In their book The Electronic Oracle: Computer Models and Social Decisions, Donella Meadows and Jennifer Robinson discuss four modeling paradigms used in the study of social systems: system dynamics, econometrics, input-output analysis, and optimization. Here are some excerpts from their descriptions of the system dynamics modeling paradigm:
"System dynamics is a subset of the field of simulation modeling.... "
"Simulation modeling is widely practiced in many traditional disciplines such as engineering, economics, and ecology. Since the formulation of differential equations to simulate the progression of systems through time is nearly a free-form exercise, with very few paradigmatic constraints, simulation modeling is usually shaped by the paradigm of discipline more than by the modeling technique. The concept of simulating a system is too general and unstructured to be in itself a paradigm that helps one organize questions and observations about the world."
"System dynamics, however, includes not only the basic idea of simulation, but also a set of concepts, representational techniques, and beliefs that make it into a definite modeling paradigm. It shapes the world view of its practitioners..."
"System dynamicists are not primarily concerned with forecasting specific values of system variables in specific years. They are much more interested in general dynamic tendencies; under what conditions the system as a whole is stable or unstable, oscillating, growing, declining, self-correcting, or in equilibrium."
"The
primary assumption of the system dynamics paradigm is that the persistent
dynamic tendencies of any complex system arise from its internal causal
structure - from the pattern of physical constraints and social goals, rewards,
and pressures that cause people to behave the way they do and to generate
cumulatively the dominant dynamic tendencies of the total system.
A system dynamicist is likely to look for explanations of recurring
long-term social problems within this internal structure rather than in external
disturbances, small maladjustments, or random events."
"The central concept that system dynamicists use to understand system structure is the idea of two-way causation or feedback. It is assumed that social or individual decisions are made on the basis of information about the state of the system or environment surrounding the decision-makers. The decisions lead to actions that are intended to change (or maintain) the state of the system. New information about the system state then produces further decisions and changes (add link here to graphic like Figure 2.4). Each such closed chain of causal relationships forms a feedback loop. System dynamics models are made up of many such loops linked together. They are basically closed-system representations; most of the variables occur in feedback relationships and are endogenous. When some factor is believed to influence the system from the outside without being influenced itself, however, it is represented as an exogenous variable in the model."
"Feedback processes do not operate instantly; the timing of system behavior depends on the presence of system elements that create inertia or delays. These inertial elements are referred to as state variables or levels. Each level is an accumulation or stock of material or information. Typical levels are population, capital stock, inventories, and perceptions."
"System elements representing the decision, action, or change in a level (often, but not always, induced by human decision-makers) is called a rate. A rate is a flow of material or information to or from a level. Examples are birth rate, death rate, investment rate, or rate of sales from inventory."
"The representation of a system by means of feedback, levels, and rates requires a careful distinction between stocks and flows of real physical quantities and of information. In the system dynamics paradigm physical flows are constrained to obey physical laws such as conservation of mass and energy. Information, on the other hand, need not be conserved, it may be at more than one place at a time, it cannot be acted upon at the same moment it is being generated, and it may be systematically biased, delayed, amplified, or attenuated."
"Two kinds of feedback loops are distinguished. Positive loops tend to amplify any disturbance and to produce exponential growth. Negative loops tend to counteract any disturbance and to move the system toward an equilibrium point or goal. Certain combinations of these two kinds of loops recur frequently and allow system dynamicists to formulate a number of useful generalizations or theorems relating the structure of a system (the pattern of interlocking feedback loops) to the system's dynamic behavioral tendencies."
"These and other structure-behavior theorems are the main intuitive guides that help a system dynamicist interpret the observed dynamic behavior of a real-world system, specify causal hypotheses about that behavior, and detect structural insufficiencies in a model. They permit identification of isomorphisms in very different systems that can be expected to have similar behavioral patterns. For example, to a system dynamicist, a population with birth and death rates is structurally and behaviorally the same as an industrial capital system with investment and depreciation rates. They look like this:... (add link here to graphic example like in book) ...and from their structure can be expected to grow exponentially, decline exponentially, or oscillate, but not to exhibit sigmoid growth (because of time delays)."
"As these examples illustrate, time delays can be crucial determinants of the dynamic behavior of a system. System dynamics theory emphasizes the characteristics and consequences of different types of delays, both in information and in physical flows. System dynamicists expect and look for lagged relationships in real systems, and represent such lags carefully in their models."
"Non-linearities are also considered important in explaining system behavior. Non-linear relationships can cause feedback loops to vary in strength, depending on the state of the rest of the system. Linked non-linear feedback loops thus form patterns of shifting loop dominance - under some conditions one part of the system is very active, and under other conditions another set of relationships takes control and shifts the entire system behavior. A model composed of several feedback loops linked non-linearly can produce a wide variety of complex behavior patterns, and can represent an evolving or adapting system structure. System dynamicists are trained to be very sensitive to non-linearities and to expect that proper identification of them will help in understanding how a system works."
"A final distinguishing characteristic of the system dynamics paradigm is its emphasis on underlying causal mechanisms, whether directly observable or not, rather than on observed correlations. In social systems models, any representation of causation must include human motivations. System dynamicists are trained to be aware of and to include explicitly such factors as desires, expectations, perceptions, and goals. Information about these behavioral factors is gained from social and psychological theory, from interviews with decision-makers in the system being simulated, and from observations of the actual decisions made under a variety of external circumstances."
"System dynamics models are usually intended for use at the general-understanding or policy-design stages of decision-making. Therefore, they tend to be process-oriented, fairly small, aggregated, and simple. Most fall within the range of 20-200 endogenous variables. (Footnote: There are some notable exceptions to this generalization, such as the National Economic Model, constructed by Jay Forrester and his colleagues at M.I.T., which contains over 2,000 endogenous variables). The individual model relationships are usually derived directly from mental models and thus are intuitive and easily understandable. The paradigm requires that every element and relationship in a model have a readily identifiable real-world counterpart; nothing should be added for mathematical convenience or historical fit. Thanks to the high standards set by Forrester, system dynamics models are usually well-documented and easy to reproduce."
"The system dynamics paradigm assumes that the world is composed of closed, feedback-dominated, non-linear, time-delayed systems and thus the method must be most applicable to systems that do indeed possess these characteristics. In general, such systems will be characterized by distinctive dynamic patterns, long time horizons, and broad interdisciplinary boundaries."
Meadows, D.H., and J. Robinson (1985) The Electronic Oracle: Computer Models and Social Decisions. Chichester, England: John Wiley and Sons. [to quote]