Agent–Based Models of World System Formation
Joachim K. Rennstich
March 25, 2010
Abstract
This paper tracks the first attempt to develop a modular agent-
based world system model using NetLogo. After a brief introduction
into complex systems theory, the paper discusses the possibilities of
modeling world system evolution and processes using agent-based mod-
els (ABMs) as opposed to equation-based models or relying on histori-
cal analysis, currently the dominant approaches. It proposes a modular
approach that allows a step–wise and –collaborative approach, using
an ABM that simulates the development of polarity structures based
on various hypotheses proposed in the literature as an example.
Prepared for the 51th International Studies Association (ISA) Annual
Convention, New Orleans, LA, February 17–20, 2010.
Department of Political Science, Fordham University, 113 W 60TH ST, RM 916, New
York, NY 10023. Contact the author at rennstich@fordham.edu. Further information
about the author can be found at www.rennstich.com.
1

1 Too Big To Handle?
The mere thought of aiming to capture globalization — or rather: the evo-
lution of the current world system — in all of its complexity in depth and
width, connecting so many actors through so many layers, seems futile at
best, and foolish to most. It is easy to understand this reaction. With tra-
ditional formal modeling techniques widely in use in international relations
(IR) any such attempt would prove rather difficult. In recent years, however,
more social sciences have started to make use of new computational modeling
tools. Traditionally, researchers have relied on static equation–based meth-
ods of capturing the effects of actors’ behavior on each or the constraining
or enabling effects of a given environmental set of factors. With the lower
costs of powerful computer technology and correspondingly powerful, yet
easy to use software tools, researchers now can aim to simulate agency as
it takes place in our observed world that takes into account not only the
effect of agency on other agents, but the effect of environments on agents
and the effect of agency on the environment in which this agency takes place.
These interactions can produce a dynamic that is very hard to capture in
equation–based modeling, especially if larger groups of agents or actors are
involved.
This paper aims to introduce researchers who are interested in using these
new technologies in their own work on globalization processes, especially the
long–term kind. It uses the argument that the world system as such is
a complex system that follows certain dynamic regularities and allows for
the emergence of a system and order without the need for a centralized
organizational plan or planner as an example, of how agent–based models
(ABMs) can be useful to scholars engaged in not only this area, but many
others as well. It first discusses the theory of complex systems and the use
of ABMs in a more general manner. It then continues to describe two such
models — “Giant Component” and “Polarity” — as concrete examples of the
uses of ABMs in research on world system development.
2 Modeling Complex Systems
The evolution of social structures understood as complex systems that form
larger world system developmental processes have been a of interest to stu-
dents of IR since Axeldrod’s early work on cognitive mind-maps ?, further
developed in his seminal work on the evolution of cooperative strategies (?).1
1For a collection that traces his work over two decades, see ?.
2

A growing literature (for example, ????????????????) on the special in-
sights gained from studying social systems as complex ones now yields a
substantial body of work that can be employed in evolutionary global and
world system approaches.
Complex systems analysis offers us insights into the way systems estab-
lish order without a singular or initial ordering entity. Yet an order (or
developmental logic) does emerge in such systems, based on feedbacks re-
sulting from trial and error, adaptation, and system-wide learning, resulting
in a system that features self-organization.2
The formation of a world system understood as a long-term social system
(involving economic, political, and cultural processes) forming a global social
world resembles such an emerging ordered system without a single orderer.
No single power, whether an empire, state, or any other unit, has transformed
the human social world over the last five hundred or thousand years (or any
other period) into the world we experience it today. Rather, globalization
thus understood has been the result of a number of reoccurring processes
of trial and error, adaptation, system-wide learning and thus: a complex
system based on the principle of self-organization.
Employing the general lessons derived from the study of complex systems,
it is possible to identify the general (or meta) developmental logic of the
long-term globalization process while at the same time leaving room for
divergent schools of explanations on the factors that influence important
order-structuring factors such as learning or adaptation in the system.
All socio–political forms derive ultimately from human agency. We there-
fore need to focus on the link between the individual, structure, and the sys-
tem, in other words decision-making (whether constrained or freely made).
However, with increased socio-political complexity comes greater need to
manage information. This, so ?, 121–2, leads to the following internal dy-
namic: complex socio-political systems will unavoidably increase in complex-
ity, with the consequence of a higher rate of decision-making. As a result,
the need for more efficient information processing rises, in order to respond
to decision-making requirements produced by its own degree of complexity
and the structures it employs. This complexity dynamic not only increases
at first the opportunity for socio-political innovation to occur, but also po-
tential instability (see also ???).
2For a very accessible summary of complex system theory, see ?, esp. Parts II and
III; for a brief general overview with a focus on items especially important to students
of international relations, see ???; for a recent volume on complexity and IR, see ?;
for examples of an application of a similar approach, see ??; however, see also ?? for
alternative discussions.
3

As the complexity of the system increases, the range of possible deci-
sions promoting further growth will decrease. In other words, an increasingly
complex system will become increasingly path-dependent and loses its adap-
tive flexibility. Unhindered hypercoherent option-narrowing will eventually
precipitate collapse, which will be a sudden occurrence. In other words,
socio-political complexity will rise to the point of edge of chaos.
? and Devezas and his collaborators (???) discuss the relationship be-
tween innovation and systemic development as being based on information
processing and decision-making and prompted by interaction, generating a
successional model of endogenous socio-political change, where the succes-
sor system might follow a similar trajectory to that of the system which it
replaces, collapsing itself in time and being replaced by yet another socio-
political system. Progression takes place stochastically through trial and
error, random non-average fluctuations. In this process, the system self-
organizes and learns to configure and reconfigure itself toward increasing
efficiency and in this manner with each iteration it performs some activ-
ity better. Each stage corresponds to a given structure that encompasses
previous self-organization, learning and the current limitations (?, 22).
As pointed out earlier, decision-making (and thus the process of agency)
does not take place in an isolated environment but rather a strongly con-
textual one. Agency affects the environment in which it unfolds, but also
is formed by it. Thus, it is important not only to focus on the agents (on
a multitude of levels) but also to identify the contextual environment in
which this agency takes place. Change in complex systems, whether in the
direction of greater or lesser complexity, produce a trajectory or historical
path, limiting future options and thus becoming path-dependent in this way.
As a consequence, complex systems as the ones observed here–the nested
global economic, political, social, and cultural processes that form the world
system we identify as the global complex system–exhibit a tendency to self-
organization, that is the endogenous ordering into hierarchies giving them a
system-wide form. Thus, these complex systems exhibit morphogenesis (i.e.,
the development of an organism or of some part of one, as it changes as a
species) based on processes that are partly independent of agency, although
they require agents to both initiate them and enact them (??). The way
the interrelationships between parts of the systems are established (i.e., the
structure of the networks) thus becomes crucial for our understanding of the
dynamics of these coevolving structures.
Information
Computation
4

Randomness or Order?
Evolution
3 Modeling a World System
Evolutionary systems theory also takes into account the role of agents (i.e.,
individual actors, groups, etc.), as it acknowledges that social change is
driven by innovations from these agents. As pointed out earlier, change
is necessarily at the heart of an evolutionary model. However, given the
scope of the analysis (in terms of levels of analysis, temporal and geograph-
ical orientation, as well as areas of inquires), it is impossible to focus on all
effective processes with equal weight. Yet, what changes (or rather structures
of change) are important? Any generalization must be simplifying while pre-
serving its value as an explanatory tool for a wider range of questions. Thus,
the main goal of such an evolutionary model must be to unravel the main
change processes that drive the system and provide a point of reference over
a multitude of levels of analysis and areas of inquiry.
A general evolutionary systems framework for the social sciences that
focuses on the development of a global world world system (especially one
that views it as a global complex system) will always be based on the fol-
lowing evolutionary logic (in some shape or form) that explains the creation
of possibility space or in other words the potential options for change open
to the systems and its parts (?). This evolutionary logic driving the global
system process is based on the following set of epistemological assumptions
borrowed here from evolutionary economics (?, 15):
 agents (e.g., individuals, groups, organizations, etc.) can never be
“perfectly informed” and thus have to optimize (at best) locally, rather
than globally;
 an agent’s decision-making is (normally) bound to rules, norms, and
institutions;
 agents are to some extent able to imitate the rules of other agents
(imitation), to learn for themselves, and are able to create novelty
(innovation);
 the processes of imitation and innovation are characterized by signifi-
cant degrees of cumulativeness and path-dependency (but may inter-
rupted by occasional discontinuities);
5

 the interactions between the agents take place in situations of disequi-
libria and result in either successes or failures of commodity variants
and method variants as well as of agents; and
 these processes of change are non-deterministic, open-ended, and irre-
versible.
The point of developing such a general evolutionary systems framework
is not, however, as pointed out many decades ago by ?, to produce all-
encompassing, broad, and general theories of everything, nor to spend time
in empirical accounts per se. Rather, what we can aim for is to formalize,
empirically test, use, and extend theoretical models able to shed light on the
causal mechanisms that are behind the complexity of empirical phenomena.
The use of of agent–based models as an additional and novel tool is therefore
not an attempt to model world system development as it emerged, but rather
the causal links and processes that have led to its emergence.
3.1 Challenges and Limitations
3.2 Creating a Common Platform
3.3 Agent-based Models vs Equation–based Models
Modeling has a long-standing tradition in the social sciences and Interna-
tional Relations of course. In fact, many publications feature — and pub-
lishers require — some element of computational model to provide empirical
support for theoretical arguments developed in the paper or book. Most
of these computational models are mathematical or equation–based mod-
els, such as structural equation models found in quantitative sociology and
political science.
Their explanatory strength is usually measured in the degree of fit be-
tween some data and an equation describing the relationship between vari-
ables and how they are related to each other. Why this relationship exists
however is something that often is not captured very well in these equa-
tions. The knowledge about the processes and mechanism connecting agents
and/or observable factors expressed in the data is therefore often missing
and added on to the analysis afterwards.
Agent-based models are far better equipped to help us understand those
processes and mechanisms, yielding data that in turn can be used for equation-
based analyses, because they model not only interaction between actors but
also interaction of agents within an environment. These relationships remain
6

not static as either agents or the environment change, a dynamic that agent-
based models capture extremely well. ?, 14–16 summarizes the spectrum of
current agent–based modeling discussing some of its characteristic features:
Ontological Correspondence: Compared to equation–based models, it is
far easier to include agents in an computational model that directly
correspond with real-world actors.
Heterogenous Agents: Rather than following a reductionist approach or
assuming a “typical” representation of an actor, agent–based models
enable the computation of models that allow for agents to operate
according to their own preferences or even rules of action.
Representation of the Environment: Not on physical factors that im-
pact the behavior of agents or outcomes of their actions can be in-
corporated into models, but also the effects of other agents in the
surrounding locality, or effects of crowding or depletion of resources.
Agent Interactions: Interactions between agents can be simulated in agent–
based models, even in terms of exchange of messages or information
and the computation of these messages.
Bounded Rationality: Far from assuming a rational or randomly-behaving
actor, agent–based models make it easy to model agents as bounded
rational actors, allowing for a clear specification of these boundaries
that mimic real–world situations far more accurately.
Learning: The ability to simulate learning is another aspect that truly sets
agent–based models apart from equation–based models, either as (1)
individual learning (learning from own experience); (2) as evolutionary
learning (population of agents learn collectively through generational
iterations); or (3) social learning (learning through imitation or by
being taught by other agents of individually learned, but collectively
distributed experiences).
3.4 Types of Models
Agent–based models are not uniform. Some models are used primarily to
formalize basic theories about dynamics or processes. Others are used as
general descriptions for more specific social phenomena whereas other even
aim to replicate very closely empirically observed specific target phenomena.
Taking up Cioffi-Revilla’s (?) call for a more unified methodology for com-
plex social simulations, the following discussion highlights briefly the role
7

these different types of models can play in the development of a common
platform for an agent–based model of world system development.
3.4.1 Abstract models
3.4.2 Middle range models
3.4.3 Facsimile models
3.5 NetLogo as a Computational ABM Platform
NetLogo is a programmable modeling environment for simulating natural
and social phenomena, authored by Uri Wilensky in 1999 and has been in
continuous development ever since at the Center for Connected Learning and
Computer-Based Modeling (?). As a computational modeling tool, NetLogo
is ideal for modeling complex systems developing over extended periods of
time, with instructions given to hundreds or thousands of agents all operating
independently but interactively. This makes it possible to explore the con-
nection between the micro-level behavior of individuals and the macro-level
patterns that emerge from the interaction of many individuals.
4 “Giant Component” and “Polarity” — A Simula-
tion of Structural Emergence
In the following section, we will use one one of the publicly available mod-
els that come standard with NetLogo, as well as model that uses elements
and code from other NetLogo models, to demonstrate the utility of a large,
preexisting user–base for a project that aims to capture the dynamics and
evolution of the entire world system.
4.1 Giant Component
NetLogo provides users with a handful of models to explore the possibilities
of the modeling platform. One such model is the “Giant Component” model
(?). As discussed earlier, the world system as a complex system is viewed
as such a network, connecting a wide range of agents over a multiple layers.
In a network, a “component” is a group of nodes that are all connected to
each other, directly or indirectly. If a network has a “giant component,” that
means almost every node is reachable from almost every other.3 The “Giant
3Other authors use different terminology, but describe the very same phenomenon, as,
for example, ? who use the term “web of webs.”
8

Component” NetLogo model shows how quickly a giant component arises if
you grow a random network.
4.2 Polarity
5 Next Steps
9