A recent overview of the integration of System Dynamics

and Agent-based Modelling and Simulation

Graciela d. C. Nava Guerrero*1, Philipp Schwarz*1, Jill H Slinger1,2

1 Faculty of Technology, Policy and Management, Delft University of Technology, Netherlands

2 Faculty of Civil Engineering and Geosciences, Delft University of Technology, Netherlands

Correspondance should be addressed to [email protected] and [email protected]

March 22, 2016

Abstract: Modelling and simulation aim to reproduce the structure and imitate the behavior of real-life
systems. For complex dynamic systems, System Dynamics (SD) and Agent-based (AB) modelling are two
widely used modelling paradigms that prior to the early 2010’s have traditionally been viewed as mutually
exclusive alternatives. This literature review seeks to update the work of Scholl (2001) and Macal, (2010) by
providing an overview of attempts to integrate SD and AB over the last ten years. First, the building blocks of
both paradigms are presented. Second, their capabilities are contrasted, in order to explore how their integration
can yield insights that cannot be generated with one methodology alone. Then, an overview is provided of
recent work comparing the outcomes of both paradigms and specifying opportunities for integration. Finally,
a critical reflection is presented. The literature review concludes that while paradigm emulation has contributed
to expanding the applications of SD, it is the dynamic combination of the two approaches that has become the
most promising research line. Integrating SD and AB, and even tools and methods from other disciplines,
makes it possible to avoid their individual pitfalls and, hence, to exploit the full potential of their
complementary characteristics, so as to provide a more complete representation of complex dynamic systems.

Word count: 4974

Keywords: System Dynamic· Agent-Based Modelling · Hybrid Models · Complex Dynamic Systems ·
multi-paradigm approach · Literature Review

1 Introduction
Modelling and simulation of complex social systems aim at increasing the understanding of the system
and testing policies with the objective to support decision-making and at times policy implementation
(Meadows and Robinson, 2002). The advantage of computational models are their capability to embrace
complex real-life systems characterized by dynamic nonlinear relationships. Another substantial benefit
is that what-if scenarios can be tested, but intervention in reality is not required.
Agent-based (AB) modelling and System Dynamics (SD) are two widely used methodologies in
modelling complex dynamic system. While System Dynamics has a long tradition since it was founded
in the late 1950s by Forrester (1958), AB is as yet in its infancy - implying that its complete potential
has not yet been utilized (Bonabeau, 2002). Both approaches have been applied to many socio-economic
problem domains including health care (Demarest, 2011; Figueredo, Aickelin, & Siebers, 2011;
Figueredo, Siebers, Aickelin, Whitbrook, & Garibaldi, 2015; Kirandeep, Eldabi, & Young, 2013;
Mellor, Smith, Learmonth, Netshandama, & Dillingham, 2012), supply chains (Angerhofer &
Angelides, 2000; Georgiadis, Vlachos, & Iakovou, 2005; Gjerdrum, Shah, & Papageorgiou, 2001; Tako
& Robinson, 2012; Xue, Li, Shen, & Wang, 2005) and technology adoption (Chen, 2011; Fisher,

-1-
Norvell, Sonka, & Nelson, 2000; Moser & Barrett, 2006; Schwarz & Ernst, 2009; C Swinerd &
McNaught, 2014; Zhang & Nuttall, 2007).

More than a decade ago, Scholl (2001) made a call for joint research between SD and ABM by
comparing and contrasting both approaches, and more recent works have enriched those comparisons
(Lättilä et al., 2010; Macal, 2010). However, during the last decade, and particularly during the last five
years, an explosive growth in computational capacity has enabled the emergence of more, and more
diverse, joint research in the field of modelling and simulation (Pruyt, 2015).
This article seeks to update the work of Scholl (2001) and Macal (2010) by providing an overview
of attempts to integrate SD and AB over the last ten years, with an emphasis on hybrid SD-AB models
published over the last five years. The research strategy comprised a systematic literature review.
Combinations of the following key words were used: agent-based modeling, combining, differential
equations models, system dynamics, and hybrid models. The objective was to compile literature related
to the ongoing discussions on the complementary potential of integrating SD and ABM, and to provide
an overview of recent case studies. The research question was formulated as:

What are the potential benefits of integrating System Dynamics and Agent-based
and what is the state-of-the-art in its application?

The reviewed literature was retrieved from several research databases, including ACM Digital Library,
Elsevier, Springer-link, EBSCO Host, and IEEEXplore. The works by Scholl (2001), Lättilä et al.
(2010), Macal (2010), Schieritz and Grobler (2003) and Behdani (2012) were used as a guide in
structuring the research process.
The remainder of this paper is ordered as follows: Section 2 gives a short overview about the SD
and AB paradigms, including theories behind the paradigms and building blocks and characteristics of
the resulting models. Section 3 contrasts the capabilities of SD and AB, in order to explore how their
integration can yield insights that cannot be generated with only one methodology alone. This section
draws from a review of recent studies that combine both paradigms. Section 4 presents how both
methods have been integrated during the last decade, and explores expected developments in this field.
Lastly, Section 5 concludes by answering the research question and delineating opportunities for future
research.

2 Single Paradigms: System Dynamics and Agent-based

Prior to the 2010s, the SD and AB paradigms developed as separate schools of modeling and simulation
(Pruyt, 2015), in spite of both paradigms being used for the analysis of complex dynamic systems
(Phelan, 1999). This Section presents an overview of the fundamental theories behind each paradigm
and of the building blocks and characteristics of their corresponding models.

2.1 System Dynamics (SD) Models

More than 50 years ago, Forrester (1958) founded SD around two notions from systems theory (Phelan,
1999): first, aggregated-level variables affect each other through feedback loops; second, system’s
structure drives system’s behavior. These notions challenge the predominant rather simplistic cause-
and-effect thinking of traditional science, decoded into independent and dependent variables. Instead,
systems theory explains the behavior of complex dynamic systems endogenously: it identifies feedback
effects that are often hidden because they are delayed at large time scales. Consequently, systems
dynamics modelling targets the underlying causes of problems instead of only treating their symptoms
(Forrester, 1958; Sterman, 2000).
-2-
 In practice, the building blocks in specifying an SD model are stocks, flows and auxiliary
variables (Forrester, 1958; Sterman, 2000). Stocks represent the accumulation of material and
information, caused by the action of inflows and outflows. While stocks are mathematically described
by integral equations, flows are described by differential equations (Macal, 2010; Parunak et al., 1998)
. The solution of these sets of equations describes the aggregated state of the system. This state changes
continuously over time and depends on the previous state of the system. These sets of equations are
solved through numerical integration at discrete time steps (Forrester, 1958; Meadows, 2009; Sterman,
2000).

2.2 Agent-based (AB) Models

The theory of complex adaptive systems (CAS) states that systems do not have central control and do
not have a fixed structure. Based on this theory, the AB paradigm models the structure of a system as
the result of decentralized decisions of individual entities or agents over time (Macal, 2010; Macal and
North, 2006). Therefore, instead of assuming a given system structure, agents’ decisions shape and
change the state and structure of the system. In turn, agents react to the dynamic changes in the system,
which can potentially alter their decision rules.
It follows that the main building blocks of AB are autonomous agents, their decision rules and
actions, and the environment in which they interact (Bonabeau, 2002; Epstein and Axtell, 1996; Phelan,
1999). Although agents’ decision rules usually govern agents’ behavior to achieve individual benefits
(Macal and North, 2006), collective intelligence may also emerge when agents coordinate their
decisions to achieve common goals (Phelan, 1999). Therefore, analyzing solely the internal mechanism
of agents does not explain the macro level observations (Epstein, 2006). Moreover, agents’ decision
making is typically based on limited observed knowledge (their view on the world) rather than on
complete knowledge of the entire state of the system (Jennings et al., 1998).

3 Potential benefits of integrating System Dynamics and Agent-based

The contrasts between SD and AB, including scope, the focus on system behavior or on emergent
behavior, aggregation level and the current capacity to study heterogeneity and spatial variability, make
the application of each paradigm more suited to different situations (Macal, 2010; Scholl, 2001; Teose
et al., 2011; Wakeland et al., 2004). Nevertheless, knowledge about the differences between SD and
AB does not necessarily result in an appropriate choice of paradigm: one paradigm alone cannot always
provide enough insights to analyze the complex system of interest (Lättilä et al., 2010; Macal, 2010;
Rahmandad, 2004; Scholl, 2001; Shafiei et al., 2013a).
In this Section, five characteristics in which SD and AB differ fundamentally are explained first.
Second, the potential benefits of combining both paradigms are clarified.

3.1 Contrasting SD and AB - five fundamental differences

The applicability, strengths and weak points of SD and AB paradigms have been compared by designing
independent models of the same system and contrasting their outcomes. Recent contributions include,
but are not limited to Figueredo and Aickelin (2011), Macal, (2010), Milling and Schieritz (2003),
Parunak et al. (1998), Rahmandad and Sterman (2008), Schryver et al. (2015). For this article, a number
of such comparisons were reviewed and five fundamental characteristics in which SD and AB differ
were identified. These aspects include the paradigms’ capacity to model continuous aggregated and
discrete disaggregated system states; physical space, topographies, and network structures; stochastic
& deterministic phenomena; learning and adaption; and ease of model building and interpretation. The
paragraphs below elaborate on each of these aspects.

-3-
3.1.1 System states: continuous aggregated vs. discrete disaggregated
SD and AB paradigms differ in the level of aggregation and their handling of time. On the one hand,
SD excels at representing continuous aggregated systems. This paradigm can account for a wide range
of feedback effects, at the cost of reducing real world diversity to aggregated average values by
assuming homogeneity and perfect mixing within stocks and flows (Parunak et al., 1998; Rahmandad
and Sterman, 2008; Sterman, 2000). However, While SD excels at modeling continuous processes, it
has difficulties in coping with discrete events (Parunak et al., 1998). Therefore, AB is more appropriate
to model discontinuous system properties (Bonabeau, 2002).
In contrast to SD, the AB paradigm inherently includes heterogeneity between agents. To account
for the diversity of agents in the real world, agents act according to properties and decision rules that
can be derived from distribution functions (Bonabeau, 2002; Epstein, 2006; Macal, 2010). By
accounting for the diversity within and between agents, AB is suitable to study problems where the
distribution of resources, costs or benefits is the focus of interest (Bonabeau, 2002; Osgood, 2007).
Empirical research has emphasized a tension between the level of analysis and the scope of the
system under study when using SD or AB alone (Alam Napitupulu, 2014; Cherif and Davidsson, 2010;
Figueredo et al., 2015; Silva et al., 2011; Thompson and Reimann, 2010). While SD can study large
systems by handling highly aggregated data, AB typically studies heterogeneous systems with relatively
limited scope.
In practice, choosing a paradigm to describe a system at an appropriate level of analysis is not
straight forward. In reality, this aspect is observer dependent: the same system can be described with
both discrete and continuous representations. Rahmandad & Sterman (2008) demonstrate that the
outcomes of equivalent SD and AB models are alike under many conditions. Other authors have come
to the same conclusion by comparing single SD and AB models in the fields of health sciences (Ahmed
et al., 2013; Figueredo et al., 2015; Figueredo and Aickelin, 2011b), economy (Alam Napitupulu, 2014),
transportation (Silva et al., 2011), software development (Cherif and Davidsson, 2010), land use (Haase
and Schwarz, 2009) and education (Thompson and Reimann, 2010).

3.1.2 Stochastic & deterministic phenomena

SD and AB can both model deterministic systems: systems which do not contain randomness and thus
yield the same result from a given initial state (Brock, 1986). However, in AB models, decision rules,
actions and properties are normally derived from distribution functions, and are therefore probabilistic
(Bonabeau, 2002).
Due to its stochastic character, the AB paradigm can naturally account for outlier values that
would not be shown in an aggregated system representation. Outlier values represent random events,
such as Black Swans, that are unlikely but can alter the system radically. Therefore, when assumptions
of homogeneity and perfect mixing can be made for a particular study, SD and AB can produce
outcomes that are not statistically different (Ahmed et al., 2013; Rahmandad and Sterman, 2008).
However, when heterogeneous clustered agent networks are central for answering the problem, AB is
usually a more appropriate paradigm to study the problem (Rahmandad and Sterman, 2008).
However, there is a trade-off between the stochasticity of an AB model and its computational
requirement (Osgood, 2007; Rahmandad and Sterman, 2008). A conflict in goals arises between the
richness of feedback structure captured endogenously, the number of agents and their complexity of
interaction, and the exhaustiveness of the sensitivity analysis (Rahmandad & Sterman, 2008). As a
result, AB can be discarded as the preferred method in modeling and simulation studies due to its high
computational resource demands (Ahmed, Greensmith, & Aickelin, 2013; Figueredo & Aickelin, 2011;
Figueredo et al., 2011; Figueredo et al., 2015; Silva, Coelho, Novaes, & Lima Jr, 2011).

-4-
3.1.3 Physical space, topographies, & network structures
Inherently, SD was not designed to cope with spatial diffusion and propagation processes, but to model
the aggregate properties of such systems and so provide strategic insight into their behaviour. When the
number of entities is small and when the entities are highly dispersed or clustered, this can be
problematic (Rahmandad & Sterman, 2008). Emerging paradigms, such as spatial system dynamics
(SSD), are trying to overcome this limitation (Ahmad and Simonovic, 2004; Neuwirth and Peck, 2013).
SSD is based on coupling SD with geographic information systems (GIS) to provide feedback effects
across physical space (Ahmad and Simonovic, 2004).
In contrast, AB has the capability to distinguish physical space, topographies, and other network
structures (Bonabeau, 2002; Parunak et al., 1998; Rahmandad & Sterman, 2008). The former allows
the explicit study of the dynamics across landscapes or networks (Osgood, 2007). Hence, AB models
have proven attractive for classes of modelling problems where topographies (particularly irregular and
clustered) are crucial with respect to understanding the problem and the assessment of policies.
Furthermore, the characteristics of mobile agents in a network, able to alter system structure, can be
utilized to account generally for evolving systems in which relations disintegrate and are created
dynamically over time (Scholl, 2001). This property and the possibility to construct goal-oriented agents
makes AB models ideally suited to model many social systems and implement concepts from social
and behavioral science such as bounded rationality (Edmonds, 1999; Manson, 2006; March and Simon,
1958).

3.1.4 Learning & adaption processes

Experience based learning effects and adaptation processes such as the “eroding quality standards”
archetype are frequently modelled in SD.. Nevertheless, explicit individual learning and adoption
processes are a focus within AB models (Bonabeau, 2002). For this, machine learning algorithms are
used to design agents that have the ability to modify their own decision rules (Parunak et al., 1998;
Phelan, 1999; Scholl, 2001; Stone and Veloso, 2000).

3.1.5 Ease of model building and interpretation

As the previous examples demonstrate, AB model have numerous virtues in specific contexts. However,
these virtues often come at the cost of more time consuming modeling simulation and interpretation
processes (Osgood, 2007). Indeed, the interpretation of AB model outputs at aggregate level is still in
its infancy. Whereas the formulation of an SD model makes use of system level observables to identify
the feedback loops that govern the system’s behavior (Rahmandad & Sterman, 2008), the construction
of an AB model requires not only knowledge of the system at an aggregated level, but also in-depth
insights on decision processes of agents and their behavior (Macal, 2010; Macal & North, 2006).
Moreover, AB models require knowledge on the disaggregated distributions of agent properties for
parametrization (Macal, 2010; Macal & North, 2006).
Next, as described in the paragraphs devoted to stochastic and deterministic phenomena, AB
models have considerably longer simulation times than their SD counterparts.
Additionally, the interpretation of simulation results is typically easier for SD models than for
AB models, because the underlying dynamics of these models are transparent and the toolbox for
analyzing and understanding simulation results is already well developed. This availability of methods
facilitates the rapid development of small models to explore the driving dynamics of current ‘hot’ issues
(Pruyt, 2013; Pruyt et al., 2009).
Finally, SDs popularity has been facilitated by the availability of several drag-and-drop software
tools for constructing and analyzing models, including Vensim® (www.vensim.com), Stella®
(www.stella.com) and PowerSim® (www.powersim.com) (Borshchev and Filippov, 2004). Until
recently, one of the obstacles for wider adoption of AB had been the limited availability of easy to use
-5-
tools that do not require programming skills (Parunak et al., 1998; Wilensky, 1999). However, the
emergence of software such as AnyLogic® (“Multimethod Simulation Software and Solutions,” n.d.)
and NOVA® (Salter, 2013) may facilitate faster adoption.

3.2 Potential benefits of combining SD and AB

Despite fundamental differences, both modelling approaches are effective in describing and simulating
complex dynamic systems. AB has received increasing attention because it holds promise of significant
benefits compared with other modeling paradigms, including SD (Bonabeau, 2002; Epstein, 2006;
Jennings et al., 1998; Macal and North, 2006). Nevertheless, there is no simple dividing line indicating
when which modelling approach will provide superior results. Instead, the choice of paradigm depends
on the problem and the purpose of the modelling exercise and should take into account the paradigms’
capabilities, limitations and tradeoffs (Figueredo and Aickelin, 2011b; Parunak et al., 1998).
By combining SD and AB, some components can be modelled discretely and in a disaggregated
fashion when this is needed, while other components can be modelled continuously and in an aggregated
fashion, based on the different system characteristics and the specific model purpose (Osgood, 2007).
In this way, a hybrid SD-AB model facilitates the definition of appropriate levels of aggregation for
each component of the system. Furthermore, for many modelling problems, a combination of SD and
AB can reduce computation times, provide the strategic overview characteristic of SD, while still
capturing relevant elements of the individual heterogeneity and stochasticity of entities and processes.
Another potential advantage of combining SD and AB is that this can be seen as a way to enhance
the capability of SD models to cope with spatially explicit problems. The resulting models permit
arranging agents in a spatial or network structure, while integrating important properties of SD, such as
continuity and non-linear multi-loop feedback. This approach can be refined when the individuals are
mobile and consequently the spatial dimension becomes dynamic. Besides this, it is possible to use
multiple SD sub-models to create different properties across a spatial grid. As a result, individuals
interact with a different SD sub-models depending on their position (Vincenot et al., 2011). Agents can
plausibly even interact with more than one SD sub-model at a time.

4 Recent efforts to integrate SD and AB

While no unified definition exists for hybrid SD-AB models, countless architectures are possible for
coupling or matching SD and AB. This section discusses first, how AB features have been incorporated
through emulation into the field of SD. Then, it presents three classifications of possible architectures
for hybrid SD-AB models. Finally, it summarizes recent efforts and breakthroughs in the design of
hybrid SD-AB models, and sketches the state-of-the-art of SD and AB integration. The focus lies on
work conducted within the last decade and particularly in the last five years.

4.1 Emulation of AB features within the SD field

In the field of SD, some authors have made attempts to emulate the capabilities of AB without changing
the overall SD paradigm. Pasaoglu et al. (2016), Powell and Coyle (2005) and Wu, Kefan, Hua, Shi,
and Olson (2010), for instance, integrated an AB perspective in the construction of an SD model. Teose
et al. (2011) embedded SD notions into AB models using Gillespie’s τ-leap algorithm, an equation that
connects the paradigms by interpreting rates of flow into movement of agents.
While paradigm emulation has contributed to expanding the applications of SD during the last
few years, it is the appropriate combination of the two approaches that has become the most promising
research line (C Swinerd & McNaught, 2014). Integrating SD and AB makes it possible to avoid their
individual pitfalls and, hence, to exploit the full potential of their complementary characteristics, so as

-6-
to provide a more complete representation of complex dynamic systems (Scholl, 2001; Stemate et al.,
2007).

4.2 Hybrid SD-AB architectures

Swinerd and McNaught (2012), Kirandeep et al. (2013) and Vincenot et al. (2011) proposed different
architectures for hybrid SD-AB models.
Based on Shanthikumar and Sargent (1983), Swinerd and McNaught (2012) presented three
classes that vary depending on how the model’s modules, either SD or AB single paradigm meta-
models, interact to produce the model’s outcome. In the first class, the sequential class, the outcome of
each module forms the input for the next module; the outcome of the final module represents the
model’s outcome. The second class, the interfaced class, includes non-sequential combinations of
modules that do not influence each other but combine their independent outcomes to produce the model
outcome. Lastly, in the integrated class, modules and even model outcomes provide feedback to one
another.
The second classification, developed by Kirandeep et al. (2013), presents two classes that are
analogous to the aforementioned sequential and integrated ones.
Vincenot et al. (2011), in turn, identified four reference cases or typical SD-AB structures. Case
1 refers to AB agents interacting within their environment, an SD module. Emergent properties from
the AB module can dynamically parameterize the SD module. Case 2 refers to AB agents containing
SD modules that determine their dynamic decision rules and spatial structures. In Case 3, individuals
interact with an environment made of more than one SD module. Unlike Case 1, Case 3 is spatially
explicit and the SD module with which an agent interacts depends on the agent’s position and the SD
module’s area of influence. Finally, Case 4 refers to SD-ABM model swapping. This case reduces
computation time by allowing only modules of the same paradigm to run at any given time. During the
run, threshold values or events cause the change from modules of one paradigm to modules from the
other one.
However, the architectures of Swinerd and McNaught (2012), Kirandeep et al.’s (2013), and
Vincenot et al. (2011) are non-exhaustive in nature. While Chris Swinerd and McNaught (2012)’s
interfaced class implies that modules in a hybrid model do not necessarily have to be connected during
the simulation, all the reference cases by Vincenot, Giannino, Rietkerk, Moriya, and Mazzoleni (2010)
consider interaction between the modules during the simulation. In practice, the architecture of hybrid
SD-AB models is usually based on the specific needs of the problem under study. Examples are
provided in the following sub-section.

4.3 Recent hybrid SD-AB models and modeling environments

Hybrid SD-AB models have proven useful in studying diffusion processes of technological innovation.
In their independent studies, Swinerd and McNaught (2014) and Shafiei et al. (2013) embedded
individual agents in an SD environment. In Swinerd and McNaught’s model, an SD module is
embedded in each agent to dynamically parameterize its properties. Similarly, Swinerd and McNaught
(2015) simulated the international diffusion of consumer technology by modeling nations as agents,
with internal decision processes consisting of SD models, and global diffusion processes with an
equation-based rate model.
Hybrid SD-AB models have also been developed in other fields. Jo et al. (2015) designed a
dynamic alternative to cost benefit analysis for infrastructure projects. This work integrates AB and SD
modules by enabling dynamic feedback from the SD states to the AB environment, and from the AB
environment to the SD rates of change. Tran (2016) developed a multi-paradigm framework to analyze
techno-behavioral dynamics in networks, and to assess the impact of technology on society. This
framework integrates the notions of system dynamics to explore the most aggregated and macro layers
-7-
of the system, and notions from agent-based to study network structures and individual behavior. Lewe
et al. (2014) studied intercity transportation by integrating SD and AB modules to represent macro-
level, and micro-level variables, respectively. Kolominsky-Rabas et al. (2015) developed the
framework ProHTA, a hybrid SD-AB tool the aim of which is to the assessment of innovative health
technologies prior to their launch.
Other examples explicitly include discrete event simulation models, in addition to the SD and
AB components. For instance, a study of the elements of a hybrid simulation model for blood supply
chains (Onggo, 2015); a feasibility assessment of hybrid approaches in the context of complex
healthcare operation management (Viana, 2014); an analysis of real workforce choices (Flynn et al.,
2014); a hybrid approach to integrate safety behaviour into construction planning, by Goh and Askar
Ali (n.d., in press) and the study of reusability in hybrid simulation by Djanatliev et al. (2014), to
mention but a few.
The availability of modelling environments that can handle multiple paradigms, including SD, AB
and discrete events has also increased. For instance, Salter (2013) reports on NOVA®, a modeling and
simulation platform that supports the integration of both paradigms. Moreover, this work envisions the
integration of Geographic Information Systems (GIS) within the platform. Other platforms include
Anylogic® (“Multimethod Simulation Software and Solutions,” n.d.), which supports modeling and
simulation with SD, AB, discrete events and incorporates certain GIS features, as well as NetLogo
(Wilensky, 1999), a free and open source modeling environment with similar capabilities.

4.4 Exploring the next generation of hybrid paradigms

As explained in the previous Sections, integrating the SD and AB paradigms is a promising approach
to overcome the limitations of each single paradigm. However, the integration of SD and AB is only a
piece in a bigger puzzle (Pruyt, 2015). Recent innovations suggest that, in the future, mainstream
research frameworks and methods to model complex dynamic systems will reach beyond the boundaries
of SD, AB, and even beyond the reach of hybrid SD-AB paradigms.
Currently, the adoption and diffusion of methods and techniques from other disciplines, such as
data analytics and machine learning, are turning modeling and simulation into an interdisciplinary field
(Pruyt, 2015). This process of blending tools and methods across disciplines, which has just started, is
enabling the emergence of a new generation of computational models with radically expanded
capabilities that promise to deliver significant breakthroughs.
For several reasons, the development of this new generation of computational models is likely to
occur using high-level programming language, such as Python, R Project and Java, instead of
commercial and closed source modeling environments (Pérez, Granger, & Hunter, 2011). First, many
scientific disciplines use these languages for scientific computing and quantitative data analysis. The
open source environment fosters transparency and reproducibility of research, while these languages
facilitate the balance between full flexibility of general-purpose programming languages and ease of
use. In addition their object-orientation supports the implementation of multi-model approaches.
Examples of the methodological innovations that will lead to the new generation of models
include Exploratory Model Analysis (EMA) (Kwakkel and Pruyt, 2015, 2013) and data analytics using
tools such as PySD (Houghton, and Siegel, 2015).

5 Conclusion
This literature review seeks to update the work of Scholl (2001) and Macal (2010) by providing an
overview of attempts to integrate SD and AB over the preceding decade, with a particular focus on the
last five years. The review described the building blocks of both paradigms and contrasted their
-8-
capabilities to explore how their integration can yield insights that cannot be generated with one
methodology alone. Five fundamental characteristics in which SD and AB differ were identified. These
characteristics are the paradigms’ capacity to model continuous aggregated and discrete disaggregated
system states; physical space, topographies, and network structures; stochastic & deterministic
phenomena; learning and adaption; and ease of model building and interpretation.
This article also provided an overview of recent work on the integration of SD and AB paradigms,
and the development of multi-paradigm and multidisciplinary modeling and simulation frameworks.
However, he unique contribution of this paper is the conclusion that while paradigm emulation has
contributed to expanding the applications of SD, the dynamic combination of the two approaches is the
most promising research line. Integrating SD and AB, as well as tools and methods from other
disciplines, makes it possible to avoid their individual pitfalls and, hence, to exploit the full potential of
their complementary characteristics, to provide more complete representations of complex dynamic
systems.
Ultimately, the widespread adoption of hybrid SD-AB models will depend on the development
of tools that are able to effectively integrate different modelling paradigms. Therefore, an area of
research that should be encouraged is the development and refinement of free and open source hybrid
modelling tools that they are easy to use and in which models can be documented.
Furthermore, this review concludes that although SD and AB are only a piece in the bigger puzzle
of innovative modeling and simulation environments, their integration into hybrid models plays an
important role in these exciting times. Breakthroughs in the integration of SD and AB can yield insights
in how to build and use smarter modeling tools to support decision-making.

References
Ahmad, S., Simonovic, S.P., 2004. Spatial System Dynamics: New Approach for Simulation of Water
Resources Systems. J. Comput. Civ. Eng. 18, 331–340. doi:10.1061/(ASCE)0887-
3801(2004)18:4(331)
Ahmed, A., Greensmith, J., Aickelin, U., 2013. Variance in System Dynamics and Agent Based
Modelling Using the SIR Model of Infectious Disease. ArXiv Prepr. ArXiv13072001.
Alam Napitupulu, T., 2014. Agent based solution of system dynamics simulation modeling: a case of
rice stock by the National Logistics Agency of Indonesia. J. Theor. Appl. Inf. Technol. 62.
Behdani, B., 2012. Evaluation of Paradigms for Modeling Supply Chains As Complex Socio-technical
Systems, in: Proceedings of the Winter Simulation Conference, WSC ’12. Winter Simulation
Conference, Berlin, Germany, pp. 413:1–413:15.
Bonabeau, E., 2002. Agent-based modeling: Methods and techniques for simulating human systems.
Proc. Natl. Acad. Sci. 99, 7280–7287. doi:10.1073/pnas.082080899
Borshchev, A., Filippov, A., 2004. From System Dynamics and Discrete Event to Practical Agent Based
Modeling: Reasons, Techniques, Tools. Presented at the The, in: July 25 - 29, 2004. Presented
at the 22nd International Conference of the System Dynamics Society., Oxford, England.
Brock, W.A., 1986. Distinguishing random and deterministic systems: Abridged version. J. Econ.
Theory 40, 168–195. doi:10.1016/0022-0531(86)90014-1
Cherif, R., Davidsson, P., 2010. Software development process simulation: multi agent-based
simulation versus system dynamics, in: Multi-Agent-Based Simulation X. Springer, pp. 73–85.
Djanatliev, A., Bazan, P., German, R., 2014. Partial paradigm hiding and reusability in hybrid
simulation modeling using the frameworks Health-DS and i7-AnyEnergy. IEEE, pp. 1723–
1734. doi:10.1109/WSC.2014.7020022
Edmonds, B., 1999. Modelling Bounded Rationality in Agent-Based Simulations Using the Evolution
of Mental Models, in: Brenner, T. (Ed.), Computational Techniques for Modelling Learning in
Economics, Advances in Computational Economics. Springer US, pp. 305–332.
-9-
Epstein, J.M., 2006. Generative Social Science: Studies in Agent-Based Computational Modeling, STU
- Student edition. ed. Princeton University Press.
Epstein, J.M., Axtell, R.L., 1996. Growing Artificial Societies: Social Science from the Bottom Up
[WWW Document]. MIT Press. URL https://mitpress.mit.edu/books/growing-artificial-
societies (accessed 1.1.16).
Figueredo, G.P., Aickelin, U., 2011a. Comparing System Dynamics and Agent-based Simulation for
Tumour Growth and Its Interactions with Effector Cells, in: Proceedings of the 2011 Summer
Computer Simulation Conference, SCSC ’11. Society for Modeling & Simulation International,
Vista, CA, pp. 52–59.
Figueredo, G.P., Aickelin, U., Siebers, P.-O., 2011b. Systems dynamics or agent-based modelling for
immune simulation?, in: Artificial Immune Systems. Springer, pp. 81–94.
Figueredo, G.P., Siebers, P.-O., Aickelin, U., Whitbrook, A., Garibaldi, J.M., 2015. Juxtaposition of
System Dynamics and Agent-Based Simulation for a Case Study in Immunosenescence. PloS
One 10.
Flynn, T., Tian, Y., Masnick, K., McDonnell, G., Huynh, E., Mair, A., Osgood, N., 2014. Discrete
choice, agent based and system dynamics simulation of health profession career paths, in:
Simulation Conference (WSC), 2014 Winter. Presented at the Simulation Conference (WSC),
2014 Winter, pp. 1700–1711. doi:10.1109/WSC.2014.7020020
Forrester, J.W., 1958. Industrial dynamics: a major breakthrough for decision makers. Harv. Bus. Rev.
36, 37–66.
Goh, Y.M., Askar Ali, M.J., n.d. A hybrid simulation approach for integrating safety behavior into
construction planning: An earthmoving case study. Accid. Anal. Prev.
doi:10.1016/j.aap.2015.09.015
Haase, D., Schwarz, N., 2009. Simulation models on Human Nature interactions in urban landscapes:
A review including spatial economics, system dynamics, cellular automata and agent-based
approaches. Living Rev. Landsc. Res. 3, 1–45.
Houghton, J., Siegel, M., 2015. Advanced data analytics for system dynamics models using
PySD, in: July 19 - July 23, 2015. Presented at the 33rd International Conference of the System
Dynamics Society, Cambridge, USA.
Jennings, Sycara, K., Wooldridge, M., 1998. A Roadmap of Agent Research and Development. Auton.
Agents Multi-Agent Syst. 1, 7–38. doi:10.1023/A:1010090405266
Jo, H., Lee, H., Suh, Y., Kim, J., Park, Y., 2015. A dynamic feasibility analysis of public investment
projects: An integrated approach using system dynamics and agent-based modeling. Int. J. Proj.
Manag. 33, 1863–1876.
Kolominsky-Rabas, P.L., Djanatliev, A., Wahlster, P., Gantner-Bär, M., Hofmann, B., German, R.,
Sedlmayr, M., Reinhardt, E., Schüttler, J., Kriza, C., 2015. Technology foresight for medical
device development through hybrid simulation: The ProHTA Project. Technol. Forecast. Soc.
Change 97, 105–114. doi:10.1016/j.techfore.2013.12.005
Kwakkel, J.H., Pruyt, E., 2015. Using System Dynamics for Grand Challenges: The ESDMA Approach.
Syst. Res. Behav. Sci. 32, 358–375. doi:10.1002/sres.2225
Kwakkel, J.H., Pruyt, E., 2013. Exploratory Modeling and Analysis, an approach for model-based
foresight under deep uncertainty. Technol. Forecast. Soc. Change 80, 419–431.
doi:10.1016/j.techfore.2012.10.005
Lättilä, L., Hilletofth, P., Lin, B., 2010. Hybrid simulation models–when, why, how? Expert Syst. Appl.
37, 7969–7975.
Lewe, J.-H., Hivin, L.F., Mavris, D.N., 2014. A multi-paradigm approach to system dynamics modeling
of intercity transportation. Transp. Res. Part E Logist. Transp. Rev. 71, 188–202.
doi:10.1016/j.tre.2014.09.011
Macal, C.M., 2010. To agent-based simulation from System Dynamics, in: Simulation Conference
(WSC), Proceedings of the 2010 Winter. Presented at the Simulation Conference (WSC),
Proceedings of the 2010 Winter, pp. 371–382. doi:10.1109/WSC.2010.5679148
Macal, C.M., North, M.J., 2006. Tutorial on Agent-Based Modeling and Simulation PART 2: How to
Model with Agents, in: Simulation Conference, 2006. WSC 06. Proceedings of the Winter.

- 10 -
 Presented at the Simulation Conference, 2006. WSC 06. Proceedings of the Winter, pp. 73–83.
doi:10.1109/WSC.2006.323040
Manson, S.M., 2006. Bounded rationality in agent-based models: experiments with evolutionary
programs. International Journal of Geographic Information Science 20, 991– 1012.
March, J.G., Simon, H.A., 1958. Organizations. Wiley.
Meadows, D., 2009. Leverage Points: Places to Intervene in a System. Donella Meadows Inst. 1, 41–
49.
Meadows, D.H., Robinson, J.M., 2002. The electronic oracle: computer models and social decisions.
Syst. Dyn. Rev. 18, 271–308. doi:10.1002/sdr.239
Milling, P., Schieritz, N., 2003. Modeling the Forest or Modeling the Trees - A Comparison of System
Dynamics and Agent-Based Simulation, in: Proceedings of the 21st International Conference
of the System Dynamics Society : July 20 - 24, 2003, New York City, USA. System Dynamics
Soc., Albany, NY, pp. 1–15.
Multimethod Simulation Software and Solutions [WWW Document], n.d. URL
http://www.anylogic.com/ (accessed 3.22.16).
Neuwirth, C., Peck, A., 2013. A conceptual spatial system dynamics (SSD) model for structural changes
in grassland farming. Presented at the 20th International Congress on Modelling and
Simulation, 1–6 December, Adelaide, Australia.
Onggo, B.S., 2015. Elements of a hybrid simulation model: A case study of the blood supply chain in
low- and middle-income countries. Presented at the Proceedings - Winter Simulation
Conference, pp. 1597–1607. doi:10.1109/WSC.2014.7020011
Osgood, N., 2007. Using Traditional and Agent Based Toolsets for System Dynamics: Present
Tradeoffs and Future Evolution. Presented at the The 2007 International Conference of the
System Dynamics Society.
Parunak, H.V.D., Savit, R., Riolo, R.L., 1998. Agent-Based Modeling vs. Equation-Based Modeling:
A Case Study and Users’ Guide, in: Sichman, J.S., Conte, R., Gilbert, N. (Eds.), Multi-Agent
Systems and Agent-Based Simulation, Lecture Notes in Computer Science. Springer Berlin
Heidelberg, pp. 10–25.
Pasaoglu, G., Harrison, G., Jones, L., Hill, A., Beaudet, A., Thiel, C., 2016. A system dynamics based
market agent model simulating future powertrain technology transition: Scenarios in the EU
light duty vehicle road transport sector. Technol. Forecast. Soc. Change 104, 133–146.
doi:10.1016/j.techfore.2015.11.028
Phelan, S.E., 1999. A Note on the Correspondence Between Complexity and Systems Theory. Syst.
Pract. Action Res. 12, 237–246. doi:10.1023/A:1022495500485
Pruyt, E., 2015. From Building a Model to Adaptive Robust Decision Making Using Systems Modeling,
in: Janssen, M., Wimmer, M.A., Deljoo, A. (Eds.), Policy Practice and Digital Science, Public
Administration and Information Technology. Springer International Publishing, pp. 75–93.
Pruyt, E., 2013. Small System dynamics models for big issues : triple jump towards real-world
complexity. Delft University of Technology, Delft, The Netherlands.
Pruyt, E., Slinger, J.H., van Daalen, C.E., Yucel, G., Thissen, W., 2009. Hop, Step, Step and Jump
Towards Real-World Complexity @ Delft University of Technology. Presented at the
Proceedings of the 27th International Conference of the System Dynamics Society,
Albuquerque, USA., pp. 1–9.
Rahmandad, H., 2004. Heterogeneity and network structure in the dynamics of contagion: comparing
agent-based and differential equation models. Presented at the 22nd International Conference
of the System Dynamics Society., System Dynamics Society, Oxford, England.
Rahmandad, H., Sterman, J., 2008. Heterogeneity and Network Structure in the Dynamics of Diffusion:
Comparing Agent-Based and Differential Equation Models. Manag. Sci. 54, 998–1014.
doi:10.1287/mnsc.1070.0787
Salter, R.M., 2013. Nova: A modern platform for system dynamics, spatial, and agent-based modeling.
Procedia Comput. Sci. 18, 1784–1793.
Schieritz, N., Grobler, A., 2003. Emergent structures in supply chains-a study integrating agent-based
and system dynamics modeling, in: System Sciences, 2003. Proceedings of the 36th Annual
Hawaii International Conference on. IEEE, p. 9–pp.

- 11 -
Scholl, H.J., 2001. Agent-based and system dynamics modeling: a call for cross study and joint research.
Presented at the System Sciences, 2001. Proceedings of the 34th Annual Hawaii International
Conference on, IEEE, p. 8 pp.
Schryver, J., Nutaro, J., Shankar, M., 2015. Emulating a System Dynamics Model with Agent-Based
Models: A Methodological Case Study in Simulation of Diabetes Progression. Open J. Model.
Simul. 03, 196–214. doi:10.4236/ojmsi.2015.34019
Shafiei, E., Stefansson, H., Asgeirsson, E.I., Davidsdottir, B., Raberto, M., 2013a. Integrated agent-
based and system dynamics modelling for simulation of sustainable mobility. Transp. Rev. 33,
44–70.
Shafiei, E., Stefansson, H., Asgeirsson, E.I., Davidsdottir, B., Raberto, M., 2013b. Integrated agent-
based and system dynamics modelling for simulation of sustainable mobility. Transp. Rev. 33,
44–70.
Silva, V.M.D., Coelho, A.S., Novaes, A.G., Lima Jr, O.F., 2011. Remarks on collaborative maritime
transportation’s problem using system dynamics and agent based modeling and simulation
approaches, in: Adaptation and Value Creating Collaborative Networks. Springer, pp. 245–252.
Stemate, L., Pasca, C., Taylor, I., 2007. A comparison between system dynamics and agent based
modeling and opportunities for cross-fertilization, in: Simulation Conference, 2007 Winter.
Presented at the Simulation Conference, 2007 Winter, pp. 2376–2376.
doi:10.1109/WSC.2007.4419891
Sterman, J.D., 2000. Business Dynamics. Systems Thinking and Modeling for a Complex World,
Har/Cdr. ed. Mcgraw-Hill Education Ltd, Boston.
Stone, P., Veloso, M., 2000. Multiagent Systems: A Survey from a Machine Learning Perspective.
Auton. Robots 8, 345–383. doi:10.1023/A:1008942012299
Swinerd, C., McNaught, K.R., 2012. Design classes for hybrid simulations involving agent-based and
system dynamics models. Simul. Model. Pract. Theory 25, 118–133.
doi:10.1016/j.simpat.2011.09.002
Swinerd, C., McNaught, K.R., 2014. Simulating the diffusion of technological innovation with an
integrated hybrid agent-based system dynamics model. J. Simul. 8, 231–240.
Swinerd, C., McNaught, K.R., 2015. Comparing a simulation model with various analytic models of
the international diffusion of consumer technology. Technol. Forecast. Soc. Change 100, 330–
343. doi:10.1016/j.techfore.2015.08.003
Teose, M., Ahmadizadeh, K., O’Mahony, E., Smith, R.L., Lu, Z., Ellner, S.P., Gomes, C., Grohn, Y.,
2011. Embedding system dynamics in agent based models for complex adaptive systems.
Presented at the IJCAI Proceedings-International Joint Conference on Artificial Intelligence,
Citeseer, p. 2531.
Thompson, K., Reimann, P., 2010. Patterns of use of an agent-based model and a system dynamics
model: The application of patterns of use and the impacts on learning outcomes. Comput. Educ.
54, 392–403.
Tran, M., 2016. A general framework for analyzing techno-behavioural dynamics on networks.
Environ. Model. Softw. 78, 225–233. doi:10.1016/j.envsoft.2015.12.004
Viana, J., 2014. Reflections on two approaches to hybrid simulation in healthcare. IEEE, pp. 1585–
1596. doi:10.1109/WSC.2014.7020010
Vincenot, C.E., Giannino, F., Rietkerk, M., Moriya, K., Mazzoleni, S., 2011. Theoretical considerations
on the combined use of System Dynamics and individual-based modeling in ecology. Ecol.
Model. 222, 210–218. doi:10.1016/j.ecolmodel.2010.09.029
Wakeland, W.W., Gallaher, E.J., Macovsky, L.M., Aktipis, C., 2004. A comparison of system dynamics
and agent-based simulation applied to the study of cellular receptor dynamics. Presented at the
System Sciences, 2004. Proceedings of the 37th Annual Hawaii International Conference on,
IEEE, p. 10 pp.
Wilensky, U., 1999. Center for Connected Learning and Computer-Based Modeling. NetLogo.

Ahmed, A., Greensmith, J., & Aickelin, U. (2013). Variance in System Dynamics and Agent Based Modelling
Using the SIR Model of Infectious Disease. arXiv preprint arXiv:1307.2001.

- 12 -
Angerhofer, B. J., & Angelides, M. C. (2000). System dynamics modelling in supply chain management: research
review. Paper presented at the Simulation conference, 2000. proceedings. winter.
Chen, Y. (2011). Understanding technology adoption through system dynamics approach: A case study of RFID
technology. Paper presented at the Embedded and Ubiquitous Computing (EUC), 2011 IFIP 9th
International Conference on.
Demarest, J. B. (2011). An Agent-based Model of Water and Health in Limpopo Province. University of Virginia.
Figueredo, G. P., & Aickelin, U. (2011). Comparing system dynamics and agent-based simulation for tumour
growth and its interactions with effector cells. Paper presented at the Proceedings of the 2011 Summer
Computer Simulation Conference.
Figueredo, G. P., Aickelin, U., & Siebers, P.-O. (2011). Systems dynamics or agent-based modelling for immune
simulation? Artificial Immune Systems (pp. 81-94): Springer.
Figueredo, G. P., Siebers, P.-O., Aickelin, U., Whitbrook, A., & Garibaldi, J. M. (2015). Juxtaposition of System
Dynamics and Agent-Based Simulation for a Case Study in Immunosenescence. PloS one, 10(3).
Fisher, D. K., Norvell, J., Sonka, S., & Nelson, M. J. (2000). Understanding technology adoption through system
dynamics modeling: implications for agribusiness management. The International Food and
Agribusiness Management Review, 3(3), 281-296.
Georgiadis, P., Vlachos, D., & Iakovou, E. (2005). A system dynamics modeling framework for the strategic
supply chain management of food chains. Journal of food engineering, 70(3), 351-364.
Gjerdrum, J., Shah, N., & Papageorgiou, L. G. (2001). A combined optimization and agent-based approach to
supply chain modelling and performance assessment. Production Planning & Control, 12(1), 81-88.
Kirandeep, C., Eldabi, T., & Young, T. (2013). A conceptual framework for hybrid system dynamics and discrete
event simulation for healthcare. Journal of Enterprise Information Management, 26(1/2), 50-74.
Mellor, J. E., Smith, J. A., Learmonth, G. P., Netshandama, V. O., & Dillingham, R. A. (2012). Modeling the
complexities of water, hygiene, and health in Limpopo Province, South Africa. Environmental science
& technology, 46(24), 13512-13520.
Moser, C. M., & Barrett, C. B. (2006). The complex dynamics of smallholder technology adoption: the case of
SRI in Madagascar. Agricultural Economics, 35(3), 373-388.
Powell, J. H., & Coyle, R. (2005). Identifying strategic action in highly politicized contexts using agent-based
qualitative system dynamics. Journal of the operational research society, 56(7), 787-798.
Schwarz, N., & Ernst, A. (2009). Agent-based modeling of the diffusion of environmental innovations—an
empirical approach. Technological Forecasting and Social Change, 76(4), 497-511.
Shanthikumar, J., & Sargent, R. (1983). A unifying view of hybrid simulation/analytic models and modeling.
Operations research, 31(6), 1030-1052.
Silva, V. M. D., Coelho, A. S., Novaes, A. G., & Lima Jr, O. F. (2011). Remarks on collaborative maritime
transportation’s problem using system dynamics and agent based modeling and simulation approaches
Adaptation and Value Creating Collaborative Networks (pp. 245-252): Springer.
Swinerd, C., & McNaught, K. (2014). Simulating the diffusion of technological innovation with an integrated
hybrid agent-based system dynamics model. Journal of Simulation, 8(3), 231-240.
Swinerd, C., & McNaught, K. R. (2012). Design classes for hybrid simulations involving agent-based and system
dynamics models. Simulation Modelling Practice and Theory, 25, 118-133.
Tako, A. A., & Robinson, S. (2012). The application of discrete event simulation and system dynamics in the
logistics and supply chain context. Decision Support Systems, 52(4), 802-815.
Teose, M., Ahmadizadeh, K., O'Mahony, E., Smith, R. L., Lu, Z., Ellner, S. P., . . . Grohn, Y. (2011). Embedding
system dynamics in agent based models for complex adaptive systems. Paper presented at the IJCAI
Proceedings-International Joint Conference on Artificial Intelligence.
Vincenot, C. E., Giannino, F., Rietkerk, M., Moriya, K., & Mazzoleni, S. (2010). Theoretical considerations on
the combined use of system dynamics and individual-based modeling in ecology. Ecological Modelling,
222(1), 210-218.
Wu, D. D., Kefan, X., Hua, L., Shi, Z., & Olson, D. L. (2010). Modeling technological innovation risks of an
entrepreneurial team using system dynamics: an agent-based perspective. Technological Forecasting and
Social Change, 77(6), 857-869.
Xue, X., Li, X., Shen, Q., & Wang, Y. (2005). An agent-based framework for supply chain coordination in
construction. Automation in construction, 14(3), 413-430.
Zhang, T., & Nuttall, W. J. (2007). An agent based simulation of smart metering technology adoption.

- 13 -