Database Reference
In-Depth Information
lengths of each time interval for which state variables are updated. Choose al-
ternative integration techniques. (In the STELLA program, for example, reduce
the time interval DT by half and simulate the mode again to see if the results
are the same.)
8. Vary the parameters to their reasonable extremes and see if the results in the
graph still make sense. Revise the model to repair errors and anomalies.
9. Compare the results to experimental data. This may mean shutting off parts of
your model to mimic a lab experiment, for example.
10. Revise the parameters, perhaps even the model, to reflect greater complexity
and to meet exceptions to the experimental results, repeating steps 1-10. Frame
an enlarged set of further questions. Consider the analogies to your model. Can
these analogies further inform your model?
Do not worry about applying all of these steps in this order as you develop your
models and improve your modeling skills. Do check back to this list now and then
to see how useful, inclusive, and reasonable these steps are.
You will find that modeling has three possible general uses. First, you can experi-
ment with models. A good model of a system enables you to change its components
and see how these changes affect the rest of the system. This insight helps you ex-
plain the workings of the system you are modeling. Second, a good model enables
prediction of the future course of a dynamic system. Third, a good model stimulates
further questions about the system behavior and the applicability of the principles
that are discovered in the modeling process to other systems.
Remember the words of Walter Deming: “All models are wrong. Some are use-
ful.” To this we add: “No model is ever complete.”
1.10 Questions and Tasks
1. We have stated that the key attributes of a model aspiring to reality are feedbacks,
delays, and randomness. The earlier models show the feedbacks and delays. Can
you add uniform randomness to the multiplier in AWARENESS LEVEL?
2. Add an explicit delay in the GETTING SICK control variable in Figure 1.14.
Try different levels of delay to see if you can get a permanent cycle in the SICK
population.
3. Suppose that AWARENESS was a function in part of the SICK TIME. How
would you represent such a possibility in the model in Figure 1.15?
