Database Reference
In-Depth Information
remember that if the flow equations must be understood as differential equations,
DT needs to be sufficiently small.
Choose Time Specs from the RUN menu to change DT. Change DT to reflect
ever-smaller periods until the change in the critical variable is within acceptable
tolerances. Start with a DT = 1 and reduce it to 0.5, 0.25, 0.125 and so forth for
subsequent runs, each time cutting it into half of its previous value. Before each run
note the values of your state variable at the end of the previous run, as STELLA
will erase the graphs if the DT is changed. To lock graphs, and thus not lose model
results when you change the DT, click on the lock in the lower right-hand corner
of the graph pad. This preserves the results. Then double-click on the graph pad to
open its dialog box, click on the upward-pointing triangle to generate a new page for
the graph pad, and select the variable(s) to be plotted. Now, one page of the graph
pad contains the results for one choice of DT; the other contains the results after
changing DT.
To get an even more accurate reading of your model results at each point in time,
create a table. Choose the table icon in STELLA, place it in the model diagram,
double-click anywhere on the table and select SICK as the input, then double-click
on the head of the column where it reads “SICK” and choose “Free Float” as the
“Precision.” With this specification, your table will report the results with an accu-
racy of more than the two decimal places that would otherwise be listed. (In this
case, though, reporting the number of individuals who are sick with more accuracy
than whole numbers makes little sense, unless the unit of measurement is, for exam-
ple, in millions of people.) However, irrespective of the level of precision at which
you report the results, the computation itself is not affected by that choice. Click
OK and run the model. Note the results (perhaps lock the results of this page of
the table, much as you would lock results on a page of a graph), and then proceed
to change the DT to a smaller value. Compare the results from one model run to
the next. Repeat this process for different solution methods, and observe changes in
model errors. Other sources of errors are discussed in more detail in the following
section.
1.8 Sources of Model Error
Error is associated with virtually every aspect of a model. As we have discussed
above there are errors involved in the algorithms—and sometimes even at the hard-
ware level of the computer—used to numerically solve the model. A set of errors
is also associated with the conceptualization of the model and is the topic of this
section
9
.
Any model is an abstraction. In the process of making this abstraction, a limited
set of system components and their interactions are considered. Features that are
irrelevant to the system's dynamics over the temporal and spatial range or to the
9
The discussion of sources of errors presented in this section follows Westervelt (2001).
