Reviewed by
J. Theodore Anagnoson
Department of
Political Science, California State University, Los Angeles,
USA.
This collection of 10 chapters plus an introduction and conclusion covers topics ranging from federalism, to neural networks, pattern recognition, and congressional campaign contributions. The authors are mostly academics in political science or government departments, plus several individuals from information sciences, economics, and decision science. The articles have in common that they all deal in some way with a non-linear model as a way of understanding the political complexity we so often see. Many of the chapters relax classical assumptions and study complex situations with many actors rather than the classical two. The book is accessible to beginners in many places; the articles have a good historical sense of what the literature has done with the particular substantive problem in the past, as well as complete bibliographies. Readers need know only the basics of multivariate techniques, including regression and probit/logit analysis.
Diana Richards begins with an introduction that asks the meaning of non-linear in political science and reviews the contributions of each author. Non-linear implies, as Richards states, an independent variable with a variable effect on a dependent variable. The essays have as their goal showing that non-linear modelling can "be a constructive enterprise that yields interesting hypotheses about a wide range of political topics." (p. 3). Within political science, the range of approaches that can be termed non-linear is wide, including non-linear dynamic systems, neural networks, symbolic dynamics, or spatial models of agent-based interaction. Non-linear methods thus cut across a wide range of methodologies, including "dynamical systems, game theory, spatial voting models, time series analysis, non parametric estimation, and logit/probit models." Richards' overview is gives a good sense of where the fields is and where it has come from and is recommended on that basis alone.
In "Landscapes as Analogues of Political Phenomena," D. Scott Bennett explains landscape theory, previous literature using landscape theory in political science, the assumptions involved, and compares landscape theory with cluster analysis and realism theory. He explores the consequences of relaxing classical assumptions and details the substantive areas where the theory might be applied. This chapter is an interesting analysis of the theory and methodology of landscape analysis.
Chris Brooks, Mel Hinich, and Robert E. Molyneux, in "Episodic Non-linear Event Detection: Political Epochs in Exchange Rates," show how standard economic data, in this case exchange rates, can be used to detect epochs related to political and economic events. It will be of interest to those who are interested in time series analysis using sophisticated techniques to detect higher order non-linearity.
In "Congressional Campaign Contributions, District Service, and Electoral Outcomes in the United States: Statistical Tests of a Formal Game Model With Non-linear Dynamics," Walter R. Mebane Junior asks the question why U. S. congressional incumbents are able to accumulate large war chests, and the next rather counter-intuitive question, why would a campaign contributor want to make an election non-competitive? The argument here is that there is no simple, linear relationship among contributions, district service, challenger quality, and election outcomes - thus there is a non-linearity in this strategic interaction. Mebane has a two stage game theoretic model with some rather interesting results - most importantly, that voter preferences only partially determine the outcome of the game even though candidates and contributors are all constrained by their anticipation of what voters are likely to do. This is one of the articles that would be of strong interest to those concerned with substance, in this case, legislative behaviour.
Susanne Lohmann in "I Know You Know He or She Knows We Know You Know They Know: Common Knowledge and the Unpredictability of International Cascades," investigates the paths taken by "dynamic informational cascades." Cascade effects occur when the decisions of citizens are interdependent over time. Each person's payoff depends on the number of other people who do so at the same or later time. This essay has a very interesting discussion of interdependence and the implications of interdependence with a seemingly simple result - the paths taken by these dynamic informational cascades are "unpredictable" and the outcome for society is "fragile."
In "Non-linear Dynamics in Games: Convergence and Stability in international Environmental Agreements," Diana Richards asks whether one can explain and predict the kind of international environmental treaty that nations will agree to. She uses the approach called "learning in games," which is an exploration of the dynamics of the actors' strategic choices over time. She focuses on the number of states that must agree to a treaty and whether those states have a shared perception of the environmental threshold. Her exploration suggests that even simple variables like the number of participants or the presence of scientific consensus can have counter-intuitive effects on the stability of the players' interaction.
Daniel P. Carpenter examines the old Herb Kaufman question about whether government organisations are immortal, using Kaufman's own data, in "Stochastic Prediction and Estimation of Non-linear Political Durations: An Application to the Lifetime of Bureaux." He uses a non-linear modelling approach to this problem. Anthony Downs stated that governmental bodies are most vulnerable to being terminated just after they have been established and that thereafter the probability that they will be terminated declines. Carpenter finds that with a non-linear approach, this finding is incorrect. This article has a good discussion of the informational delegation approach to the establishment of bureaux and a nice sense of how this problem has been handled in the literature thus far.
In "Neural Network Models for Political Analysis," Langche Zeng compares the results of analysis with neural networks for political data that have already been analysed with linear methods, such as regression and probit analysis. This article is a good introduction to this complex topic. Zeng uses both Monte Carlo data and older previously analysed data to show how neural network models offer improvements in the ability to predict and explain. This article is quite accessible to those who don't know much about this topic.
David H. Bearce, in "Economic Sanctions and Neural Networks: Forecasting Effectiveness and Reconsidering Co-operation," considers whether real world needs for forecasting in practical terms might make a neural network preferable to traditional (and linear) statistical analysis. The particular case he examines is the effectiveness of economic sanctions, using roughly 100 quantitative cases first examined in the 1980s. He finds that the neural network forecasts twice as well as traditional statistical methods. As with the other cases, there is an excellent survey of how political scientists and economists have treated this problem.
In "Pattern Recognition of International Crises Using Hidden Markov Models," Philip A. Schrodt demonstrates the use of a non-linear sequence recognition technique, hidden Markov models, to differentiate crises in the Behavioural Correlates of War data set. The sequences are then applied to a contemporary data set on the Middle East. Schrodt has used the method before. This article is a good introduction to an entirely different method for recognising patterns in data.
There is a conclusion in which the editor, Diana Richards, ties the essays together.
All in all, this is an important work for graduate students who have taken courses in the basic linear statistical approaches and wish to learn more about what can be done to relax their assumptions, go further with similar methods, or use a completely different method to study the particular phenomena whose existence or change is of interest. Many of the chapters are more accessible than one might think, and the editor is to be commended for asking the authors to have a sense of how the literature and the discipline have treated the problem at hand in the past. Strongly recommended.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2002