Reviewed by
Warren Thorngate
Department of
Psychology, Carleton University, Canada.
The contagion of scientific ideas and methods usually begins with the informal circulation of messages among scientists, followed by the organisation of conferences, the founding of institutes, the publication of edited readings, the development of journals and the creation of textbooks spreading the word to a new generation. Indicative of the growth of computer simulation in the social sciences is the founding of JASSS and the publication of several edited books describing the simulations of their authors. At least a dozen such books have been published in the past decade, bringing a feeling of legitimacy to a nascent specialisation and marking a new stage in the evolution of a methodology.
One of these books has been recently edited by Liebrand, Nowak and Hegselmann: Computer Modeling of Social Processes. Following Liebrand's concise introduction to the history and current issues of computer simulation, the book's nine remaining chapters are equally divided into three modeling themes: simulations, neural networks and data analysis. Included are the following papers.
Troitzsch provides a useful discussion of differences among approaches to modelling and computer simulation, offering a taxonomy to organise some important points of distinction. Hegselmann gives a readable introduction to cellular automata with clear examples of the evolution of co-operation and support networks. De Vries presents a brief discussion of self-organising conceptual networks. Alas, I found its relation to the theme of the book, social processes, a bit unclear.
Nowak, Vallacher and Burnstein give us the longest of the three chapters covering various aspects of neural networks. It is an excellent introduction to neural network concepts and their possible use in modelling social dynamics, particularly concepts of attractor networks full of feedback loops. The authors end their chapter with a valuable perspective on possible limits of attractor network concepts, and suggest how the concepts might be expanded to be more suitable for exploring social processes. Gernert provides a second neural network chapter, discussing two important implications of using neural networks for classification and prediction: network reliability and responsibility. The former sets limits on the predictive validity of a neural network; the latter exposes limits of the people who use neural networks for classifications and predictions that affect the lives of others. He ends his short chapter with a discussion of the explanatory limits of neural networks and their implications for understanding. Adèr and Bramsen provide the third neural network chapter, comparing and contrasting structural equation models with neural network models that predict post traumatic stress disorder from a large data set. Though both types of models do a fair job of prediction, the comparative exercise nicely illustrates some of the practical considerations of neural network training and shows how neural models can supplement structural equation models in revealing the influences of predictor variables.
The final three chapters of the book concern data analyses. Yung, Chan and Bentler, in a primarily statistical chapter, analyse bootstrap tests of cross-classification data. Geuze, Ouwerkerk and Mulder present a thoughtful discussion of methodological issues arising when relating behavioural observations to physiological correlates. Finally, Klovdahl leads us on a tour of manual and computer methods developed to graph large and complex networks, noting the paucity of computer programs well-adapted to visualising complex social networks.
As I read this book, I formed two impressions. First, each chapter taught me something. Though some taught me more than others all are worth the attention needed to read them. Second, fewer than half the chapters taught me something about computer modelling of social processes, the title and putative theme of the book. Perhaps because of shrinkage in my own neural network, I was unable to make the conceptual stretch needed to understand how Monte Carlo simulations of null hypothesis tests, predictions of post traumatic stress disorder and a list of characteristics of electrophysiological data were related to computer modelling of social processes. It was interesting to learn about these and other unrelated topics. I fear, however, that the chapters containing them will miss the wider or more appropriate audiences they deserve, simply because they have been packaged between the wrong covers.
In sum, half of this edited volume contains valuable chapters about computer modelling of social processes, chapters suitable as introductions to central issues in the area. The remaining half of the book contains valuable chapters about other topics. Those looking for a book only about computer modelling of social processes will probably finish this one feeling half-empty. Those wanting a more varied diet will find the additional chapters filling as well.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 1999