Supporting Material is available for this work. For more information, follow the link from
the Table of Contents to "Accessing Supporting Material".
Table of Contents
Reusability in System Dynamics:
Current Approaches and Improvement Opportunities
Ricardo Sotaquira, Gerly Carolina Ariza Zabala
Systems Thinking Research Group
Universidad Aut6noma de Bucaramanga
Calle 48 Nr.39-234 Bucaramanga, COLOMBIA
Tel. +76 43 61 11 ext. 347
E-mail: rsotaqui@unab.edu.co - gariza@unab.edu.co
http://fis.unab.edu.co/gps/sistemika/index.html
Abstract
Several companies in Colombia, and also some public institutions, are beginning to use formal methodologies for
strategic analysis. Unfortunately, System Dynamics modeling is an almost unknown option for these potential
users. One answer to this poor diffusion is based on the concept of model reusability. It will be revealed, by
means of a framework of levels of reusability derived from the object-oriented approach, that the state of the art
in system dynamics reusability shows the same evolution stage of software reusability before the 1960s. This
interpretation will expose too a major opportunity for encourage the practice of model reuse in our System
Dynamics community and for widening the spectrum of users. A first implementation of a software tool for
system dynamics modeling with higher level of reusability is being developed.
Keywords: __ reusability, model reuse, generic structures, object-oriented approach,
inheritance hierarchy
1. Introduction
Several companies in Colombia, and also some public institutions, are beginning to use formal
methodologies for strategic analysis. Unfortunately, System Dynamics modeling is an almost
unknown option for these potential users. They usually choose a less robust and more
qualitative approach like scenario planning or balanced scorecard. Furthermore the local
community of system dynamics practitioners is quite small.
This situation is very similar to the one presented by Graham Winch (2002) about the barriers
to entry for system dynamics (SD) in small-medium enterprises of industrialized countries. In
both cases, a pertinent contribution could be to offer an alternative where “the participation of
the consultant/ specialist can be partially, or ideally completely, removed” (Winch 2002, p.342).
Winch proposes to use a simulator based on a uset-parameterized generic model. From a wider
perspective, this alternative is just one member of a seties of proposals based on the concept
of model reusability.
From system archetypes to molecules, a history of model reuse has been written in the system
dynamics practice. But this paper will try to prove that the state of model reusability in our
field is similar to the state of software reusability before the 1960s, and that we have to pay
close attention to the concept of reusability, as developed by software engineering, in order to
achieve a methodological and technological improvement and a wider diffusion of SD.
Therefore, the theoretical construction of the reusability concept in object-oriented software
engineering constitutes the first part of the paper’. There is no just one approach to software
reusability, but there is a series of reusability levels developed during the growing of software
engineering field. On the basis of the definition of each level, it will propose a corresponding
reusability level in system dynamics. Later, this interpretive framework about reusability will be
used to examine an state of the art of system dynamics proposals relative to model reuse:
System archetypes, molecules, predefined components, user defined components, etc. Finally,
this interpretation will clearly reveal the limitations of the current meaning of reusability in our
field, and it will expose too a major opportunity for methodological and technological
improvement.
2. Levels of Reusability
As Meyer (1988) pointed out it would be unfair to say that reuse only occurs in software
engineering when object-oriented approach arose. Accordingly, it would be unfair to affirm
that reuse only occurs in SD until the emergency of system archetypes. But these generic
structures constitute one of the first proposals moved by an explicit intention of model reuse.
The reusability concept in software engineering appears when the deliberated intention of code
reuse was formulated and this event occurs in the domain of the object-oriented approach
(OOA).
Reusability concept in OOA has been evolved. From its beginnings on the 1960s to its formal
specification several years later, software reuse has been enhanced its power. In order to offer
an organized view of this progress we propose a framework composed of four levels of
reusability: (1) loosely coupled units of state and behavior, (2) close mapping between
components and problem domain or mental model entities, (3) composability /decomposability
and white-box encapsulation, and (4) inheritance hierarchy and polymorphism. Current
research in software development is still increasing these basic levels with higher ones (i.e.
multiple inheritance), but these four levels are shared by current standards in the field.
2.1. First Level of Reusability: Loosely Coupled Units of State and Behavior
Usually the concept of object, or formally defined the concept of class, in OOA is introduced
as a “solution” to the gap between data and code, or between state and behavior. This
“solution” appears in the context of one of the first computer simulation languages: Simula
(Nygaard 1962; Dahl, Myhrhaug and Nygaard 1968). The first implementation of Simula, in
1962, features an innovation in computer programming. The program was organized in
independent parts that combine data and code: “customers” and “stations”. (Simula was a
programming language oriented to queuing models). But in the second version of Simula, they
“replaced these [parts] with ove concept that could describe all the different ‘objects’ that
should participate in a simulation. This concept was called a ‘process’ and could play both an
active [station] and a passive [customer] role in simulations” (Krogdahl 2003, p.2).
! The research process presented in this paper is based on an interpretive holistic approach: the Interpretative
Systemology (Fuenmayor 1991). Particularly the methodological guidelines of its theory of organizations
(Fuenmayor 2001) have been followed.
The version named Simula 67 changed the concept of process by the concept of class. The
concept of class, the formal definition of an object, was the cornerstone for the formulation of
a software construction methodology based on reusable components, it was the first feature of
software reusability.
By definition, system dynamics transcends the gap between state and behavior. However
system dynamics modeling generally is not oriented to build models by using independent
units. In fact, feedback loops can go through several subsystems or submodels. Therefore one
conceivable difference between OOA concept of class and the corresponding concept in SD is
that the latter sometimes correspond to independent units, but in the general case SD classes
correspond to /oosely coupled units. That is, they would be units with a minimum coupling
between each other, just the minimum interdependence to support feedback loops.
2.2. Second Level of Reusability: Close Mapping between Components and Problem
Domain or Mental Model Entities
One of the early observed advantages of OOA, with respect to previous software development
approaches, was the close relationship between the concept of class/object and the implicit
concept of entity that we commonly use to understand our world. This is the reason why
OOA practitioners suggest that “object-oriented design is a natural approach: the world being
modeled is made of objects ... and it is appropriate to organize the model around computer
representations of these objects” (Meyer 1988, p.51). In a less positivist fashion, we could
affirm that the concept of object allowed to software engineers a one-to-one mapping between
the software model and the mental models about the problem domain.
This feature has promoted software reusability. For example, the entire technology in
Graphical User Interfaces is supported by a set of common notions about human visual
interaction: window, menu, icon, etc. These notions have given to the developers a common
set of objects that have been increasingly enhanced with their programming efforts. The one-
to-one relationship between software object and real/mental entity encourages collective work
around common issues.
In some cases this feature of OOA could be in contradiction with the orientation of SD to
underlying structure and its foundation on the concept of feedback information system.
Sometimes it could not be possible to draw a one-to-one connection between a class (object or
component) in SD and a corresponding mental entity. But to encourage a close, instead of
one-to-one, mapping between both domains, through the concept of SD class, means to
promote model reusability.
problem domain/ ris ~N
mental model ,~ s
4
a nom it \
system dynamics | 3) (2 + mle |
model | , , |
a | se |
i! LOM | oO!
| |
| |
| |
| |
| |
\ |
7
Figure 1. Close mapping between components and problem domain or mental model entities
2.3. Third Level of Reusability: Composability/Decomposability and White-box
Encapsulation
The original version of Simula offered to the user two basic predefined classes, station and
customer, but it was only until Simula 67 when user-defined classes appear. This event opened
a vast land of possibilities for software reuse. The user then could create new classes composed
of simple classes previously defined by the programming language or by himself. On the other
hand, a developer could easily observe the decomposition of a system into its constitutive
components, avoiding extensive changes to the entire system. The formal names of these two
OO properties are composability and decomposability, respectively (Meyer 1988, p.12-13).
These object-oriented attributes are complemented by the concept of encapsulation. Any
component must have a clear defined boundary that separates its state and its behavior from
other components, this is an evident requirement for decomposability. Thus components
resemble capsules. In OOA encapsulation is implemented by applying the principle of
information hiding, for example in Smalltalk, another object-oriented language of the 1960s,
“everything we can describe can be represented by the recursive composition of a single kind
of behavioral building block that hides its combination of state and process inside itself and
can be dealt with only through the exchange of messages” (Kay 1993).
Information hiding and encapsulation usually imply that components are black boxes. But
recently it has been proposed in OOA a white-box approach that also satisfies the
requirements of encapsulation and controlled communication between components. The
white-box approach would be the appropriate alternative for encapsulation in SD classes and it
also fits the Composability/Decomposability conditions.
2.4. Fourth Level of Reusability: Inheritance Hierarchy and Polymorphism
It would seem that an enough level of reusability is achieved when a software technology
satisfies the requirements established by the first three levels of reusability. But “classes, as
seen so far, are not sufficient. They do provide a good modular decomposition technique. [...].
But more is needed to fully achieve the goals of reusability and extendibility. [...]. Progress in
either reusability or extendibility demands that we take advantage of the strong conceptual
relations that hold between classes: a class may be an extension, specialization or combination
of others. [...]. Inheritance provides this support” (Meyer 1988, p. 217).
Observe that inheritance relationships and composition relationships are different. In the latter
ones some classes are integrated into a new and more complex class. This kind of relationship
between superclass and subclasses is the relationship between the whole and its compounding
parts. On the other hand, if a subclass inherits from a superclass it means that the subclass will
extend or specialize the former features of its superclass. Inheritance is a descendant-ancestor
kind of relationship. When a programming language supports inheritance then any updates in
an ancestor are mirrored automatically in every descendant.
The complement for implementing inheritance is the property of polymorphism. It relates to
the ability to replace an instance of an ancestor class by an instance of any of its descendant
classes. For example, an object-oriented organizational system has defined a class operation and
some descendant classes like sales, product design and manufacture. Any reference in the system to
an instance of operation could be rightly satisfied by instances of sa/es or any other descendant.
Inheritance hierarchy and polymorphism introduced a new dimension of reusability, and they
constituted the more distinctive features of reusability in OOA. Classes or components should
be sufficiently generic to have a wide domain of use, but also they should offer a simple way to
adapt them to a particular use. This dilemma is solved by inheritance and polymorphism.
The application of this level of reusability in SD yields the possibility to have generic structures
but, at the same time, to have an efficient technique to fit these structures to particular model
circumstances in a quite more flexible and robust way than the approach of parameterizing.
3. Current Approaches to Reusability in System Dynamics
The interpretive framework about reusability formulated in previous section will be used now
to examine an state of the art of system dynamics proposals related to model reusability.
All of the examined proposals satisfy at least the first level of reusability, that is the capability
of define classes or components. There are other approaches to reusability in SD not oriented
to components, for example parameterization of generic models (Winch 2003). These offer a
more restricted reusability than any level presented here, so they will not be interpreted in this
framework’, The following table summarizes the findings of this interpretive study.
2 Generic models approach also exhibits other shortcomings as Corben et. al. pointed out (1999): a generic model
could not be well fitted to the modeled situation, and the group learning process could be hardly limited by the
reduction of the modeling process.
coupled units
of state and
behavior
generic (abstract)
causal loop
structure.
(abstract) structure
made of stocks, flows
and auxiliary elements.
components are based
on the concept of
generic structures.
Levels of System Predefined Predefined Limited user defined
Reusability/ | Archetypes components components and components
Approaches (i.e. Molecules) limited user defined | (i.e. Powersim Studio
to Reusability components 2003 and Myrtveit)
in SD (i.e. Liehr)
1. Loosely Yes. Archetype: | Yes. Molecule: generic | Yes. Predefined model | Yes. A component is a
model piece that can be
used as a building block
of another component.
2. Close
mapping
between
components
and problem
domain or
mental model
No. Because an
archetype can
involve more
than one
(mental/real)
entity, and the
archetype is
Sometimes the
molecule level can map
one (mental/real)
entity.
Yes. Predefined model
components must be
designed and named to
correspond to entities
of mental model.
Powersim does not
force that components.
correspond to mental
model entities, the
choice depends of the
user.
ility and white-
box
encapsulation
are dissolved
when thi
substructures are
integrated into a
model.
when these
substructures are
integrated into a
model.
the model and they can
be viewed as white
boxes depending on
the level of
aggregation.
entitie: causal-oriented
not entity-
oriented.
3; No. Because the | No. Because the Yes. The predefined Yes. The predefined
Composability | boundaries of | boundaries of the model components can | model components can
/Decomposab | the archetypes _| molecules are dissolved | be distinguished into _ | be distinguished into the
model and they can be
viewed as white box
depending on the level
in the top-down
hierarchy.
4, Inheritance
hierarchy and
Polymorphism
No.
No. The molecules are
organized through a
map that pretends to
contain inheritance
relations. But the
implementation does
not support ancestor-
descendant
relationships, there is
no inheritance
hierarchy.
No. There is
composition hierarchy:
submodels compose
models. This is a
hierarchy of different
levels of detail or
aggregation.
Inheritance hierarchy is
plicitly not included
in this platform.
ex
No. The components
are based in the
different levels of model
structures. This
approach does not
ibit inheritance
hierarchy, but
composition hierarchy.
Table 1. State of the Art in System Dynamics Reusability
The previous table shows that several approaches relate reusability concept with system
dynamics. Some of these approaches incorporate only the first levels of reusability, system
archetypes (Senge 1990) and molecules (Eberlein and Hines 1996) belong to this stage. In
system archetypes, a causal structure cannot be clearly associated with an entity of a mental
model because this structure is an abstract representation of a common behavior then it can
6
simultaneously correspond with different entities. By this reason it could not be achieved a
close mapping between mental model entities and causal structures. The same happens to
molecules, these substructures can be related to different entities at the same time. In both
proposals, when one of these structures is introduced into a model, it loses the possibility to
distinguish itself from rest of the model.
Liehr’ (2002), Myrtveit (2000) and Powersim (2003) have proposed other approaches with a
greater level of reusability. In the theoretical proposal of Liehr the predefined components
would be generic structures that correspond with mental model entities. In the case of Myrtveit
and Powersim the decision to establish the correspondence between mental entities with
components depends on the modeler. In both approaches, when this kind of structure is
introduced into a model, this can be recognized from the other structures of the model (white-
box approach), depending on the level of aggregation. Moreover in Powersim these
components can be edited and reformulated on the basis of modeler needs, or new
components can be created from more basic components’,
4. The Elusive Fourth Level of Reusability in System Dynamics
In the following section three proposals of reusability in SD will be closely examined in order
to prove that in spite of all of them refer to concepts like inheritance or polymorphism,
however the fourth level of reusability is still elusive for current approaches in SD.
Eberlein and Hines (1996) organize their molecules by a diagram. It draws different types of
relationships between the molecules. Maybe some of these molecular linkages could be
inheritance relationships. But these potential inheritance relations disappear when user is
modeling. Because these relationships belong to the diagram, they are not implemented by
software. Molecules do not satisfy one previously mentioned requirement of any
implementation of inheritance hierarchy, that is, any updates in an ancestor must be mirrored
automatically in every descendant during modeling time.
Powersim (2003) exhibits this latter limitation also. The modeler can build a component as an
extension of another previously defined. But there is no connection between the former
component and the derived one. This is not an ancestor-descendant relationship. On the other
hand, the methodological proposal of Myrtveit suggests that “Polymorphism is achieved
through the component interfaces, as components with equal interface are interchangeable”
(2000, p.1). Later he writes “As an example, a model of a company may have sales channel as a
high-level building block” (ibid, p.5). This notion of polymorphism can work if and only if the
fundamental concept of inheritance is developed. But this is not the case of Myrveit’s paper.
As table 1 shown, both Liehr and Myrtveit/Powersim regard a concept of composition
hierarchy, a different kind of hierarchy than inheritance one. Explicitly Liehr’s platform moves
away from inheritance concept or “principle of heredity” (2002 p.3) as it named.
Arguments presented hete confirm that recent approaches to reusability in SD do not satisfy
the requirements imposed by the fourth level of reusability: inheritance hierarchy and
polymorphism.
5 Currently the proposal of Liehr is theoretical, that is, there is not a matching software tool.
4 Barros, Lima and Horta (1991) proposed a similar approach. He defined a limited concept of class with
parameterization ability, but neither he did apply the inheritance concept.
5. The Potential Future of Reusability in System Dynamics
The situation of reusability in SD is very similar to the conditions in software reusability before
the emergence of Simula language. Simula 67 incorporated all the basic features of reusability
described in section 2. So, it is valid to assert that, the state of the art in SD reusability shows
the same evolution stage of software reusability before the 1960s, in spite of recent interest in
the SD community about this issue.
The availability of a SD modeling technique with inheritance hierarchy and polymorphism
would be considered a major instrament for encourage the practice of model reuse in our
community of practitioners and for widening the spectrum of users. On one hand, the
application of inheritance principles promotes a permanent exercise of generalization-
specialization so model reuse and class design could became usual practices in SD field. On the
other hand, the ancestor-descendant technique is a lot more powerful way of extending and
adapting predefined classes than parameterization technique. The latter one does not allow
adding new features to the predefined classes, so its adaptation ability is quite limited. While
the opposite occurs with inheritance, even a non expert user would be benefited by having a
set of white-box SD classes fully adaptable to their particular modeling needs. Both effects of
reusability through inheritance would yield a powerful alternative for SD modeling as an
alternative for strategic analysis, specially under the previously mentioned circumstances of
Latin-American countries or even those of small-medium enterprises in industrialized nations.
It appears to be a pertinent answer for the scarce diffusion of SD approach in the strategic
management field.
Finally, a team of our research group is currently developing a first implementation of a
software for SD modeling with fourth level of reusability, it is expected to be available in the
mid of 2004.
Acknowledgement
This research project has been funded by Instituto Colombiano pata el desarrollo de la Ciencia
y la Tecnologia “Francisco José de Caldas” (COLCIENCIAS), Universidad Auténoma de
Bucaramanga (UNAB) y Pensemos Ltda. Our thanks to the Engineering team at Pensemos for
their contributions to this project.
References
Bartos M.; Lima C.; Horta G. 2001. From Metamodels to Models: Organizing and Reusing
DomainKnowledge in System Dynamics Model Development, Proceedings of System
Dynamics Conference, Atlanta, Georgia, USA. July. System Dynamics Society.
Corben D.; Stevenson R.; Wolstenholme EF. 1999. Holistic oild field value management: using
system dynamics for ‘intermediate level’ and ‘value-based’ modelling in the oil industry,
Journal of the Operational Research Society, 50: 383-391.
Dahl O; Myhrhaug B; Nygaard K. 1968. SIMULA 67 Common Base Language, Norwegian
Computing Center.
Eberlein RJ, Hines JH. 1996. Molecules for Modelers, Proceedings of the 14th International
System Dynamics Conference (CD-ROM), Cambridge, Massachusetts, USA. July. System
Dynamics Society.
Fuenmayor R.L. 1991. Truth and Openness: An Epistemology for Interpretive Systemology,
Systems Practice, 4(5): 473-490.
Fuenmayor, R. 2001. Interpretando Organizaciones, 1st edn. Universidad de los Andes
(Venezuela).
Kay A. 1993. The Early History of Smalltalk, ACM SIGPLAN Notices, 28(3): 69-95, ACM.
Krogdahl S. 2003. The Birth of Simula, Proccedings of the HiNC 1 Conference, Trondheim,
June. IFIP.
Liehr M. 2002. A Platform for Systems Dynamics Modeling: Methodologies for the Use of
Predefined model Components, Proceedings of the 20th International System Dynamics
Conference(CD-ROM), Palermo, Italy. July. Systems Dynamics Society.
Meyer B. 1988. Object-Oriented Software Construction, 1st edn. Prentice Hall International
(UK) Ltd.
Myrtveit M. 2000. Object Oriented Extensions to System Dynamics, Proceedings of the18th
International System Dynamics Conference(CD_ROM), Bergen, Norway, August. System
Dynamics Society.
Nygaard K. 1962. SIMULA: An Extension of ALGOL to the Description of Discrete-Event
Networks, Proceedings of the IFIP congress 62, Munich, North-Holland Publ.
Powersim. 2003. Powersim Studio Express 2003 Online Help. Powersim AS.
Senge PM. 1990. The Fifth Discipline. Doubleday.
Winch G. 2002. User-parameterised generic models: a solution to the conundrum of modelling
access for SMEs?, System Dynamics Review, 18(3): 339-357.
CC BY-NC-SA 4.0