalues N = 416 ∗

Estimate ∗

0.1685 0.1827∗ 0.00135∗ 0.0000125∗ λ1 = 0.169951375

t-score

p(t)

3.5 2.99 5.58 8.33 λ2 = −0.001438875

0.0005 0.0029 0.0001 0.0001


before the bandwagon could begin. Perhaps it is because this is a longer time-series than our other cases, but more likely it is the violence that characterized Romania to make all parameters statistically significant. It is our only such case, and therefore represents an outlier. Romania was the only east European case we have where the regime excluded itself from the Soviet bloc. The other members of the bloc fell without (or without much) violence, but Romania looked nowhere else for policy guidance. The country was also poorer both in standards of living and education than our other cases. As in Bulgaria, it took longer for acceptable transition officials to succeed. Our time series runs from November 15, 1989 to January 19, 1991. As Fig. 4.7 shows, there were a series of bandwagon mobilizations in Romania.

Fig. 4.8 Romanian mobilization, 1989–1991

62


1400

1200

1000

800

600

400

200

0 21 42 63 84 105 126 147 168 189 210 231 252 273 294 315 336 357 378 399

Fig. 4.9 Romanian repression, 1989–1991

In order to understand the episodic mobilization in Romania, it helps to have a view of the repression, which was stronger in Romania than in any of our other cases. Figure 4.9 shows Romanian repression, heavy at the outset, but then recurring occasionally along the time series until late 1990, when it decreased significantly. Romania is a different case because repression never ceased fully during the bandwagon period. Figures 4.8 and 4.9 show how waves of mobilization rose most just after repression was high. This is the sort of backlash protest we see under normal conditions. In this, the most interactive case we have, there is much less evidence of logistical mobilization. More spikes of mobilization emerge toward the end of the period in the figures than at the outset, after repression virtually disappeared. Romania represents a much more interactive bandwagon than we see in any other case. This accounts for the results shown in Table 4.6, our single situation where all the model’s parameter estimates are statistically significant.

4.4 Discussion We know now that all of these bandwagon mobilizations are stable, but also that their dynamics vary a good deal. All arise logistically, but tend to end either suddenly or to peter out in extended fashion. Beginnings are short and long. Sustained mobilization is also both short and long. Can we infer anything from the statistical results and graphical evidence about the dynamics of bandwagon mobilization? Let us proceed systematically.

4.4 Discussion

63

First, what accounts for the acceleration rate of the bandwagon? The two most important factors are the level of repression and the probability of making a difference. If repression is present, mobilization will arise slowly; if it is hidden but available, the rate of mobilization should be higher; and if it is nonexistent or overwhelmed, the rate of mobilization should be highest. Typical of bandwagon mobilizations is repression at the outset or arising suddenly to end mobilization. If it occurs at the outset, then once it subsides or dissolves, especially if the state admits it was wrong to repress, bandwagons can start; after all, if the state admits it was wrong, it is a weak state and therefore vulnerable to dissent. Bandwagon dissidents act only when it appears that a regime transition might occur. If the repression is nonexistent until the end of the mobilization period, then bandwagons are logically possible as well. If mobilization swells beyond the state’s expectation and its coercive capacity, sometimes it attempts to increase its repression ability behind the scenes. If it is able to generate sufficient coercion ability, then it waits until the bandwagon begins to falter and moves to shut it down. This is what happened in France in 1968. Second, what halts a bandwagon? Obviously severe repression can do the trick, but bandwagons typically overwhelm state capacity. Endings of bandwagons have not been analyzed, or even much considered, but we have data and graphical information, so we can think about this problem systematically. There are several possibilities: (1) the regime grants sufficient concessions to satisfy most of the dissident population; under this circumstance, DeNardo and probably Granovetter would say that most would simply resume their normal lives (DeNardo 1985; Granovetter 1978); (2) the regime augments its coercion capacity and begins to apply it; bandwagon dissidents do not tolerate repression and would go home, which was what happened in Burma in 1989 (see Chapter 3); and (3) the state grants concessions in intervals with some repression; this is the situation that obtained in Bulgaria, the GDR and Romania; it leads to several bandwagons, but usually with smaller mobilization in each; if the regime gives up, mobilization rises; if it represses, bandwagons halt. Of the bandwagon ending possibilities, the first is represented by Czechoslovakia, Poland and France. In the Czechoslovak case, the Husák regime really did surrender to dissent. Once Vaclav Havel and Alexander Dubˇcek stood together in Prague in front of hundreds of thousands of citizens, mobilization stopped. The Polish Solidarity mobilization arose swiftly and then once all the divisions were created, street mobilization stopped. The French government gave some indication that it would implement reforms, but it did not. Bandwagons did not arise after it was clear that the DeGaulle regime had stonewalled the young dissidents. Bandwagons are difficult to restart after mobilization has collapsed totally. We have no instances of the second bandwagon ending possibility in our sample, mainly because after the fact these instances are not considered bandwagons. A good example is what happened in Burma after the 1988 election was annulled by the military junta (see Ferrara, 2003). There was massive mobilization at levels beyond even the Burmese government’s ability to repress. So the regime, saying it would accede to the dissident demand to free political prisoners, actually freed

64


rapists, armed robbers, and murderers. The population instantly left the streets and retreated to their homes, locking themselves in. Another example is represented by the quote at the outset of this chapter. The Hungarian revolution was definitely a bandwagon mobilization, but its repression by the Soviet Red army dominates our impressions today. As noted earlier, we have three instances of the third possibility: piecemeal concession to dissident demands. Here we expect that the bandwagons would subside and regenerate, albeit at lower levels, and that appears to happen in the figures (4.4, 4.6 and 4.7) representing these cases. Once the regime itself surrenders, bandwagon dissidents are more likely to be present than are standard protesters, since bandwagons are for the future and one’s place within it.

Bibliography Artle, Roland and Christian Avernous. 1973. “The Telephone System as a Public Good: Static and Dynamic Aspects.” Bell Journal of Economics and Management Science 4(1):89–100. Brookmeyer, Ron. 2006. Statistics: A Guide to the Unknown, chap. Modeling an Outbreak of Anthrax, pages 197–209. Belmont, CA: Thomson. DeNardo, James. 1985. Power In Numbers: The Political Strategy of Protest and Rebellion. Princeton, NJ: Princeton University Press. Ferrara, Federico. 2003. “Why Regimes Create Disorder: Hobbe’s Dilemma during a Rangoon Summer.” Journal of Conflict Resolution 47(3):302–325. Granovetter, Mark. 1978. “Threshold models of collective behavior.” American Journal of Sociology 83(6):1420–1443. Granovetter, Mark and Roland Soong. 1983. “Threshold Models of Diffusion and Collective Behavior.” Journal of Mathematical Sociology 9:165–179. Harvey, Anna L. 1996. “The Political Consequences of Suffrage Exclusion: Organizations, Institutions, and the Electoral Mobilization of Women.” Social Science History 20(1):97–132. Kuran, Timur. 1991. “Now Out of Never: The Element of Surprise in the East European Revolution of 1989.” World Politics 44(1):7–48. Kuran, Timur. 1995. Private Truths, Public Lies: The Social Consquences of Preference Falsfication. Cambrdige, MA: Harvard University Press. Lichbach, Mark I. 1995. The Rebel’s Dilemma. Ann Arbor: University of Michigan Press. Lichbach, Mark I. 1996. The Cooperator’s Dilemma. Ann Arbor: University of Michigan Press. Mathieu, Lilian. 2008. “The Spatial Dynamics of the May 1968 Demonstrations.” Mobilization 13(1):83–97. Rohlfs, Jeffrey. 1974. “A Theory of Interdependent Demand for a Communication Service.” Bell Journal of Economics and Management Science 5(1):16–37. Tsebelis, George and John Sprague. 1989. “Coercion and Revolution: Variations on a PredatorPrey Model.” Mathematical and Computer Modelling 12:547–559.

Chapter 5

Dynamics and Stability in Civil Wars

Both [North and South] deprecated war, but one of them would make war rather than let the nation survive, and the other would accept war rather than let it perish, and the war came. Abraham Lincoln

5.1 Introduction Interstate and colonial wars have now mostly vanished, leaving civil wars as the most common form of armed conflict. We know a great deal about the correlates of civil wars, but almost nothing about their underlying dynamics, such as their underlying structures or mechanisms. Are they stable? If these conflicts stay in equilibrium, what sort of equilibrium is it? If they are unstable, how? Is there an oscillation pattern or a divergent path toward infinite growth or total decay? How far and how fast does the war move away from a stable point or plane? These are the questions this chapter attempts to answer. The concept of stability is prior in importance to almost anything else, even the correlates of civil wars. It is surprising then that there is no theory predicting the dynamics of civil war. Certainly, some papers have attempted to model the dynamics of at least civil war duration, but not the mechanism of the war itself (DeRouen and Sobek (2004)). We work instead by necessity from a series of inductive conjectures. In order to understand the critical nature of stability, it is necessary to consider each civil war as a system. Both sides act against one another, mostly without direct communication. Nonetheless, they interact almost constantly on the battlefield. A civil war could remain stable (i.e., in equilibrium) if the conflict casualties during most of the war formed a white-noise random pattern or was relatively constant in casualties over time. That civil wars might periodically dip into stability during sieges is clear, but even some battlefield conditions might promote equilibrium. Great-great grandfather Allen Francisco, a lieutenant in the 126th regiment of the Illinois Volunteers, fought with the Union army during what he called “The War Between the States.” Late in the war General Grant sent my ancestor’s and two other Illinois regiments to Arkansas to clear out patches of remaining confederates. After a long, trench-based fire fight, Lt. Francisco realized that both sides had probably run out of ammunition. So he shouted across the trenches simply to ask the rebels if they were out. A yell came back, “yes!” The default military policy at the time was a bayonet charge, but this was late in the war, and the Union army had often R.A. Francisco, Dynamics of Conflict, DOI 10.1007/978-0-387-75242-6 5, C Springer Science+Business Media, LLC 2009

65

66

5 Dynamics and Stability in Civil Wars

eschewed pursuit of rebels lacking ammunition. So great-great grandpa Allen suggested that everyone simply turn around and go home. The rebels quickly agreed, and that battle ended peacefully. This suggests to me that rational soldiers and their lower-grade officers often resort to tactics that protect both sides from further harm. The individual soldier’s preferences should lead to stability in a time-series. This view is bolstered in many other accounts of civil wars. Younger (1968) for example, points out that in the Irish civil war: . . .there was among many Republicans a lack, not of conviction or courage, but of heart in the fight. They wanted to make their protest as urgently as they could and to keep on making it; they wanted to stop the Provisional Government from working the Treaty and building up an administration. They did not want to take life if they could avoid it, and neither did most of the Provisional Government troops, and so flights of bullets hurtled through the air harmlessly as migrating birds. The air above Ireland was crisscrossed with busy bullets with no particular object in view.

These soldier-based views increase the likelihood that civil wars remain in equilibrium. Soldiers are risk-averse; their casualties are generally far lower than are civilian casualties in a civil war, which are often nine times higher (see Mason, 2004). Nonetheless, there are also many reasons to predict that civil wars would be unstable. After all, unlike normal, everyday protest, civil wars last a long time. Their organization arises from battles, not more common street mobilization. Civil wars therefore should be cyclical, since no army fights a battle continuously. Armies and terrorists are paid fighters; hence, they are likely to fight when required and to be idle most of the remainder of the time. All these factors lead one to believe that unlike most protest, civil wars oscillate and remain unstable. This might seem like a purely academic concern. In part it certainly is, but since civil wars absorb the highest human costs, suffering is likely to be even greater in unstable wars. A conflict with expanding oscillation creates the condition for citizens approaching Olson’s (1993) roving bandits. Raids and battles are unpredictable. A quiet village in the morning becomes a cauldron of battle in the afternoon. Stability renders a war less harmful, both to troops and to civilians. High levels of instability would certainly help to explain why civil wars have emerged as the worst form of social conflict. The effects of civil wars extend far beyond the battlefield: the actual process conflict loses predictability both for commanders and citizens; civilian casualties always outnumber military losses; and civil wars generate the highest proportion of casualties of all types of conflict. Readers can find in E.L. Doctorow’s The March (2005) a gruesome, horror-filled account of an apparently unstable civil war. Most of the published literature on civil wars concerns the correlates of war: their causes, endings, and an array of factors that affects them. We know that there are civil wars spawned by greed versus those born of grievance (Collier and Hoeffler, 2001; Regan and Norton, 2005; Ron, 2005; Dunning, 2005; Fearon, 2005; Humphreys, 2005). We have learned that mountain ranges inhibit movement, yet increase mobilization; ethnic rivalries fuel conflict, but poverty dominates ethnic conflict as a spur to mobilizing rebellion (Fearon and Laitin, 2003). Both ethnic homogeneity and foreign military intervention raise the level of casualties in civil wars (Lacina, 2006). Veto players are critical in the duration of civil wars

5.2 The Cases

67

(Cunningham, 2006). Military victories often lead to genocide but reduce the probability of recurrence of war. Intervention and negotiated settlement preclude postwar genocide, but bring a much higher probability of recurrence (Licklider, 1995). We know much less about the fundamental mechanisms of these conflicts. Steven Garrison (2002, 2008) has shown that most civil wars begin as tactical co-evolution: both sides adapt to one another. Either one side defeats the other early on or both adapt at the same rate, allowing the conflict to continue and to escalate. It is this co-evolution and two-sided adaptation that bolsters the conjecture that civil wars are unstable. After all, if both sides augment their capability over time, we would expect casualties to increase sharply. This chapter investigates the most fundamental mechanism of civil wars: their equilibrium (see Bendor and Swistak, 1997; Adam, 2003). Do they oscillate or inflate exponentially as expected by many observers, or is conflict constrained such that stability is maintained, as foot soldiers would prefer? If civil wars vary in terms of stability, divergence, or oscillation, what accounts for such differences? Are geographically widespread wars more likely to oscillate, whereas spatially contained ones remain stable? Or do wars that end suddenly with a military victory show stability, while wars that have a long period of final battles (such as the US or the Russian civil war) do not? Were that to be the case, then an unfinished, still violent civil war (such as Colombia) is far more likely to be unstable.

5.2 The Cases We explore the question of stability and dynamics with seven dissimilar (in type, scope, space, and time) civil wars: (1) the Archidamian war (the first 10 years of the ancient Greek Peloponnesian war [431-421 BCE]); (2) the US civil war from 1861 to 1865; (3) the Russian civil war, including the Allied intervention and the Polish-Russian war, 1917–1920; (4) the Irish civil war, 1922–1923; (5) the Spanish civil war, with German and Italian intervention, 1936–1939; (6) the El Salvador civil war, 1979–1991; and (7) the ongoing Colombian civil war, for which we have data from 1988 to 2004. The principle of most-different systems (Przeworski and Teune, 1970) formed the basis for our case selection. While all these conflicts are civil wars, they occur at different times and in different regions. Moreover, each has a unique cause that sparked the fighting. While the population of the countries differs greatly, size is always uniquely relative to the conflict. In a small war, rebel casualties should be lower, but so also should state casualties. It is their relation to one another that matters, not absolute size. Geographic scope matters much more. The larger the physical size of a country, the more probability that there are multiple fronts. Multiple fronts complicate matters and generate higher likelihood of a more diffuse and oscillating conflict. In order to provide an orientation to the character of these varied conflicts, we present a short capsule description of each civil war. Sparta attacked the city–state of Athens in ancient Greece in 431 BCE. The first 10 years of this classic Peloponnesian civil war is called the Archidamian war after

68


the name of the original warring Spartan leader. It is one of the best documented ancient conflicts, which enabled us code it. We include it here because we are interested in civil war mechanisms. Were ancient conflicts, before firearms even existed, different in their dynamic processes than more modern wars? There were certainly greater periods with no events in this war, but also extremely intensive conflict with heavy casualties. The US civil war occurred in the nineteenth century because groups in the South sought to retain their position as slave-holding elites. Although abolitionists in the North actively opposed this state of affairs, South Carolina initiated the war at the federal Fort Sumpter in the spring and summer of 1861, after which a cascade of southern states left the Union. In the early years, the war might have gone either way, but eventually, the Union amassed more resources and military troops than could the Confederates. In April 1865, General Lee surrendered to General Grant, followed later by Johnston’s surrender to Sherman. The last battle of the war occurred in Texas in April 1865, and the war ended resolutely when the last skirmish took place in May in Missouri. A far different state of affairs served as backdrop in Russia. The Russian civil war was the result of the Bolsheviks’ refusal to accept a free-election result in the Constituent Assembly. As the Bolsheviks destroyed the assembly, conservative White forces, leftist Social Revolutionary groups (who had won a majority of seats) and ethnic Cossacks and Chechen groups organized to fight them, as did the Western allies of World War I, Japan, and the Poles later in the war. In the end, Leon Trotsky was able to mobilize and discipline the Bolshevik forces and garner sufficient resources to defeat every foe except the Poles, who signed a peace treaty that allowed the communists to shut down the only remaining White opposition in 1920. The Irish civil war emerged from a fight between factions of the Irish Republican Army. When the British offered Ireland “free state” status in the form of a treaty on December 6, 1921, most IRA members saw it as a victory over British imperialism. Many others saw the continued British presence in the six counties of Ulster as unacceptable. Tempers flared in the spring and summer of 1922 when a full-scale civil war began. It lasted barely a year, but the consequences of the war have dominated Irish politics to this day. The Spanish civil war also began as a response to an election victory: the 1936 triumph of communists, socialists, and anarchists. The military, with which the new government clashed, rose from a base in Morocco and soon controlled most of western and northern Spain. General Franco ascended to the apex of command in 1936 and brought in large-scale military aid from Nazi Germany and fascist Italy. Democratic European governments sympathetic to the Republicans in Madrid remained neutral during the civil war. This posture doomed the Republicans; the government was forced to yield all the gold in Spain’s treasury in order to obtain poor quality weapons from the USSR. The Republicans further reduced their probability of winning when the communists violently attacked both socialists and anarchists. Slowly at first, and then working systematically, the Nationalist fascists defeated the Galicians and Basques in the north, then the Catalonians within and around Barcelona, and finally Madrid itself in the spring of 1939.

5.3 The Data

69

It should be clear that the Archidamian, US, Russian, Irish, and Spanish civil wars are independent of each other and not linked in any way. If our Latin American cases relate to each other, it would be a remote connection and one that should not affect the underlying mechanism of the wars. El Salvador’s civil war began with military subversion of an election in 1979, escalated quickly to a civil conflict and ended with a negotiated agreement in early 1992. This was a war punctuated by death squads (comprised of police and military troops) murdering teachers, union officials, and peasants in the cities, as well as three horrific 1981 massacres in rural villages, while the leftist forces remained entrenched in caves and jungle sanctuaries. Everywhere, citizens experienced many terror tactics, especially kidnapping, bombing, and raids on prisons and police stations. In South America, the Colombian conflict represents our only case of a continuing civil war. This war began 40 years ago, progressed in fits and starts, then accelerated sharply in 1989, around the time the El Salvador battle in central America was winding down. The Colombian civil war’s great escalation began when leftist guerrillas undertook a set of market and contract solutions (Lichbach, 1995) of collective action with cocaine cartels. The guerrillas protected the cocaine business in exchange for large sums of money to pay leftist fighters and to procure weapons. This caused the United States to step up aid to the Colombian government. These augmented resources sustained both sides and considerably worsened the conflict.

5.3 The Data Civil war data required for stability tests were unavailable until recently. The standard forms of existing data (e.g., the Correlates of War) preclude the kinds of tests we need to conduct. Typically civil war data heretofore are aggregated yearly and in ordinal form. We need interval capture and casualty data aggregated daily in order to determine the dynamics of interaction on the smallest available time scale. The seven civil wars investigated in this chapter are the only ones I am aware of that offer or yield the requisite form of data. The Archidamian war was coded with six history books, of which Thucydides was clearly the most important. I used six reference books compiled by military historians to code the US civil war. Even though the Confederates burned all their military records at the end of the war, this case was the easiest to code. The Russian civil war required almost twenty volumes representing all facets of the conflict. The brief Irish civil war was covered well by two military histories and one anti-treaty general’s memoirs. The Spanish civil war took 10 histories as well as two sampled microfilmed newspapers, the Times of London and the New York Times. The data for our El Salvador case comes from Steve Garrison’s coding from Lexis-Nexis. Presented in raw form, these data require extensive work to clean and expand event series. Nonetheless, they have events sufficiently detailed to aggregate daily. All these data, including El Salvador, are linked in their original format on my data site (see http://web.ku.edu/ronfran/data/index.html). Our Colombian data were generously provided from a project led by Michael Spagat, a

70


professor of economics at the University of London. These dense, archival coded data run from 1988 to 2004 but include only aggregated injuries and deaths: there are no data on captured troops. We have sufficiently dense interval data on all of our civil wars for daily aggregation. This may seem surprising in such durable conflicts. Most people are unaware that in civil wars battles or skirmishes occur virtually every day in various locations and on different fronts. Such was the case in all of our civil conflicts. The data density allows us to use mainstream mathematical and statistical methods to analyze stability and equilibrium; in fact, because the data are so dense and not nearly as episodic as most history textbooks indicate, we have a higher probability of stability.

5.4 Models This chapter’s principal goal is to test the dynamics and stability of civil wars. The putative opposite of stability is divergence, oscillation or what Mandelbrot and Hudson (2004) call wild randomness. In their terms, we seek to find regularity, but “sometimes this regularity can be direct and awesome, at other times strange and wild” (Mandelbrot and Hudson, 2004, 30). The kind of regularity we seek is at best white-noise random. We could perform spectral analysis (see Hamilton, 1994) to test for white-noise randomness, but this is a univariate procedure. We are interested in interactive, system-level effects. For the complete mechanism, we must turn to a proper system of equations model. The model appropriate for civil wars should match the goals of each side. As we noted in Chapter 1, the best choice for a model is based on the principle of competitive exclusion (Luenberger, 1979, 328). The competing-species model is based on this principle and is therefore the structure of choice when investigating civil wars, since it reveals the mechanism of two opponents trying to defeat one another. Its underlying thesis is the principle of competitive exclusion, precisely what each side in a civil war seeks to accomplish. This model allows us to discover stability in its differential-equations sense, although we use difference equations because we cannot assume continuity. We compute the eigenvalues of the parameter estimates (i.e., partial derivatives) in the Jacobian matrix of the competing species model to evaluate the mechanism of the war system; this is an appropriate structure given our data (Murray, 1993). Because this is the only chapter that uses this model, we introduce it again. The competing species model is an ordinary differential-equations model:1

1 The standard competing species model in biology divides terms on the right side of the equation by K, a measure of the ecological carrying capacity. Since there is no significant limit on carrying capacity in civil wars, we assume K = 1.

5.5 Results

71

dR = a Rt − m(Rt St ) dt dS = bSt − n(St Rt ) dt

(5.1)

where, R is the rebel-induced state casualties; S is the state-induced rebel casualties; a is the state casualties in the absence of interaction; m is the the decline of rebel and state casualties with interaction; b is the rebel casualties in the absence of interaction; and n is the the decline of state and rebel casualties with interaction. Since we use interval data aggregated daily, we cannot assume continuity. Even in civil wars, fighting is not continuous. Soldiers, unlike protesters, do not return home at night, but neither do they (usually) fight both day and night. Therefore, we need to transform the above model into a system of difference equations. We are interested in how the resolute loss of fighters affects each force. It makes little sense to employ the total number of troops on each side as a variable in civil war analysis; the number rises in great battles and falls in skirmishes and quiescence. A better method is to consider the actual loss of soldiers. So, as noted earlier, we use the number of state forces that were lost for rebel power and conversely how many rebel casualties the state caused. The other side must be defeated in a civil war; therefore the objective is to eliminate as many opposing troops as possible. With these principles in mind, the tested difference equation model becomes (with R, S and the parameters retaining the meaning designated in the differential equation model above): Rt = aSt−1 − m(Rt−1 St−1 ) St = b Rt−1 − n(Rt−1 St−1 )

(5.2)

The parameter estimates emerging from each test form a Jacobian matrix. Our two exactly identified equations yield a square matrix that allows us to compute determinants as well as eigenvalues and eigenvectors. Our principal indicators of stability are eigenvalues that arise from the Jacobian matrices. The interpretation of these eigenvalues is similar in the worlds of difference and differential equations, but not identical. See Chapter 1 for details.

5.5 Results Do battles happen occasionally with great casualties, rendering the mechanism of civil wars unstable? Or can soldiers maintain sufficient control of the war so that it is able to maintain equilibrium? We test the entire systemic mechanism of civil wars. We take our conflicts chronologically, starting with the ancient Greece Archidamian war. Tables for each civil war indicate parameter estimates, t-values, and the probability that the t-score. The eigenvalues from the Jacobian matrix are also displayed in each table.

72

5 Dynamics and Stability in Civil Wars Table 5.1 Competing species test results for the Archidamian war (431-421 BCE) Rt = aSt−1 − m(Rt−1 ∗St−1 ) St = b Rt−1 − n(Rt−1 ∗St−1 ) Parameter a b m n Eigenvalues N = 3, 209 ∗

Estimate ∗

0.2459 0.3221∗ 0.00016 0.000212∗ λ1 = 0.2461096

t-ratio

p(t)

13.47 17.61 1.92 2.07 λ2 = 0.00000241681

0.0001 0.0001 0.0547 0.0387


Table 5.1 shows the ancient Archidamian war results. While this is a conflict almost one and one-half millennial old, it fit the competing species model well. Three parameter estimates are statistically significant and the fourth (m) is close to significance. The eigenvalues portray a conflict with much interaction, but a fully stable mechanism. Phase-plane plots of this war system would show motion only toward the origin. It is somewhat remarkable that a conflict that had no firearms, no mechanized weapon system and no motorized ships still fit the model parameters well. This indicates that like size, primitive weapons systems do not deter battles and killing. Since both sides used arrows as weapons, the soldiers were often almost as far apart as in more modern wars. Our next case is the US civil war. As we will see in the Spanish civil war, the rebels in the US civil war showed more initiative and activity in the absence of interaction than did the Union forces. Interaction accelerated Confederate casualties, even while it damped Union losses. The US data indicate a quiescent war in Virginia, an active one in the south and west during during the early years of the war, and almost complete Union dominance everywhere from 1864 onward. The system’s eigenvalues (Table 5.2) indicate stability. In fact the eigenvalues easily meet the stability standard for difference equations. The US civil war is a study in extraordinary mobilization (on the Union side) to elude quick defeat (by the rebel South). The Union lost a great many more troops

Table 5.2 Competing species test results for US civil war (1861–1868) Rt = aSt−1 − m(Rt−1 × St−1 ) St = b Rt−1 − n(Rt−1 × St−1 ) Parameter

Estimate

t-ratio

p(t)

a b m n Eigenvalues N = 1, 599

0.1559∗ 0.2629∗ 0.0000349∗ 0.0000086812∗ λ1 = 0.1559588

3.84 6.8 6.39 2.01 λ2 = −0.000050153

0.0001 0.0001 0.0001 0.0446

∗


5.5 Results

73

over the course of the war, but its huge resource advantage fueled its victory, albeit in stages. Victory in Virginia in April 1865 was followed by the surrender of Johnston to Sherman and then finally to triumph in the west. The course of the war, and even its uneven ending stayed within the bounds of stability. The Russian civil war is the most complex in our sample of large-scale civil conflicts. More armies faced the Bolsheviks than faced any other government. The Bolsheviks were challenged by several White armies, Social Revolutionaries, Chechens, Cossacks, Americans, British, French and Japanese intervenors as well as a fullscale series of battles with Poland. In 1920 the Poles fought the Bolsheviks to a draw, after which the Bolsheviks secured a peace treaty with Poland, and then set about defeating the remaining clutches of White armies around Petrograd and in Ukraine. They had already driven all the foreign Western troops out, and chased the Japanese from Siberia. Leon Trotsky marshaled his forces well after the first year of the war when outcomes were seriously in doubt. The Bolsheviks mobilized resources and troops to a higher level than any of their many opponents. When the war ended, it ended resolutely with White forces running for boats in Crimea. From late 1919 there was little doubt that the Bolsheviks would defeat the international challengers in Siberia as well as the array of White forces in the west. These victories, save against the Poles, were decisive. Stability makes sense in this type of varied, but abrupt resolution of a conflict, and that is what we found. In our statistical tests, all these players were arrayed against the hegemony of the Bolsheviks. This was a multi-front war fought over huge distances and for three years. These characteristics seem to imply chaos, but the Russian civil war is not unstable (Table 5.3). Instead it is stable, in other words its phase-plane is damped to the point that the war would move into stability at the origin. The Irish civil war is the shortest of our seven conflicts, with the active phase lasting only about one year. Nonetheless, it was a multi-front war with a great deal of violence, especially in the summer of 1922. By the fall of 1922 the anti-treaty rebels had been sufficiently vanquished that they henceforth fought a guerrilla campaign. We would expect such a war to be more likely stable than unstable, and that is what it is. The eigenvalues are bounded by −1 and 1; nonetheless, it is curious that only one of the model parameters in Table 5.4 is statistically significant, and that is a, the rebel-induced state casualties. But as in the US civil war, the state had overwhelming Table 5.3 Competing species test results for the Russian civil war Rt = aSt−1 − m(Rt−1 ∗St−1 ) St = b Rt−1 − n(Rt−1 ∗St−1 ) Parameter

Estimate

t-ratio

p(t)


0.00189 0.7535∗ 0.000425 −0.000881∗ λ1 = 0.01929293

0.05 8.03 0.51 4.5 λ2 = −0.01651193

0.9636 0.0001 0.6077 0.0001

∗


74

5 Dynamics and Stability in Civil Wars Table 5.4 Competing species test results for Ireland (1922–1923) Rt = aSt−1 − m(Rt−1 ∗St−1 ) St = b Rt−1 − n(Rt−1 ∗St−1 ) Parameter a b m n Eigenvalues N = 595 ∗

Estimate ∗

0.349 0.002534 −0.000938 −0.00348 λ1 = 0.3490067

t-ratio

p(t)

5.23 0.06 1.4 0.37 λ2 = −0.003487

0.0001 0.9509 0.1625 0.709


resources. The fact that the conflict’s intense phase was short, followed by low-level skirmishes and guerrilla tactics probably accounts for the worst model fit in our seven conflicts. The Irish civil war is the only one in our sample that did not last at least one thousand days. The Spanish civil war saw a great deal of foreign intervention, albeit only on the side of the fascists. General Franco was able to mobilize aid from his allies, while the leftist government could not force France and the United Kingdom out of their neutrality. The mechanism of the war is stable; the entire system remained in equilibrium. Parameter b in Table 5.5 represents rebel casualties in the absence of interaction. Interaction accelerated the casualties on the state side, rather than damping them as the competing species model specifies. But the system’s eigenvalues indicate the same sort of convergence we saw in the other cases. Both eigenvalues are positive and bounded by zero and one. El Salvador received some resources for the military government, but little for the rebels. The war was largely confined within the El Salvador northern “red zone” and had few major battles. It was a guerrilla-based conflict and it remained in equilibrium as it wound down toward a negotiated settlement. The El Salvador civil war is a paragon of equilibrium (Table 5.6). Its eigenvalues both indicate stability, indicating that the system stays in equilibrium. Why the stability? First, it was a long war. The data begin at its commencement and end near the negotiated cessation of active conflict. Second, while its casualties were

Table 5.5 Competing species test results for the Spanish Civil War (1936–1939) Rt = aSt−1 − m(Rt−1 ∗St−1 ) St = b Rt−1 − n(Rt−1 ∗St−1 ) Parameter

Estimate

t-ratio

p(t)


0.052905 0.6323∗ −0.00000992 0.000178∗ λ1 = 0.05302369

1.12 22.55 0.16 3.69 λ2 = 0.00005930696

0.2612 0.0001 0.8765 0.0002

∗


5.5 Results

75 Table 5.6 Competing species test results for the El Salvador Civil War Rt = aSt−1 − m(Rt−1 ∗St−1 ) St = b Rt−1 − n(Rt−1 ∗St−1 ) Parameter

Estimate

t-ratio

p(t)


0.004843 0.0804∗ 0.00048 0.000416∗ λ1 = 0.0092254315

0.32 5.15 1.44 4.1 λ2 = −0.00396315

0.7497 0.0001 0.1491 0.0001

∗


among the lowest of the seven cases we considered, much of the killing was done by state-sponsored death squads. Interspersed in the data are three brutal massacres that literally left no one to lash back at the state. Third, the war was fought on two fronts: (1) in major cities as a one-sided conflict at night (via death squads) and (2) in rural areas as an occasional two-sided conflict on the battlefield by day. The Colombian civil war is the longest-running conflict we consider. For over four decades the insurgents have battled Colombian state forces. Our data comprise the most violent and active era of the whole conflict, from 1988 to 2004. Three of the model parameter estimates are statistically significant in our analysis, showing a good deal of intense interaction. Only the n parameter is insignificant in Table 5.7, and that is an artifact of serial correlation correction. The fact that even Colombia, an unfinished, long-term civil war, is stable is an important finding. This is the case one might predict would move out of equilibrium and oscillate. But it stays in stable equilibrium. Why? It appears that the most basic reason is that all the diverse civil wars we tested are stable. That even Colombia is appears to be a victory of the foot soldier over commanders. It is well known that Colombian army conscripts sell rifles to insurgents and that the two sides often compromise with one another locally. What is our score at this point? We have seven civil wars in secure stability. We could not have more consistent empirical results. While different conflicts met different parameters with statistical significance, our principal question is answered: civil wars are stable; they remain in equilibrium for the whole of the conflict. The

Table 5.7 Competing species test results for colombia (1988–2004) Rt = aSt−1 − m(Rt−1 ∗St−1 ) St = b Rt−1 − n(Rt−1 ∗St−1 ) Parameter

Estimate

t-ratio

p(t)


0.2296∗ 0.2271∗ 0.001734∗ 0000181 λ1 = 0.23163848

15.66 16.04 2.63 0.21 λ2 = −0.00185748

0.0001 0.0001 0.0085 0.8307

∗


76

5 Dynamics and Stability in Civil Wars Table 5.8 Model fit by civil war (statistical significance) Rt = aSt−1 − m(Rt−1 ∗St−1 ) St = b Rt−1 − n(Rt−1 ∗St−1 ) Civil war

a

b

m

Archidamian US Russia Ireland Spain El Salvador Colombia

x x

x x x

x x

n x x

x

x

x x x

x x x

preference of the foot soldier seems to prevail. Let us examine the model fit by parameter for each of our conflicts (Table 5.8). Table 5.8 tells an interesting tale about the dynamics of civil wars. Parameter b is statistically significant in every case except Ireland, the shortest conflict. So the most likely occurrence in civil war mobilization is the state rising in the absence of interaction. The state is more active than the insurgents in Russia, Spain, and El Salvador in the absence of interactive conflict. The state and its enemies were equally likely to capture, injure, and kill each other in the Archidamian, US and Colombian civil wars (parameters a and b). Only in Ireland was the a parameter statistically significant without its state counterpart. The interactive terms also show differences. The parameter measuring the state-induced rebel casualties equation interaction (n) is statistically significant in the US, Russia, Spain, and El Salvador cases. Only in the Archidamian and Colombian civil wars are the rebel-induced state casualties equation parameter of interaction (m) confirmed statistically without its state-induced rebel casualties equation interaction parameter significant as well.

5.6 Discussion Both sides of a civil war seek to win; but barring victory, the worst outcome is a defeat or negotiated settlement that provides territory or special rights to the opposing side. Without an outright military victory, peace is a hazard. This is what impels a civil war to co-evolve. It causes both the state and its opponents to mobilize fighters, resources and better weapons. In large measure, a civil war is a different kind of conflict from those we generally study in protest and repression. Most modern civil wars do not begin in a fully military fashion as did the US and Spanish civil wars, but rather as persistent protest against a state. A state that is unable to deter, co-opt or defeat its dissent, incurs a higher probability of a civil war. It is a great surprise to find that all of our seven civil conflicts are stable and in complete equilibrium. Almost any veteran can attest that aside from moments of high intensity, most of the time war is intensely boring. Civil wars offer less down time, less predictability, and usually more intense battles than do other kinds of fighting. These factors make civil wars the worst form of conflict human beings

5.6 Discussion

77

have developed. Colombia’s civil war wages on in its fifth decade. As an unfinished war, it is surprising that it operates like any other case in our diverse sample. The US civil war ended unevenly, with battles raging even a month after Grant accepted Lee’s surrender at Appomattox. Yet despite this messy conclusion, US civil war remains in equilibrium. The Russian civil war was a wide-ranging contest of many foes against resolute Bolsheviks. Its theater ranged over an incredibly wide arena, from Poland and Petrograd to Crimea and Siberia and the Pacific ocean. Denikin’s White army and the foreign intervenors were defeated in 1919, Poland was subdued sufficiently to agree to a truce in mid-1920, and the Wrangel White army collapsed against Bolshevik pressure in the autumn of 1920. Weak and lacking resources, the opponents succumbed. Mostly fighting only on one or two fronts at time the Bolsheviks claimed a decisive military victory. Resources matter. The side that can mobilize and assemble the greatest resources has a good chance to win a civil war. But resources do not affect stability as much as they might appear. Conflicts in our sample that mobilized the most resources are most likely, in relative order, the United States, Russia, Archidamian, Colombia, Spain, El Salvador, and Ireland. Only one side successfully mobilized resources in Spain. El Salvador and Ireland were clearly the “cheapest” wars in terms of resources used. A naval blockade was imposed along the entire Atlantic and Gulf coasts during the US civil war. And although the British and Confederates were able to breach the blockade, in general it held and deprived the south of urgently needed military goods and food. It is fitting that Robert E. Lee surrendered during the siege of Petersburg, Virginia when he could not fight his way outside Grant’s forces and the crippling naval blockade. The Bolsheviks, after a slow start, ramped up their resources and fighter mobilization quickly. Trotsky ran the Bolshevik army in the field in an armored train, which allowed him to move easily from one theater of operation to another. The Bolsheviks could also meet payrolls more easily than could their opponents. One should not minimize the difficulty of finding soldiers who will fight with no compensation. The mobilization in ancient Greece was relatively easy and surprisingly mobile. Since the war was fought in a relatively small area, all surrounded by sea, it was not difficult to move large numbers of troops quickly to a new fighting venue. The present Colombian civil war is the model for resources: (1) the drug cartels have paid insurgents to protect them; (2) the United States sends the arms needed by the state, but troops too readily sell rifles and ammunition to the insurgents; and (3) the insurgents kidnap wealthy citizens and hold them for ransom. It was the insurgent military that found necessary resources in its fascist allies, even while the government foundered on the neutrality of the UK and France. El Salvador and Ireland used the fewest resources. To be sure, the military regime in San Salvador secured large-scale help from the United States, some of which was confiscated by the rebels. But El Salvador was a classic guerrilla war against an established military leadership that sought to use air power and artillery against a disbursed enemy and mostly succeeded in killing civilians. Ireland’s difficult shedding of British imperialism led to violent confrontation of two Irish nationalist sides. The pro-treaty, provisional government side managed to secure the resources of the

78


departing British military, which gave it an immense advantage from the outset of the war. Perhaps one of the reasons we have found consistent and total stability is that we do not have civilian casualties, which are uniformly higher than military casualties in civil wars. Consider General Sherman’s comments in the US civil war: “[we will] make old and young, rich and poor, feel the hard hand of war, as well as their organized armies.” (Weigley, 1973, 149). Civilians certainly suffer more harm from civil wars than do soldiers (Mason, 2004). If we could obtain accurate and valid data on civilian casualties we could undertake stability tests, but these are data collected by no one, in large measure because the harms are diffuse and often hidden from officials. I am far more inclined to view stability as a consequence of almost constant conflict. History texts emphasize the major battles in several of the civil wars in our sample. Only in coding these conflicts does one realize how important daily skirmishes are in the portrait of the total war. Often there are small conflicts apart from major battles even as these battles rage. These violent interactions emerge consistently and create a kind of asymptotic global minimum to the data time series. As a consequence, the time series appears as a rather stable series with a few spikes that account for major battles. The surprising finding from this study is how much stability occurs in civil wars. That all of our cases would be stable is unexpected. All of this is consistent with the basic results of stability tests: most human macro-level conflicts remain in equilibrium (Francisco, 1995, 1996).

Bibliography Adam, John A. 2003. Mathematics in Nature. Princeton, NJ: Princeton University Press. Bendor, Jonathon and Piotr Swistak. 1997. “The Evolutionary Stability of Cooperation.” American Political Science Review 91(2):290–307. Collier, Paul and Anke Hoeffler. 2001. “Greed and Grievance in Civil War.” World Bank. Cunningham, David E. 2006. “Veto Players and Civil War Duration.” American Journal of Political Science 50(4):875–892. DeRouen, Karl R. Jr. and David Sobek. 2004. “The Dynamics of Civil War Duration and Outcome.” Journal of Peace Research 41(3):303–320. Dunning, Thad. 2005. “Resource Dependence, Economic Performance and Political Stability.” Journal of Conflict Resolution 49(4):451–482. Fearon, James D. and David D. Laitin. 2003. “Ethnicity, Insurgency, and Civil War.” American Political Science Review 97(1):75–90. Fearon, James D. 2005. “Primary Commodities and Civil War.” Journal of Conflict Resolution 49(4):483–507. Francisco, Ronald A. 1995. “Coercion and Protest in Three Coercive States.” Journal of Conflict Resolution 39(2):263–282. Francisco, Ronald A. 1996. “Coercion and Protest: An Empirical Test in Two Democratic States.” American Journal of Political Science 40(4):1179–1204. Garrison, Steven. 2002. The Long and Terrible Road: The Evolution of Political Protest to Civil War. Ph.D. thesis, University of Kansas.

Bibliography

79

Garrison, Steven. 2008. “The Road to Civil War: An Interactive Theory of Political Violence.” Defence and Peace Economics . Hamilton, James D. 1994. Time-Series Analysis. Princeton, NJ: Princeton University Press. Humphreys, Macartan. 2005. “Natural Resource Conflict and Conflict Resolution.” Journal of Conflict Resolution 49(4):508–537. Lacina, Bethany. 2006. “Explaining the Severity of Civil Wars.” Journal of Conflict Resolution 50(2):276–289. Lichbach, Mark I. 1995. The Rebel’s Dilemma. Ann Arbor: University of Michigan Press. Licklider, Roy A. 1995. “The Consequences of Negotiated Settlement in Civil Wars, 1945-1993.” American Political Science Review 89(3):681–690. Luenberger, David G. 1979. Introduction to Dynamic Systems. New York: John Wiley & Sons. Mandelbrot, Benoit and Richard L. Hudson. 2004. The (mis)Behavior of Markets. New York: Basic Books. Mason, T. David. 2004. Caught in the Crossfire: Revolution, Repression and the Rational Peasant. Lanham, MD: Rowman and Littlefield. Murray, J.D. 1993. Mathematical Biology. New York: Springer Verlag. Olson, Mancur. 1993. “Dictatorship, Democracy and Development.” American Political Science Review 87(3):567–576. Przeworski, Adam and Henry Teune. 1970. The Logic of Comparative Social Inquiry. New York: John Wiley & Sons. Regan, Partick M. and Daniel Norton. 2005. “Greed, Grievance, and Mobilization in Civil Wars.” Journal of Conflict Resolution 49(3):319–336. Ron, James. 2005. “Paradigm in Distress? Primary Commodities and Civil War.” Journal of Conflict Resolution 49(4):443–450. Weigley, Russell F. 1973. The American Way of War: A History of the US Military Strategy and Policy. Bloomington: Indiana University Press. Younger, Carton. 1968. Ireland’s Civil War. London: Frederick Muller.

Chapter 6

Conclusion: Stability in Conflict

Popular revolutionary violence was not some sort of boiling subterranean lava that finally forced its way onto the surface . . . and then proceeded to scald all those who stepped in its way. . . . (I)t is better to think of the revolutionary elite as rash geologists, themselves gouging open great holes in the crust of public discourse and then feeding angry matter through the pipes of their rhetoric out into the open. Simon Schama, Citizens: A Chronicle of the French Revolution

The surprising result of our investigations is almost universal stability. In a wide variety of conflictual contexts, stability was the norm. The single exception was terror in Northern Ireland, which oscillated, but in so damped a fashion that it too returns to the origin and a bed of stability. We did not expect this at the outset of the project. It certainly seemed that civil war would oscillate, moving from big battle to big battle with little in between. But the data show that there is much more action between large-scale battles than we suspected. There are skirmishes almost every day between scouting parties and front lines and with individual regiments. All these conflicts smooth the data between the really large fighting and their aftermaths. Civil war is the worst form of human conflict, but it is stable. Neither did we think that bandwagon mobilization would be stable, since it arises exponentially and generally collapses exponentially as well. Yet as we saw in Chapter 4, bandwagon mobilizations with widely varying graphs are nonetheless in equilibrium. In this chapter we summarize the most important findings in the project. We try as well to find important commonalities beyond the stability that lies at the core of our conflicts.

6.1 Stability is the Norm Why does stability stand out so clearly as a universal concept in conflict? The findings tell us that stability exists, but nothing about the reasons that caused it. Foremost among the likely reasons for stability is that human conflicts are created by and conducted by humans. As Hardin (1995) noted “How many people want to be ardently political all the time?” Certainly no one, or least few wish to be in conflict all the time either. Normal life is a fierce competitor to conflict. Most people R.A. Francisco, Dynamics of Conflict, DOI 10.1007/978-0-387-75242-6 6, C Springer Science+Business Media, LLC 2009

81

82

6 Conclusion

have jobs, families, and other responsibilities. After all, civil society alone has many requirements and conflict is certainly not one of them. One of the reasons we refuse to assume continuity is that most protesters go home at night and sleep. They do not continuously protest. In another project, we found that protesters were even loathe to give up their weekends to demonstrate. The most likely demonstration times over most of North America and Europe are during lunch from Monday to Friday. So when a rally begins, it is likely to end as soon as its point is made. A few radicals may remain, but the vast majority of dissidents find other activities more enjoyable and apparently compelling. We have discovered in other contexts that conflict is not only stable, but is usually white-noise random. Mark Lichbach has noted that this occurs naturally; as one conflict starts and proceeds, another does two days later. The first one ends and another begins three days after the second one. This is precisely how a white-noise random and stable conflict situation would be organized. It does happen that way: a hunger strike ends two days after a corporate lockout occurs; a boycott starts three days after the lockout begins. We have documented these patterns in the data that underlie this project and in a wide variety of conflict. Most instances of conflict, then, remain in equilibrium and exhibit a white-noise randomness pattern. Conflict is not a cauldron; it is a means to an end in political disagreement.

6.2 Varieties of Repression in Democracies and Dictatorships We have seen, especially in Chapters 2 and 3, that not all democracies are equally repressive on the street and that even dictatorships vary a great deal in the amount of street repression they create. While one might believe that unitary democratic countries are more likely to repress, federal systems actually have more repression. This is in part because our data represent the whole of each country. We discover protest even in rural or outlying areas. Austria, for example, showed three parameters that were statistically significant, as did Belgium, where linguistic differences and conflict have bifurcated the country’s political structure. Both these countries allow a good deal of local control. Denmark, in contrast, has only parameter estimates in the protest equation statistically significant. In Germany, a federal country, we found only the noninteractive parameters statistically significant. Greece, a less-developed country in our sample, had parameter estimates near to Belgium’s. But Iceland and Ireland showed little repression or even interaction with dissidents. Iceland was one of our early examples of linear algebra lower triangle context, which means that critical parameter estimates are near zero. Italy and Luxembourg each had but one parameter significant, while the Netherlands showed neither interaction nor statistical significance. Norway had one noninteractive parameter statistically significant, while the Iberian states shared both parameters in the protest equation statistically significant. Sweden and Switzerland, meanwhile, matched the Norwegian results closely. In none of these countries do we find much repression or interaction. In contrast, we have France and the United Kingdom with all parameters in both the protest and repression equations statistically significant. As we noted in

6.2 Varieties of Repression in Democracies and Dictatorships

83

Chapter 2, the fact that each country has only one veto player explains a good deal of the interaction on the street. If the legislature is a silly place to lobby, and one cannot talk to the leader of the country, the street is one of the only options available for political action. This is one possible explanation. If it is large countries that have these properties, then why are they absent in Germany? There is little else in common between France and UK; they have totally different political systems and UK does not even have a constitution, although both countries maintain unitary systems. So the absence of veto players is the best explanation from my perspective. If democracies are not uniform in their levels of repression, then what about dictatorships? During and after Enver Hoxha’s rule, Albania was intensely repressive, claiming some of the largest t-values I have ever seen. Meanwhile Belarus (with only weekly data) and Bulgaria were not noticeably more repressive on the street than are many West European democracies. Burma, in 1988, however, matched the brutality of Albania. Czechoslovakia’s results show that both repression equation parameters are statistically significant, indicating that the regime was active against dissidents on the street. The German Democratic Republic was much more subtle; except for a large t-score, one might mistake its results for a West European democracy. Hungary looks, if anything, better than a West European democracy. Clearly, the period after the rising in 1956 left Hungary a peaceful and politically careful country. Poland battled against Solidarity almost the entire decade of the 1980s; its three significant parameter estimates show activity on the street, but not nearly as active and violent as in Albania and Burma. Romania is a repressive regime; all the parameter estimates are statistically significant. Yet, its t-values are far higher than those in France and UK, indicating much more frequent repressive action on the street. Nor were democratic transitions equally peaceful and democratic. Romania before 1995 did not complete a total regime transition and continued to be repressive, even as a non-communist state. Albania continued to repress during its transition, although its t-values decreased markedly. Bulgaria’s results continued to make it look like a West European state. But the real drama was reserved for the other transition countries in the sample. We were amazed by the astonishing reduction of repression in Czechoslovakia, the Czech Republic, Hungary, Poland, and Slovakia. All these are examples of the linear algebra lower-triangle phenomenon. They show some of the most remarkable changes in regime behavior we have witnessed anywhere. To the great relief of the populations, pacification followed the largely non-violent revolutions in east central Europe. Bandwagon mobilizations varied greatly. The only finding they had in common was stability. From France in 1968 to the 1989 east central Europe revolutions all had widely varying graphs, but all were found to be stable. This would be less surprising if our sample included only east central Europe revolutionary mobilizations. But we also included the quite separate and major French May 1968 mobilization. And Poland’s 1980–1981 Solidarity mobilizations was not linked directly to 1989. They too are stable. The most variance we introduced into samples was in our investigations of civil wars. From ancient Greece to current Colombia we had major (Archidamian, US,

84

6 Conclusion

Russian, Spanish, Colombian) civil wars as well as smaller (Ireland, El Salvador) but in all cases, covering (quite unevenly) three millennia, we found systemic equilibrium.

6.3 Convergence in Estimations Whenever one attempts to estimate time-series data, the technical problem of convergence to efficient estimates (when the gradient is zero) must be addressed (see Greene, 2003). For the most part we experienced no difficulty in gaining convergence, even in the shortest time-series of all, the bandwagon mobilizations. We encountered convergence difficulties only two or three times, usually in long timeseries. We then attempted other estimation techniques, even ordinary least squares, to determine whether the parameter estimates we found were close to correct. We confronted no serious problems in doing this, which was surprising, given the number of time-series we estimated.

6.4 Correction of Time-Series Pathologies Time-series estimation always faces the dual hazards of serial correlation of variable error terms and heteroscedicity. The former of these is by far the most serious pathology in time-series estimation, but we tested for and corrected for heteroscedicity as well (see Gelman and Hill, 2007 or Greene, 2003). In fact, all the parameter estimates appearing in this volume are corrected for both of these pathologies. That we used seemingly unrelated regression (SUR) was helpful, because then we could compare SUR parameter estimates with corrected parameter estimates determined equation by equation. For the most part, the SUR parameter estimates were close to the corrected ones. Nonetheless, we entered only corrected parameter estimates and used these in the eigenvalue stability tests.

6.5 When Repression is Absent or Rare Beyond nearly universal stability, the real surprise from our tests is the number of cases with almost no repression. For these cases we invoked the linear algebra lower or upper triangle concept. Remarkably, these occurred infrequently in longdemocratic countries, and mainly in the democratic transitions from communist dictatorships. The new rulers in these countries, at least in the north part of east central Europe, were judiciously conservative in arrest and police violence. In the political mayhem that accompanies transition, it is a tribute to these countries that they were even less repressive than the Scandinavians. Would that the whole world could act this way!

Bibliography

85

6.6 What Have We Learned? Conflict is stable, but what else emerged from the investigation of many diverse conflicts? Certainly, the lack of repression in surprising contexts and the necessity to invoke the lower triangle concept was something we never anticipated. Also, in the course of our investigations we were at last able to test a long-standing conjecture by Mark Lichbach that inconsistent repression accelerates protest (see Lichbach, 1987). We found in German leftist protest and repression that inconsistent repression did in fact accelerate protest. This is a graphical demonstration that should be tested differently and in more than one case. Yet it is empirical confirmation of a discovery from a formal model. Now that data in interval form are available, much more of this sort of testing can be conducted on many conjectures drawn from formal theory. The fact that we had interval data drawn directly from multiple sources enabled us to invoke levels of estimation and mathematics that are only available to these forms of data. Such data are difficult and expensive to code, but the wealth of applications they open justifies much of the time and expense of coding. This volume was restricted to Europe, Burma, and one state in the United States for the simple reason that daily interval data exist only in these settings. If we had more such data on developing countries such as India, China, Brazil, and African countries, we might see instability and different outcomes. Certainly, the experience we have had in a wide variety of contexts that included revolution, though, indicates that we should expect stability even in these actively changing environments.

Bibliography Gelman, Andrew and Jennifer Hill. 2007. Data Analysis Using Regression and Multilevel/Hierarchical Models. New York: Cambridge University Press. Greene, William H. 2003. Econometric Analysis. Upper Saddle River, NJ: Prentice Hall. Hardin, Russell. 1995. One for All: The Logic of Group Conflict. Princeton, JN: Princeton University Press. Lichbach, Mark I. 1987. “Deterrence or Escalation? The Puzzle of Aggregate Studies of Repression and Dissent.” Journal of Conflict Resolution 31:266–297.

Index

A Acheson, David, 5 Adam, John A., 67 Albania, 34, 36, 43–44, 48–49, 83 Hoxha, Enver, 36, 44 Sigurimi, 36 Almond, Gabriel, 34 America, 15, 69, 82 Archidamian civil war, 67–69, 71–72, 76, 83, 84 Athens, 67–68 Sparta, 67–68 Arendt, Hannah, 13 Artle, Roland, 52 Austria, 9, 14, 16, 24, 27, 29, 31, 34, 41, 46, 82 Autocracy, 35 Avernous, Christian, 52 B Bandwagon mobilization, 3, 51–64 Bauer, Otto, 34 Belarus, 33, 34, 37, 43, 83 Belgium, 14, 16, 17, 20, 31, 82 Flemish, 17 Walloons, 17 Bendor, Jonathon, 67 Boyce, William E., 5 Brookmeyer, Ron, 52 Bruesch–Godfrey test, 8 Bruesch–Pagan test, 8 Bulgaria, 34, 37, 38, 43, 44, 53, 57, 59, 60, 61, 63, 83 Burma, 33, 34, 35, 38, 39, 43, 63, 83, 85 C Carey, Sabine C., 2 Ceaus¸escu, Nicolae, 43, 60 Chicago, 26, 27, 29, 30 Coevolution, 67

Coleman, Stephen, 2 Collier, Paul, 66 Colombia, 67, 75, 76, 77, 83 Colombian civil war, 67, 69, 75, 76, 77, 84 Competing species model, 6–7, 70–71, 72, 74 Convergence, 5, 74, 84 Correlates of War data, 69 Cunningham, David E., 67 Czechoslovakia, 38–39, 40, 44, 45, 46, 47, 63, 83 Charter 77, 38–39, 40, 45 Helsinki accords, 38–39 Jazz, 39 Klaus, Vaclav, 44 Plastic People of the Universe, 38–39 Prague, 38–39, 40, 45, 63 Czech Republic, 34, 44, 45, 46, 47, 83 D Data, 1–10, 13, 14–15, 18, 21, 23, 24, 25, 26, 29, 35, 37, 39, 42, 43, 44, 45, 46, 53, 69–70, 71, 72, 74–75, 81, 82, 83, 84, 85 interval, 2, 4, 14, 70, 71, 85 ordinal, 2, 13 Democracy or democratic, 3, 6, 9, 13–31, 33–49, 53, 57, 8 68, 82–84 DeNardo, James, 3, 13, 14, 33–34, 35, 58, 63 Denmark, 14, 17, 31, 82 DeRouen, Karl R. Jr, 65 Dictatorship, 3, 9, 14, 20, 33–49, 54–55, 57, 60, 82–84 DiPrima, Richard C., 5 Divergence, 1, 2, 4, 67, 70 Doctorow, E.L., 66 Doering, Herbert, 9, 28 Dubcek, Alexander, 57 Dunning, Thad, 66 Durbin h test, 8 Dynamics, 1, 7, 10, 13, 33–49, 51–63, 65–78

87

88 E Elaydi, Saber N., 4, 5, 8, 31, 48 Electoral law, 47 El Salvador civil war, 67, 74, 75 death squads, 69, 75 red zone, 74 Equilibrium, 2, 3, 4–5, 31, 48, 59, 65, 66, 67, 70, 71, 74, 75, 76, 78, 81, 82, 84 Exponential, 2, 5, 8, 55, 67, 81 F Fearon, James D., 66 Federalism, 9, 18, 24 Ferrara, Federico, 38, 63 Fisher, R.A., 1 France, 9, 14, 17, 26, 27, 28, 29, 30, 31, 52–53, 54, 63, 74, 82, 83 Debre, Michel, 17, 28 DeGaulle, Charles, 17, 28, 54, 63 May 1968, 52–53, 83 Francisco, Ronald A., 2, 23, 29, 35, 48, 65, 78 G Garrison, Steven, 67 Gelman, Andrew, 84 German Democratic Republic (GDR), 40, 43, 54, 57, 59, 63 Bohley, Baerbel, 40 Honecker, Erich, 54 Kampftruppen, 57 Neues Forum (New Forum), 40 Pflugbeil, Sebastian, 40 Socialist Unity Party, 57 Stasi, 40 Germany or Federal Republic of Germany, 9, 14, 18, 24, 27, 29, 31, 34, 40, 57, 68, 82, 83 Brandt, Willy, 35 ecological protesters, 18 leftists, 18, 19, 20, 68, 69, 74, 85 Gill, Anton, 34 Goldberg, Jack, 5 Goldberg, Samuel, 4, 8, 31, 48 Gorbachev, Mikhail, 36, 40, 41, 42, 44, 45, 54, 57 Granovetter, Mark, 52, 58, 63 Grant, General, 65, 68, 76, 77 Greece, 14, 20, 21, 28, 29, 31, 67, 71, 77, 82, 83 Greene, William H., 4, 7, 84 Grove, Andrew S., 51

Index H Hamilton, James D., 70 Hardin, Russell, 81 Harvey, Anna L., 52 Havel, Vaclav, 39, 44, 45, 48, 57, 63 Hill, Jennifer, 84 Hoeffler, Anke, 66 Hoffman, Peter, 34 Hudson, Richard L, 70 Humphreys, Macartan, 66 Hungary, 34, 40, 41, 42, 46, 49, 54, 83 1956 rising, 40–41 Democratic Forum, 41 Husak, Gustav, 57 I Iceland, 14, 21, 23, 29, 31, 82 Illinois, 9, 13, 14, 15, 16, 24, 26, 27–30, 31, 65 Chicago, 26, 27, 29, 30, 31 Downstate, 26, 30 Institutional theories, 9 International Monetary Fund, 41 Ireland, 14, 21, 26, 28, 29, 30, 31, 35, 66, 68, 74, 76, 77, 81, 82, 84 Irish civil war, 66, 67, 68, 69, 73, 74 anti-treaty, 69, 73 pro-treaty, 77 Italy, 14, 21, 31, 36, 68, 82 Red Brigade terror, 21 J Jacobian matrices, 71 Jakeˇs, Milo, 38, 45 Johnson, Paul, 29 K Khrushchev, Nikita, 41 Kornhauser, William, 34 Kuran, Timur, 39, 52, 54 L Lacina, Bethany, 66 Laitin, David D., 66 Lakatos, Imre, 3 Lee, Robert E., 68, 76, 77 Lenin, Vladimir, 35 Licklider, Roy, 67 Lincoln, Abraham, 65 Logistic, 52, 53, 54, 55, 58, 62 Lotka–Volterra (or predator–prey) model, 4, 5–6, 15, 16, 17, 18, 20–28, 36–47, 52, 53, 55, 57, 58, 61 Lower triangle, 21, 22, 45, 46, 49, 57, 82, 83, 85

Index Luenberger, David G., 3, 70 Lukashenko, Aleksander, 37 Luxembourg, 14, 22, 29, 31, 82 M McCelvey, Richard, 28 Maddala, G.S., 5, 7 Mandelbrot, Benoit, 70 Mason, T. David, 66, 78 Masur, Kurt, 57 Mathieu, Lilian, 53 Maung, Saw (Burmese General), 38 Meˇciar, Vladimir, 47 Merkin, David R., 4, 5, 48 Mladenov, Petar, 38, 58 Mobilization, 1, 3, 9, 14, 16, 18, 33, 34–36, 39, 40, 43, 45, 51–63, 66, 72, 76, 77, 81, 83, 84 Morton, Rebecca B, 2, 4, 5, 6 Murray, J. D., 4, 6, 70 Myanmar, see Burma Myers, Daniel J., 15 N Nagy, Imre, 41 Netherlands, 14, 18, 23, 31, 82 Nordberg, Marc, 37 Northern Ireland, 14–15, 21, 26, 29–30, 31, 81 PIRA, 25 UFF, 25 UVF, 25 Norton, Daniel, 66 Norway, 15, 23, 29, 31, 82 O Oliver, Pamela E., 15 Olson, Mancur, 3, 66 Ordeshook, Peter C., 1 Oscillation, 1, 2, 4, 8, 29, 65, 66, 67, 70 P Poland, 42, 47, 52, 53, 55, 56, 63, 83 Jaruzelski, General, 42 KOR, 42, 55 Mazowiecki, Tadeusz, 47 Solidarity, 42, 47, 52, 53, 55, 56, 63, 83 Walesa, Lech, 47 Polish-Russian war, 67 Political transition, 3, 44, 46 Portugal, 15, 23, 24, 29, 31 Balsemao, Francisco, 23 Potter, Merle C., 5 Putin, Vladimir, 37

89 R Ramet, Sabrinia P., 36, 37, 38, 39, 40, 41, 42, 44, 45, 47, 48 Regime transition, 14, 38, 41, 44, 63, 83 Repression, 3, 14, 15, 18, 19, 33, 34–35, 85 harsh, 33, 34–35 inconsistent, 3, 14, 15, 18, 20, 85 Riker, William H., 1, 2, 55 Rohlfs, Jeffery, 52 Romania, 33, 34, 36, 42–43, 48, 53, 59–62, 83 Free Romania, 42 Ron, James, 66 Roughgarden, Jonathan, 2, 5 Russian civil war, 67, 68, 69, 73, 77 Bolsheviks, 35, 68, 73, 77 Chechens, 68, 73 Cossacks, 68, 73 Whites, 68, 73, 77 S Saddle point, 5 Schama, Simon, 81 Semi-presidential government, 17, 28, 47 Sherman, William, 68, 73, 77 Slovakia, 34, 44, 45, 47, 48, 83 Sobek, David, 65 Solzhenitsyn, Aleksander, 33 Soong, Roland, 52 Spagat, Micheal, 69 Spain, 15, 24, 29, 31, 68, 76, 77 Carlos, Juan, 24 Franco, Francisco, 24, 68, 74 Spanish civil war, 24, 67, 68, 69, 72, 74, 76, 84 Sprague, John, 4, 52 Stability, 2, 4, 5, 7, 9, 16, 47, 54, 65–78, 81–85 See also Equilibrium Stalin, Joseph, 35, 41 Sweden, 15, 24, 25, 31, 82 Swistak, Piotr, 67 Switzerland, 9, 15, 24, 25, 27, 28, 31, 82 T Tabak, John, 2 Teune, Henry, 67 Thucydides, 69 Tilly, Charles, 34, 48 Toolis, Kevin, 34 Transition, see Political transition; Regime transition Trotsky, Leon, 35, 68, 73, 77 Tsebelis, George, 4, 9, 27, 28, 52 Turks (in Bulgaria), 38, 57

90 U Unitary, 17, 27, 82, 83 United Kingdom, 9, 15, 21, 25, 26, 27, 28, 29, 31, 82 House of Commons, 27–28 House of Lords, 27–28 Major, John, 25, 30 Thatcher, Margaret, 25, 30 Upper triangle, 84 US, 67, 68, 69, 72, 73, 76, 77, 83–84 US civil war, 67, 68, 69, 72, 73, 76, 77 USSR or Soviet Union, 33, 34, 36, 37, 38, 40, 41, 42, 54, 57, 60, 68

Index V Veto players, 9, 28, 66, 83 W Warsaw Pact, 38, 40, 57, 60 Weigley, Russell F., 78 Wintrobe, Ronald, 3, 33, 34 Wolchick, Sharon, 47 Y Younger, Carlton, 66 Z Zhivkov, Todor, 38, 44

Dynamics of Conflict

Conflict

Conflict

Conflict

Conflict

Conflict

Conflict of Honors

Conflict Of Empires

Conflict of Empires

A Conflict of Interest

Conflict of Laws (1999)

Conflict of Interest

Conflict of Honors

Conflict of Interest

Conflict of Interest

Patterns of Conflict

Conflict of Honors

Encyclopedia of Conflict Resolution

Conflict of Empires

Conflict of Honors

Wheels of Conflict

Conflict of Laws (2001)

The Strategy of Conflict

A Conflict of Interests

Rules of Conflict

Conflict of Empires

A Conflict of Interests

Principles of conflict economics

Hour Of Conflict

A Conflict of Interest

A Conflict of Interest

Dynamics of Conflict

Conflict

Conflict

Conflict

Conflict

Conflict

Conflict of Honors

Conflict Of Empires

Conflict of Empires

A Conflict of Interest

Conflict of Laws (1999)

Conflict of Interest

Conflict of Honors

Conflict of Interest

Conflict of Interest

Patterns of Conflict

Conflict of Honors

Encyclopedia of Conflict Resolution

Conflict of Empires

Conflict of Honors

Wheels of Conflict

Conflict of Laws (2001)

The Strategy of Conflict

A Conflict of Interests

Rules of Conflict

Conflict of Empires

A Conflict of Interests

Principles of conflict economics

Hour Of Conflict

A Conflict of Interest

A Conflict of Interest

Recommend Documents