System dynamics (SD) is the application of feedback control systems principles and techniques to managerial, organizational, and socioeconomic problems. As such, the methodology seeks to bring together multiple views or aspects of the same problem under study and integrate them into a conceptual and meaningful whole. In fact, most difficulties to fully understanding complex issues arise from looking independently at various elements of an issue instead of considering pertinent interrelations. Consequently, optimization is sought for each separate element in the system, which inadvertently leads to sub-optimization of total system performance. With SD, it is possible to take hypotheses about the separate parts of a system, to combine them in a computer simulation model, and to learn both the “local” and “global” consequences of decisions and actions, as well as the impact of these decisions and actions on short-term and long-term performance. Most of the time, the impact on short-term and long-term performance are opposite: an action that looks positive in the short-term is often very detrimental in the long-term. Conversely, an action that produces favorable long-term performance must usually suffer poor performance in the short-term.
SD extends modeling methods traditionally associated with engineering design and feedback control theory into the arena of policy evaluation and management decision making. The following characteristics distinguish SD models from traditional decision support methodologies:
Its building blocks are feedback loops;
It can accommodate non-linear relationships among variables;
It enforces causality;
It can include delays;
It can model “soft” variables;
It can model management policies; and
It presents a dynamic environment for decision analysis.
These characteristics are important because they allow SD models to capture the key structural relationships that define a social system. The structure, in turn, produces the dynamic behavior of interest. The resulting simulation mirrors reality because the underlying model structure includes the appropriate feedback loops, causality, delays, and other relationships. SD models include real-world causal logic, which allows someone to trace through the model to see why things happen the way they do.
The SD modeling and simulation approach is different from traditional statistical approaches in several ways. First, the models are more realistic because they capture cause-and-effect linkages, feedback loops, delays, non-linear relationships, and management policies. Second, the simulations are more accurate and reliable because they provide a sanity check on assumptions and are more rigorous than mental models or spreadsheets, allow for analysis of a wider range of issues, and identify the actions that are most effective (and least effective) for improving performance. Third, communication is more effective because the approach is graphical (the connections are easily seen and understood), logical (the results can be traced back to their root causes), and experiential (we learn best by doing and simulation is a good substitute for the real world).
In SD models, a “stock” and “flow” methodology is used in which stocks represent accumulations of “things” (e.g., people, inventory), and flows are the movement of these “things” into, out of, and between stocks (Figure E-14). For Scenario 1 (moderate risk) and Scenario 2 (high risk), a very basic SD model was used in which the stocks represent groups of people in the following categories (which were established based on available data):
In Science and Engineering (S&E)—The number of people employed in science and engineering positions (not considered postdoctorates).
Out of S&E—The number of people employed in areas other than science and engineering.
Unemp Seeking Work—The number of people currently unemployed but are seeking work.
Unemp Not Seeking Work—The number of people currently unemployed but not seeking work, but are not retired.
Retired—The number of people currently retired.
Postdoctorate—The number of people employed as postdoctorates.
The total number of people considered in the “workforce” is the sum of all people that are not retired. Thus, the workforce for any particular demographic group (e.g., U.S.-trained males in biomedical science) is the following:
Workforce = In S&E + Out of S&E + Unemp Seeking Work + Unemp Not Seeking Work + Postdoctorate
The flows in and out of the stocks (e.g., In 1, Out 1) are based on growth rates determined from the data for the specific demographic group and shown earlier in Tables E-13 and E-14. If the growth rate is greater than zero (i.e., positive), then people are added to the stock through the In flow. If the growth rate is less than zero (i.e., negative), then people are removed from the stock through the Out flow. The amount of people that are added or removed is based on the percentage growth rate multiplied by the current number of people in the stock. For example, if 100 people were in a stock and the growth rate is 5 percent, then 5 people would be added to the stock during that simulation step.
Figure E-14 below shows this stock-and-flow diagram for the U.S.-trained males in biomedical science. This exact same model structure is used for all other demographic