"Management is in transition, from an art based
experience alone, to a profession based on an underlying structure of principles and science.
an underlying structure of principles and science."
(Jay Forrester)
"System Dynamics is a methodology [...]
that offers faster and more reliable results than
other traditional ways of perceiving reality" (Juan Martín García)
(Juan Martín García)
A New Discipline: System Dynamics
System Dynamics (SD) is a discipline created by Jay Forrester in the 1950s that studies the dynamic behavior of all types of complex systems (commercial, biological, social, psychological, economic, etc.) based on the causal relationships that exist between the elements of a system. To do so, it uses a methodology that allows the creation of computer simulation models to facilitate decision making. Simulation is essential to verify in a practical way the theories and hypotheses established in the models.
SD emphasizes causes and their interrelationships. We live in a complex world where everything is interrelated, where there are cause-effect relationships, but also relationships where the causes may be distant in space and time. Multiple causes can also produce the same effect.
For Forrester, a dynamic system is a set of elements interrelated in such a way that a change in one element produces a change in all the other elements, that is, in the system as a whole. Although the elements are simple, the structure formed by all of them produces complexity, with non-linear behavior. Moreover, the behavior of the system as a whole cannot be explained by the behavior of its parts. Complex systems involve a large number of interrelated variables.
The philosophical, theoretical and technical foundations of SD are to be found in three important disciplines: Cybernetics, Computer Science and General Systems Theory. The first for its key concepts of feedback and control. The second for the concept of computational modeling. The third for systems thinking and its concept of general system. SD is a type of systems thinking.
Characteristics of dynamic systems
Hydrodynamic analogy.
A parallelism is established between dynamic systems and hydrodynamic systems (fluid dynamics), consisting of reservoirs (stocks) intercommunicated by channels, with flows, with or without delay, and with the assistance of exogenous factors. The state of the system at a given instant t is determined by the levels of the reservoirs. A reservoir outside the system is represented by a cloud. The flow can have 3 directions: 1) from one reservoir to another; 2) from a reservoir to a cloud (e.g. dead); 3) from a cloud to a reservoir (e.g. births).
There are valves (faucets) that regulate the flow between two reservoirs. The flows cause the level of the reservoirs to increase or decrease. The decision on the opening level of the valves is made taking into account only the levels of the tanks at each instant t.
SD deals only with systems that evolve continuously and deterministically. Decisions (actions) are made at the external level and are discrete surface phenomena over a continuous internal structure.
Feedback relationships.
The feedback relationships between two elements A and B are of two types:
Positive feedback A →+B, which indicates that A influences B in the same sense, i.e., if A increases, B also increases, and when A decreases, B also decreases. Positive feedback implies continuous growth and progressive deviation from equilibrium. It is the "snowball" effect. Examples: 1) The number of births and population. The more births, the greater the population. And as there is more population, the birth rate increases; 2) The interest on a bank account and the capital of the account. The more interest, the more capital accumulates, which in turn produces more interest.
Negative feedback A →−B, which indicates that A influences B in the opposite direction. When A increases, B decreases, and when A decreases, B increases. Negative feedback is based on a goal, which is to reach equilibrium. Examples: 1) A thermostat that regulates the temperature of a room. The lower the room temperature, the higher the thermostat level, and the higher the room temperature, the lower the thermostat level; 2) A production system between the storage (stock) of a product and the production rate. The lower the stock, the higher the production rate, and the higher the stock, the lower the production rate.
The concept of feedback or circularity is central to cybernetics, but in SD what you have is a complex structure of feedback loops.
A hyperstable system is a system made up of multiple negative feedback loops, in which any action that attempts to modify an element of the system is counteracted, not only by the loop in which that element is located, but by the entire set of negative loops acting in its support.
Causal relationships.
Causal relations are indicated by A→B, which indicates that A determines B. In simple systems, cause and effect are usually close in space and time. In complex systems, cause may be remote in space and time. Correlational relationships are relationships that exist between elements without a direct cause-effect relationship.
Diagrams.
There are two types of diagrams: causal diagrams and flowcharts (or Forrester diagrams).
There are authors who believe that causal diagrams are of little or no use; that the flow diagram is much more explicit and useful, since it clearly shows the existing flows in the system. However, the causal diagram serves as an initial conceptual model and makes it easier to move on to the flow diagram, which is a detailed diagram.
Variables.
The elements of a system are represented by variables. A model is determined by the variables and the relationships between them. Variables are of two types: endogenous and exogenous. In turn, endogenous variables are of 3 types: 1) level variables; 2) flow variables; 3) auxiliary variables. The relationship between variables is complex because there are variables that are not directly linked by a cause-effect relationship.
Level variables are equivalent to the internal state variables of a system. They constitute the memory of a dynamic system. They indicate the level of the reservoirs, and change slowly in response to variations in flow variables. A level variable cannot directly influence another level variable except through flow.
Flow variables (rates) are variables that determine the variations of the level variables. They reflect actions taken in the system that have the effect of varying the tank level. They are associated with the valves of the hydrodynamic analogy. Every reservoir level variable has at least one flow variable associated with it.
A flow variable F is usually expressed as a product: F = Fn·M·N (at all instant t), where Fn is the normal flow, M is a multiplier or factor of the normal flow and N is the associated reservoir level. In general, M is usually the product of several Mi multipliers, each a function of a Vi variable.
The simplest flow is an expression of the form TargetLevel/T, where T is the time required to reach the target (the desired level of a reservoir).
Flow variables depend on level, auxiliary and exogenous variables. The flow unit is LevelUnit/TimeUnit.
Auxiliary variables are variables that represent intermediate steps into which the calculation of a flow variable is decomposed from the values of the levels. They usually represent individual concepts and facilitate the understanding and definition of flow variables.
Exogenous variables are variables whose values are independent of those of the rest of the system. Each value represents an action of the external environment on the system. Exogenous variables appear as factors in the flow variables.
Events.
There can be discrete and continuous events. Discrete events are exogenous events. Continuous events are produced by the different flows that circulate through the channels, increasing or decreasing the deposits.
Delays.
There are temporary communication delays between elements. For example, when the price of a product suddenly decreases, the number of units sold does not increase instantaneously, but there is a delay in the effect, because the perception of the change requires some time (information transmission delay), and there is also a delay in the arrival of the new products on the market (physical transmission delay).
Endogenous changes.
Changes in a system are always internal. The causes are contained within the system structure itself. Exogenous disturbances are information that trigger the internal causes of the system. A SD system is causally closed. There are no causes other than internal ones.
Models
A model is a representation of a real system. The value of a model lies in its ability to provide a greater understanding of the system (at an internal or deep level) than by observing the real system (at an external or surface level), as well as to predict the behavior of the system under different conditions, which serves to make the most appropriate decisions.
A SD model consists of a set of elements and a set of relationships that specify the interactions between the elements. Essential to SD models is the time variable.
A model of a system must reflect the model of the mind. According to Forrester, in his 1961 work "Industrial Dynamics", a mental model is a model that represents in our thinking a real system. Each of us carries a mental model of the world: a set of concepts and relationships with which we internally represent a real system. This model is constantly evolving.
Traditionally, models were of a mathematical type, expressed by a set of differential equations. With the advent of computers, current models are of the computational type.
The development of a SD model requires two figures: 1) the expert, the person knowledgeable about the problem or the real system, who provides the conceptual model; 2) the modeler, the designer of the formal model to be implemented on computer.
Since SD is a generic theory of complex systems, it has been suggested that it could also serve as a model of the mind or as a mathematical foundation for the complexity of the mind.
Systemic archetypes
Systemic archetypes also called "generic structures are generic patterns or models of qualitative behavior that occur in many systems and in different domains. Systemic problems are not unique; there are patterns of behavior that recur. This theme is one of those addressed by Peter Senge in his books "The Fifth Discipline" and "The Fifth Discipline in Practice" [Senge, 1993, 1995].
Systemic archetypes were developed by Innovation Associates in the mid-1980s, although some of these archetypes were already described in the previous two decades by Forrester and other pioneers of systems thinking.
They constitute one of the most important theoretical contributions of systems thinking. These systemic archetypes facilitate the practice of systems thinking. Being qualitative patterns, they are more understandable.
They are states of consciousness associated with special systemic situations. They make it possible to become aware of a systemic problem and correct it by modifying both the system and the associated thought or mental model. Systemic archetypes make explicit many situations that were previously only intuited. The states at which corresponding actions must be applied to overcome systemic problems are called "leverage points".
They are based on the structure "desired state - current state - difference - action". When there is a difference between the current state and the desired state, an action is required to bring the current state closer to the desired state. Actions can be positive feedback (or reinforcing) or negative feedback (compensating). To see the result of the actions, it is necessary to consider that there may be delays between the action and the result of that action.
Several systemic archetypes have now been identified. Many are interrelated. We can mention the following, where the first two archetypes are the most basic archetypes that can appear in a system.
Reinforcing cycle (positive feedback).
A key variable in the system is accelerated up or down.
Compensating cycle (negative feedback).
The system moves directly toward a target, without delay. Or, the system moves in an oscillating fashion, due to the delay, toward a target.
Compensation between process and delay.
An action is performed in order to achieve the desired goal, but apparently no progress is apparent, so more actions of the same type are performed, which are more than necessary. The problem is that one is not aware that there is a delay between the action and the system response. This archetype reveals the essential concept of delay.
Limits of growth or sigmoidal growth.
This is a positive feedback loop that acts initially as a dominant that makes growth exponential. This process encounters limits that produce an exhaustion of the growth process, producing a negative loop that cancels the effects of the previous one, providing stability to the system, bringing it asymptotically closer to a limit value. The growth pattern is sigmoidal and is a curve that has 2 subcurves or subpatterns: 1st) first, an exponential growth; 2nd) a more moderate growth that finally becomes asymptotic. In between there is an inflection point connecting the two subcurves. Examples of systems that exhibit this behavior are: the spread of a rumor, the spread of an infectious disease, the introduction of a new technology, etc.
Addiction.
The actual state of a system is matched to the desired state when an external element is called upon in order to achieve the result more quickly.
Load shifting.
An action is taken to eliminate the symptoms of a problem because a quick, easy and effective solution is needed. The burden of the problem is shifted to this superficial solution, with seemingly positive results, but the underlying problem is not attacked. Over time, a dependency on the symptomatic solution is created and the ability to act for a definitive solution is atrophied.
Special case: shifting the burden toward intervention (toward the external factor).
When burden shifting is based on external intervention, there is symptom relief, but those responsible for the system do not learn to deal with the fundamental, underlying problems.
The system receives help to achieve its desired state from another external system. This external system is autonomous and may not provide help at any given time. The initiative for help is from the external system. In addiction, on the other hand, the initiative is from the system itself.
Goal erosion.
When there is a shift of the burden toward a short-term solution, the fundamental long-term objective deteriorates.
Goal erosion also occurs when achieving the desired state requires consuming a lot of resources or is considered impossible to achieve, so the desired state is rethought by decreasing it or even making it equal to the current state.
Escalation.
Two systems compete, for their welfare is seen to depend on achieving a relative advantage of one over the other. When one gets ahead, the other feels threatened and acts to regain its advantage, which threatens the first, which reacts in the same way, and so on.
Success for the one who succeeSD.
Two systems compete. The greater the success of one system, the more support it gets, with the other being left behind. One success is the engine of more successes.
Resistance to change.
When faced with a novel change, the system's response is one of rejection.
Tragedy of the common ground.
Several systems compete for a limited common resource. At first all goes well, as neeSD and objectives are met. As the resource is depleted, the objectives recede, inducing intensified efforts to obtain more resources. Eventually, tragedy strikes: the resource is exhausted or seriously eroded, with no capacity for regeneration.
Quick fixes that fail.
An effective short-term quick fix has unforeseen long-term consequences.
Counterproductive solutions.
A quick fix is applied to relieve symptoms temporarily. This solution seems to work at first, but the problem reappears later. The same solution that seemed to work at first is reapplied, but the problem reappears, getting progressively worse.
Accidental adversaries.
Several groups feel that they must work together to improve the performance of all of them. Eventually they clash over differences in criteria.
Rapid growth and underinvestment.
Growth is approaching a limit. To prevent growth from slowing, rapid and intense investment is needed. But what is decided is to lower expectations by underinvesting, which leaSD to even lower expectations.
MENTAL vs. System Dynamics
We can compare MENTAL with SD in the following aspects:
Systems thinking.
SD is a type of systems thinking. MENTAL represents systemic thinking par excellence because it is a system −or metasystem−, formed by the primary archetypes and their possible relationships, with which all types of systems can be developed, including SD models.
The true systemic archetypes are the primary MENTAL archetypes that establish the degrees of freedom. Those referred to as "systemic archetypes" are secondary archetypes, derived from the primary ones.
Language and paradigm.
Forrester devised a two-dimensional graphical language for performing model design using blocks (tanks, flows, and valves). Under his direction, a formal language and software (Dynamo) was also created at MIT to develop SD models. But this language is limited to the SD paradigm. In contrast, MENTAL is a general formal language that covers the SD paradigm or any other systemic paradigm, allowing the development of all types of systems.
MENTAL allows to use several paradigms to model a system, not only the hydrodynamic paradigm, which is just another paradigm; it is not a universal paradigm. You can use the functional, relational, object, aspect, agent, etc. models. MENTAL, as such, is a system and a universal paradigm.
Simplicity.
The concepts of SD are simple and easily understood. Therein lies its power and success, for its ease of model design and for using concepts close to the mind. MENTAL also uses simple concepts, but at a higher level of abstraction. The complexity results from the combination of simple elements and relationships.
Model of the mind.
MENTAL is a model of the mind and a universal model (that of possible worlSD) with which any particular system, including SD models, can be modeled. The SD model is a loose and limited model of the mind. In fact, it is a particular paradigm, not a universal paradigm. With MENTAL, the semantic gap between the mental model of a system and the formal model is eliminated.
MENTAL makes it possible to create SD models in a simple, straightforward and natural way. Diagrams may even be unnecessary. MENTAL unifies the conceptual model and the operational model in a single language. It also blurs the distinction between model and program.
Time.
Since time is essential in a SD model, and since time belongs to the physical world −so MENTAL logically does not contemplate it −, we assume that the variable t (time) belongs to the system and is permanently updated from the external medium. That is, t is an element common to the system and the environment.
The system cannot modify the value of t, but it can access its value. In MENTAL, an absolute time delay can be implemented like this:
⟨( Delay(r) = (t1 = t) // value of the start time
(t2 = t1+r) // value of the end time
(c =: (t<t2 → c)) // waiting loop
)⟩
Relations.
MENTAL allows you to easily express all kinSD of relationships between variables, not only causal ones, but also functional, sharing, interlinking, etc. Interrelationships between variables can be specified by means of generic expressions: positive and negative feedback, causal relationships, flows, etc. The value of a variable is always a function of other variables, including conditions. The update is immediate and continuous. You can specify quantities in bins and flows, i.e. including quantities and units, e.g. 514*piece, 12*(piece÷hour), etc.
Other advantages of the MENTAL paradigm over the SD paradigm:
The boundary between system and external system is diluted. In addition, the language of the system and the external environment is the same.
The distinction between endogenous and exogenous variables (such as time) is diluted.
MENTAL allows modeling continuous and discrete systems. It also allows modeling qualitative relationships.
The system itself can make decisions (as in AI), in addition to external decisions.
Fuzzy, modal, generalized logic expressions, etc. can be specified.
Allows you to specify all types of data structures and processes (rules, functions, procedures, etc.).
Allows interactivity with other systems and even the sharing of resources between different systems.
Addenda
History of SD
SD was created at the Sloan School of Management of MIT in the mid-1950s for the understanding and management of industrial processes. The first application (by Forrester, an MIT systems engineer) was the analysis of the industrial structure of an American company (Sprague Electric), a manufacturer of electronic components in which there were puzzling oscillations in orders. Forrester applied operations research techniques and performed simulations using the Monte Carlo method (a statistical method for approximating complex mathematical expressions), but was unable to discover the cause of the oscillations. Finally, he discovered that the cause was a combination of feedback structures and delays in the transmission of information.
Because of its industrial origins, SD was initially named by Forrester "Industrial Dynamics", the title of Forrester's work published in 1961, which is considered the formal start of this new discipline.
In 1971, Forrester published "World Dynamics" and "Urban Dynamics" in 1976, works showing how SD modeling is applicable to social systems and city systems, respectively.
Forrester, along with other personalities, founded in 1968 the Club of Rome, an international organization whose main objective is to raise awareness that the current world system is unsustainable and doomed to collapse.
In the 1970s, a report entitled "The Limits to Growth" was produced, based on the results provided by SD, especially inspired by Forrester's world dynamics model. This report was commissioned by the Club of Rome to MIT, and published in 1972. Its main author was Donella Meadows, a biologist and environmental scientist specializing in SD. It predicted that, under a wide range of scenarios, exponential growth would lead to economic collapse during the 21st century. This report helped to popularize SD worldwide.
Today, SD has a large number of applications. It is used for the analysis and design of all kinSD of complex systems in a wide range of fielSD, such as economics, politics, environment, health, industrial processes, business management, social sciences, security and national defense. SD has become indispensable in decision making in complex systems.
The Systems Dynamics Society is an international non-profit organization dedicated to promoting the dissemination, development and use of SD and systems thinking worldwide. It organizes annual conferences and publishes the journal The System Dynamics Review.
Software for SD
There are various software for SD: simulators and programming languages. The most important ones are:
Dynamo (DYNamic MOdels).
It was the first SD simulation language. Developed at MIT in the late 1950s for mainframe by Jack Pugh, under the direction of Forrester. A personal computer version was created in the early 1980s. Dynamo was used for the simulations that were reflected in the report "The Limits to Growth".
Stella and iThink, from iSee Systems.
Stella is a software for the development of scientific and educational SD simulation models. iThink is oriented to economic and business management models. Both feature an intuitive icon-based graphical interface.
Stella appeared in 1985 for Mac computers and was a real revolution, since it allowed designing SD models visually (by specifying the relationships between elements), instead of using the Dynamo programming language.
PowerSim, by PowerSim Software.
SD simulation software. It is oriented to the business area: financial, customer management, production, human resources, etc. models.
Vensim, from Ventana Systems.
Software for the development of SD simulation models. Models can be built in graphical or text mode.
A personal computer version is available: PLE (Personal Learning Edition). This version is free (it does not expire) and has a software, Model Reader, which allows the distribution of the developed models. This software is used for educational and research purposes at the Sloan School of Management at MIT.
SimCity, from Maxis (a division of Electronic Arts).
It is a strategy simulation game of city building, management and evolution. It was released in 1989. It was designed by Will Wright, who recognized the influence of SD and Forrester's ideas. Subsequently more simulation games appeared in other domains such as SimFarm, SimLife, SimHealth, SimAnt and Civilization. The usefulness of games is recognized, but they must be embedded in a learning environment.
System Dynamics has been popularized primarily through Stella and SimCity.
Meadows, Donella H. Thinking in Systems: A Primer. Chelsea Green Publishing, 2008.
Morecroft, John. Strategic Modelling and Business Dynamics: A Feedback Systems Approach. Wiley, 2007.
Ogata, Katsuhiko. System Dynamics. Prentice Hall, 2003.
Palm III, William. System Dynamics. McGraw-Hill Science/Engineering/Math, 2009.
Randers, Jorgen (ed.). Elements of the System Dynamics Method. Productivity Press, 1980.
Ruth, Matthias; Hannon, Bruce; Forrester, Jay W. Modeling Dynamic Economic Systems (Modeling Dynamic Systems). Springer, 1997.
Senge, Peter M. La Quinta Disciplina. Granica, 1993.
Senge, Peter M. La Quinta Disciplina en la Práctica. Granica, 1995.
Sterman, John; Sterman, John D. Business Dynamics: Systems Thinking and Modeling for a Complex World. McGraw-Hill/Irwin, 2000.
Tessem B.; DaviSDon P.J. Fuzzy system dynamics: an approach to vague and qualitative variables in simulation. System Dynamics Review, vol 10, no. 1, pp. 49-66, Spring 1994.
