© Copyright JASSS

JASSS logo ------

Agent-Based Modeling of Social Conflict: From Mechanisms to Complex Behavior

Carlos Lemos
Springer-Verlag: Berlin, 2017

Order this book

Reviewed by Armano Srbljinovic
University of Zagreb

Cover of book This book describes several extensions of Epstein’s model of civil violence (Epstein et al. 2001; Epstein 2002). The author’s goal is to develop an “abstract” agent-based model capable of replicating some “stylized facts” derived from an analysis of “Arab Spring” – the series of social conflicts in the Middle East and North Africa that began at the end of 2010 and continued well into the 2010s.

A comprehensive review of social conflict theories would require a book of its own, but the author provides a good introductory overview, covering those most relevant for the Epstein’s model and its already existing extensions, which are also briefly reviewed.

Lemos’s first extension of the Epstein’s model concerns the form of the function describing a citizen-agent’s “estimated arrest probability”, which serves as the main deterrent/catalyst for the citizen’s “rebellion”. Whereas Epstein used a continuous exponential function of the ratio between the numbers of cop-agents and already rebellious citizen-agents within the citizen-agent’s vision area, Lemos argues that a threshold model is more plausible. Introducing a threshold below which the citizen’s risk estimation falls to zero has a strong impact on the size, duration and recurrence of rebellion peaks, as the simultaneous fall in risk perception in groups of citizens leads to the phenomenon of “massive fear loss”. The author’s argumentation is bolstered by theoretical, mathematical-analytical, and agent-based simulation findings, which, taken altogether, makes it all the more convincing. Simulations also show that another parameter of the Epstein’s model – the “maximum jail term” – strongly influences the interval between successive rebellion peaks: the frequency of peaks decreases with increasing the jail term.

The second extension is the introduction of the concept of relative deprivation (RD) instead of a simple, uniformly distributed “hardship” parameter in the Epstein’s model. The RD is modelled as a citizen’s “sensitivity” to the difference between the median “value capability” (or “wealth”) within the citizen’s vision area and the citizen’s own value capability. In a default setup, the value capability is Pareto Type I distributed in the population. The “sensitivity” is a power function of the perceived difference with an adjustable exponent. The author demonstrates that, for low values of legitimacy, the RD-extended model is capable of producing different regimes (“calm”, “punctuated equilibrium” and “permanent turmoil”), strongly depending on the exponent controlling the sensitivity to the perceived gap and much less depending on the cop-agents density. This indicates that the system’s stability may depend more on citizens’ relative deprivation than on the deterrence/repression capabilities of the central authority.

The third extension is related to legitimacy. Instead of legitimacy as an exogenous parameter in the Epstein model, Lemos introduces endogenous legitimacy feedback into the model by assuming that legitimacy is roughly proportional to the current share of non-rebellious citizens in the population. Introduction of the legitimacy feedback into a low-legitimacy, high-repression, RD-extended model does not change the qualitative characteristics of the outcomes. Furthermore, the addition of the legitimacy feedback induces instability in an otherwise stable high-legitimacy, low-repression setting, and the corresponding time histories vary greatly with unpredictable transitions between the regimes.

The final extension introduces two types of networks into the Epstein’s model. A “group” network is set up by forming cliques of adjustable size between citizens. An “influential” network is set up by connecting a randomly chosen citizen (an “activist”) with other citizens (“followers”) via directed activist-to-follower links. The number and size of such “star” networks are adjustable. Each citizen can be a follower of more than one activist. Whereas in the Epstein model a citizen’s “disposition to rebellion” is individually calculated, the network extension allows modelling “dispositional contagion”: each citizen includes the dispositions of already rebellious agents within that citizen’s group network, as well as the dispositions of all the activists to which he/she is connected, into the calculation of own tendency to rebellion. The form of dispositional contagion is linear, and influence weights of both group and influential networks are adjustable. It is shown that both group and influential networks induce instability in an otherwise stable setting. Preliminary explorations indicate that the degree of connectivity (i.e. the size of cliques and star networks) has a larger impact on the magnitude of rebellion events than the influence weights.

The extensions are fairly complex with a correspondingly large parameter space which is close to impossible to explore exhaustively. The author, however, does a good job of initial exploration and sensitivity analysis, particularly for the first three extensions. He often attempts to identify plausible ranges of parameters by rule-of-thumb calculations and comparisons with the real-world socio-economic indicators, such as Fragile State Index (FSI), Freedom in the World Indicator (FWI) and All the Ginis Dataset. The model’s validation would be extremely demanding. As a first step, the author attempts to provide rough comparisons between the model’s results and the data on the magnitude, duration and frequency of conflict events gleaned from the Social Conflict Analysis Database (SCAD) database.

The language of the book is clear, and the writing style is parsimonious, which is generally commendable. At times, however, the frugality is excessive, as, for example, when descriptions of some parameters (w-group and w-infl introduced at p. 54) appear several pages further on (at p. 60) without a proper pointer, or when some parameters (peak.threshold and diff.threshold introduced at p. 76) apparently lack any description. The readability of the book would be significantly improved if more supplementary explanatory material were included.

Lemos rightly claims that the book is of general interest to researchers of social conflict, particularly those using agent-based modelling and simulation methods. Although the author does not explicitly cite the body of work in analytical sociology (e.g. Schelling 1978; Hedström & Swedberg 1998; Hedström 2005; Demeulenaere 2011), this book, with its strong emphasis on social mechanisms, stands close to this tradition and can be regarded as an engaging example of the advanced application of analytical sociology to studies of social conflict. The book can also be of interest to researchers in sociophysics (Galam 2012), as well as to those interested in quantitative research of social movements.

* References

DEMEULENAERE, P. (ed.) (2011). Analytical Sociology and Social Mechanisms. Cambridge, UK: Cambridge University Press.

EPSTEIN, J. M. (2002). Modeling Civil Violence: An Agent-Based Computational Approach. Proceedings of the National Academy of Sciences, U.S.A. 99 (3), pp. 7243-7250.

EPSTEIN, J. M., Steinbruner, J. D. & Parker, M. T. (2001). Modeling Civil Violence: An Agent-Based Computational Approach. Working Paper No. 20. Washington, D.C.: Center on Social and Economic Dynamics, The Brookings Institution.

GALAM, S. (2012). Sociophysics: A Physicist's Modeling of Psycho-Political Phenomena. Berlin: Springer-Verlag.

HEDSTRÖM, P. (1998). Dissecting the Social: On the Principles of Analytical Sociology. Cambridge, UK: Cambridge University Press.

HEDSTRÖM, P. & Swedberg, R. (1998). Social Mechanisms: An Analytical Approach to Social Theory. Cambridge, UK: Cambridge University Press.

SCHELLING, T. C. (1978). Micromotives and Macrobehavior. New York, NY: Norton.


ButtonReturn to Contents of this issue