Party Competition: An Agent-Based Model: 20 (Princeton Studies in Complexity) - Softcover

Laver, Michael; Sergenti, Ernest

Softcover

ISBN 10: 0691139040 ISBN 13: 9780691139043

Verlag: Princeton University Press, 2011

Inhaltsangabe

Party competition for votes in free and fair elections involves complex interactions by multiple actors in political landscapes that are continuously evolving, yet classical theoretical approaches to the subject leave many important questions unanswered. Here Michael Laver and Ernest Sergenti offer the first comprehensive treatment of party competition using the computational techniques of agent-based modeling. This exciting new technology enables researchers to model competition between several different political parties for the support of voters with widely varying preferences on many different issues. Laver and Sergenti model party competition as a true dynamic process in which political parties rise and fall, a process where different politicians attack the same political problem in very different ways, and where today's political actors, lacking perfect information about the potential consequences of their choices, must constantly adapt their behavior to yesterday's political outcomes. Party Competition shows how agent-based modeling can be used to accurately reflect how political systems really work. It demonstrates that politicians who are satisfied with relatively modest vote shares often do better at winning votes than rivals who search ceaselessly for higher shares of the vote. It reveals that politicians who pay close attention to their personal preferences when setting party policy often have more success than opponents who focus solely on the preferences of voters, that some politicians have idiosyncratic "valence" advantages that enhance their electability--and much more.

Über diesen Titel

Über die Autorin bzw. den Autor

Michael Laver is professor of politics at New York University. He is the coauthor of "Multiparty Government: The Politics of Coalition in Europe". Ernest Sergenti is a consultant at the World Bank.

Von der hinteren Coverseite

"Laver and Sergenti argue that politics is best viewed as a complex dynamical system that is neither random nor in equilibrium. Their use of agent-based modeling to better understand the flux of politics will be of great interest to the next generation of modelers seeking to make sense of the wide variety of political systems in the real world."--Norman Schofield, Washington University in St. Louis

"Laver and Sergenti present a bold and rigorous new approach to interparty electoral competition. Their results are remarkable and often puzzling, offering plentiful food for thought for modelers, analytical theorists, and empiricists alike. Ignore this book at your peril, or better, read it and become an admirer."--Kaare Strøm, University of California, San Diego

"Party Competition is an ambitious and pioneering work. Laver and Sergenti present a new methodology for the study of a quite central and traditional problem in political science. This is to my knowledge the first book-length treatise on the evolutionary modeling of party competition."--Hannu Nurmi, author of Models of Political Economy

"Distinct and important. The tools that Laver and Sergenti bring to bear on computational modeling will start a debate that is long overdue in the social sciences. This is another step forward in developing the methods needed to solve real-world problems that have so far resisted our best efforts."--Scott de Marchi, Duke University

Auszug. © Genehmigter Nachdruck. Alle Rechte vorbehalten.

Party Competition

AN AGENT-BASED MODELBy Michael Laver Ernest Sergenti

PRINCETON UNIVERSITY PRESS

Copyright © 2012 Princeton University Press
All right reserved.
ISBN: 978-0-691-13904-3

Contents

Preface............................................................................viiAcknowledgments....................................................................xiiiPart One: Preliminaries............................................................11. Modeling Multiparty Competition.................................................32. Spatial Dynamics of Political Competition.......................................153. A Baseline ABM of Party Competition.............................................284. Systematically Interrogating Agent-Based Models.................................56Part Two: The Basic Model..........................................................835. Benchmarking the Baseline Model.................................................856. Endogenous Parties, Interaction of Different Decision Rules.....................1067. New Decision Rules, New Rule Features...........................................132Part Three: Extensions and Empirics................................................1578. The Evolutionary Dynamics of Decision Rule Selection............................1599. Nonpolicy Factors in Party Competition..........................................18310. Party Leaders with Policy Preferences..........................................20611. Using Theoretical Models to Analyze Real Party Systems.........................22812. In Conclusion..................................................................258References.........................................................................267Index..............................................................................275

Chapter One

Modeling Multiparty Competition

We hold these truths to be self-evident:

? Politics is dynamic. It evolves. It never stops; It is never at, nor en route to, some static equilibrium. Politics evolves.

? Politics is complex. Political outputs today feed back as input to the political process tomorrow.

? Politicians are diverse. In particular, different politicians attack the same problem in different ways.

? Politics is not random. Systematic patterns in political outcomes invite systemic predictions, making a political "science" possible.

Politics in modern democracies is largely the politics of representation. It concerns how the needs and desires, the hopes and fears of ordinary citizens affect national decision making at the highest level, doing this via public representatives who are chosen by citizens in free and fair elections. Representative politics is to a large extent about party competition: about how a small number of organized political parties offer options to a large number of voters, who choose at election time between alternative teams of public representatives. Party competition is therefore a core concern for everyone, be they professional political scientist or ordinary decent civilian, who cares about politics in democratic societies.

We believe that party competition is a complex and evolving dynamic process that can be analyzed in a rigorous scientific manner. More precisely, we analyze the dynamics of multiparty competition, by which we mean competition for voters' support among more than two parties, opening up the possibility that no single party wins a majority of votes cast. Figure 1.1 plots some observations of multiparty competition in the Netherlands over the period 1970–2005. The left panel shows positions of the three main Dutch parties on a left-right scale of party ideology, estimated from their party manifestos. The right panel shows support for these same parties in the Dutch electorate, estimated using Eurobarometer surveys. While some of the plotted "variation" in party sizes and policy positions is surely the result of measurement error, by no stretch of the imagination was the Dutch party system "flatlining" in steady state during the period under observation. It was clearly a dynamic system and, as a result, there were frequent changes in the partisan composition of Dutch governments. These dynamics are clearly a central concern for all political scientists analyzing Dutch politics during this period, be they theorist or country specialist. Equivalent plots can be generated for any party system in which we might be interested.

We Need a New Approach to Modeling Party Competition

Formal models of party competition have been an abiding preoccupation of political scientists since the early 1960s. A vast body of existing work has added hugely to our understanding of party competition. Our own substantive interest, however, and we believe the substantive interest of most people who want to understand party competition in democratic societies, concerns crucial features of party competition that these models typically assume away as a price to be paid for analytical tractability. We ourselves are interested in party competition among many more than two parties. We are interested in "multidimensional" political environments in which politicians and voters care about more than one type of issue. We see politics as a continuously evolving dynamic process that never settles at some static equilibrium, to be perturbed only by random shocks. Pursuing these interests poses formidable theoretical challenges. We show in chapter 2 that dynamic models of multiparty competition, especially when voters care about a diverse set of issues, are analytically intractable. They are not just "difficult" to solve, they cannot be solved using conventional analytical techniques.

The analytical intractability of the relevant theoretical models does not make us any less interested, substantively, in dynamic multiparty competition. Indeed, this very intractability gives us an important and liberating theoretical insight. If analysts cannot use tractable formal models to find optimal courses of action in this setting, then neither can real people making real decisions about real party competition. These people still need to make decisions about what to do. If no formally provable best-response strategy is available, real humans must employ informal decision rules or heuristics. To preview a decision rule we investigate extensively in this book, a party leader might decide to move party policy toward the position currently advocated by some larger rival party, on the grounds there must be more voters who prefer this rival's policy position. We find that this decision rule (which we call Predator) is sometimes very, very good and sometimes perfectly horrid. It is certainly not a "best" response in any conceivable situation but, in the analytically intractable setting of dynamic multiparty competition, it is one of many potentially good rules that politicians may use in certain circumstances when they set party policy positions.

Agent-Based Modeling

Analytical intractability of the decision-making environment, and the resulting need for real politicians to rely on informal decision rules, suggests strongly that we use agent-based modeling to study multiparty competition in an evolving dynamic party system. Agent-based models (ABMs) are "bottom-up" models that typically assume settings with a fairly large number of autonomous decision-making agents. Each agent uses some well-specified decision rule to choose actions, and there may well be considerable diversity in the decision rules used by different agents. Given the analytical intractability of the decision-making environment, the decision rules that are specified and investigated in ABMs are typically based on adaptive learning, rather than forward-looking strategic analysis, and agents are assumed to have bounded rather than perfect rationality (Gigerenzer and Selten 2001; Rubinstein 1998; Simon 1957). ABM is a modeling technology that is ideally suited to investigate outcomes that may emerge when large numbers of boundedly rational agents, using adaptive decision rules selected from a diverse portfolio of possibilities, interact with each other continuously in an evolving dynamic setting (MacGregor et al. 2006).

Putting a particular ABM to work by manipulating its parameters and observing the associated outcomes typically involves computing the outcomes of these interactions if the underlying model is analytically intractable—as is usually the case. Such computation, does not, of its essence, involve electronic computers. One of the most influential early ABMs analyzed housing segregation by scattering black and white chips and then moving them around on what amounted to a large chess board (Schelling 1978). This model was computational in the sense that an abacus is a computer, implemented by moving pieces around a chessboard. As originally published, it did not rely on using an electronic computer. Scatter a number of black and white chips at random on a chessboard; these chips represent people of different color. Assume people have some view about the color of their neighbors; say, for example, they are unhappy if fewer than a quarter of their neighbors are the same color as them. The modeled behavior is simply that unhappy agents move to a randomly chosen close-by empty square that makes them happy. A model "run" begins with chips scattered at random. With an equal number of black and white chips, the typical person will find that 50 percent of neighbors are the same color and will be happy to stay put. There will however be some people in the random scatter who find that fewer than a quarter of their neighbors are the same color; they will move to a square that makes them happy. Everyone is given a chance to move, and to move again, using this rule until there is no unhappy agent who wants to move. The results are striking and unexpected. Even if everyone merely wants at least a quarter of their neighbors to be the same color, modeled population movement typically results in a steady state in which on average about 60 percent of a typical agent's neighbors are of the same color. If we change the key model parameter and assume people to be unhappy, and to move, when they are in a local minority (fewer than 50 percent of neighbors are the same color) then people find that on average 88 percent of neighbors are the same color in the typical steady state that emerges. The deep substantive insight from Schelling's ABM is that intense spatial segregation can arise when people do not seek this at all, but simply prefer not to be in a small minority. More generally, this model shows very nicely that simple decision heuristics can interact to generate complex and unexpected "emergent" patterns of social behavior. This is the core insight of agent-based modeling.

All good things come at a price. The price paid for using computational as opposed to formal analytical models, and thus for using agent-based modeling, is that computation involves calculating model outputs for particular parameter settings. An analytical result, if it is general, is a beautiful thing that is good for all valid parameter settings. Strictly speaking, computational results are good only for those parameter settings that have actually been investigated. Inferences about parameter settings that have not been investigated—and thus more general theoretical inferences we might want to draw from the model—are, in effect, interpolations. This is one reason why we never use computational methods when analytical results are available for the substantive problem that interests us.

The distinction between analytical and computational methods should not be overdrawn, however. A longstanding set of observations that compare models of computation with systems of formal logic, collectively known as the "Curry-Howard isomorphism," shows us that computer programs and formal proofs are in essence the same thing (De Groote 1995). Both take a set of explicit premises and manipulate these, using some system of formal logic, to prove theorems based on these premises. Consider, for example, the area, A, of a circle with radius r. It is well known that we can prove analytically the proposition: A = p · r² for any positive real r. We can also prove A [approximately equal to] p · r² for any given positive real r by various computational methods. With infinite computing power at our disposal, we could prove A [approximately equal to] p · r² for any positive real r. This would not be an "elegant" proof according to most standards of elegance, but now we are talking about aesthetics. With less-than-infinite computing power, we can sample a huge number of positive real values of r, compute A, and show in every single case that A [approximately equal to] p · r². We can draw the statistical inference, at a specified level of confidence, that A [approximately equal to] p · r² for any positive real r. If for some reason it happened that we could not prove analytically that A [approximately equal to] p · r², then this computational/statistical inference would be immensely valuable to us. If we wanted to increase our confidence in this inference, we could simply do more computing and sample more values of r. Of course, we could never be perfectly confident in this conclusion. We can show that A [approximately equal to] p · r² when r = 2.0000001 and 2.0000002; you could claim it is possible A [not equal to] p · r² when r is set between these values, at 2.00000015. Strictly speaking, this would be true. We could however show statistically, with access to enough computing power, that the probability of this exception is extraordinarily small. Furthermore, we could drive down this probability to as low a level as makes you feel happy— simply by doing more computing.

This is an issue we take very seriously indeed in this book since we do want our computational results to have effectively the same scope and precision as those derived from analogous analytical work. We address this by specifying careful procedures for systematically varying parameter settings, and rigorous methods for estimating model outputs of interest associated with these settings. If we carefully design and execute our computational work in this way, then the scope and precision of our results depend only on the volume of computation we are willing and/ or able to deploy. Since we want our own results to have the same scope and precision as typical results from formal models in this field, we are both willing and able to deploy a huge amount of computing power, taking advantage of the Harvard-MIT Data Center's high-performance cluster in order to do this. An important consequence of this is that we are confident that the computational results we present in this book can be "taken to the bank," in the formal statistical sense that, if we were to do very much more computing, or if many other people were to repeat our procedures, essentially identical results would arise. Thus, while this is a book above all about the substantively fascinating topic of multiparty competition, it is also an exercise in how to use computational methods in general, and ABMs in particular, in a way that allows us to draw confident general conclusions.

To summarize, the substantively important real-world problem that interests us is the dynamics of multiparty competition. Theoretical models are no more than intellectual tools designed to help us understand substantively important real-world problems. The technology of classical formal modeling is not a good tool to help us understand the dynamics of multiparty competition, since the resulting models are analytically intractable, with consequences for analysts and more importantly for real humans making decisions in these settings. In contrast, the empowering new technology of agent-based modeling is well suited to investigating problems that are of great substantive interest to us. Impatient for results and problem focused as we are, this book is about how agent-based modeling helps us think systematically about the dynamics of multiparty competition. We start simple and build an increasingly complex model of party competition that deals with a range of substantive matters we have 6 Although our advice to you in this case would be that you should get out more. wanted to think about for a long time but had not really been able to think about in a systematic way before the emergence of ABM.

Plan of Campaign

Chapter 2 sets up the core problem in which we are interested. To demonstrate that this problem is analytically intractable, we use compelling results from a subfield of geometry that deals with "Voronoi tessellations" (or tilings) and has powerful applications in many disciplines. Largely unnoticed by political scientists, this work addresses a problem of "competitive spatial location" that is directly analogous to the problem of dynamic competition between a set of political parties competing with each other by offering rival policy programs. One result from this field is that the problem of competitive spatial location is intractable if the space concerned has more than one dimension (we return below to discuss the meaning of a "dimension" in models of party competition), implying that there are no formally provable best-response strategies for this. This is an important and widely recognized justification for deploying computational methods, and the study of Voronoi tessellations is a major subfield in computational geometry.

(Continues...)

Excerpted from Party Competitionby Michael Laver Ernest Sergenti Copyright © 2012 by Princeton University Press. Excerpted by permission of PRINCETON UNIVERSITY PRESS. All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.

Details zur Ausgabe

• Verlag: Princeton University Press
• Erscheinungsdatum: 2011
• Sprache: Englisch
• ISBN 10: 0691139040
• ISBN 13: 9780691139043
• Einband: Tapa blanda
• Anzahl der Seiten: 296
• Kontakt zum Hersteller: Nicht verfügbar
• Verantwortliche Person: Nicht verfügbar