Database Reference
In-Depth Information
other parameters produce little such change. This knowledge tells us where to put
our research effort into refining parameter values.
Elementary to modeling is the idea that a model should be kept simple, even
simpler than the cause and effect relationships it studies. Add complexities to the
model only when it does not produce the real effects. Remember that models are
sketches of real systems and are not designed to show all of a system's many facets.
Models aid us in understanding complicated systems by simplifying them.
Models study cause and effect; they are causal. The modeler specifies initial con-
ditions and relations among these elements. The model then describes how each
condition will change in response to changes in the others. In the example of host-
parasite interactions, a larger population of hosts could provide more opportunities
for parasites to grow, reproduce, and spread—which could affect the health of hosts,
reduce the growth rate of their population, and thus negatively impact growth of the
parasite population.
Key to model design is the decision on system boundaries—what to explicitly
model and what to consider as given. Open models are models whose workings are
influenced by a large number of outside influences that are not explicitly modeled.
For example, your model of the spread of a disease may consider birth and death
rates as given, contact rates of infected with susceptible individuals fixed, and any
number of other parameters not the actual subject of your model itself. In that case,
you have an open model. As you explain more how birth, death, or contact rates
change in response to factors that are part of the model, you “internalize” these
factors and “close” your model. As you approach a closed model, its complexity is
likely to increase as ever more components are influenced by each other.
The initial or starting conditions from which a model runs could be actual mea-
surements (the number of people in a city in a given year) or estimates (the number
of people there in four years, given normal birth rate and other specified conditions).
Such estimates are designed to reflect the process under study, not to provide precise
information about it. Therefore, the estimates could be based on real data or the rea-
sonable guesses of a modeler who has experience with the process. At each step in
the modeling process, documentation of the initial conditions, choice of parameters,
presumed relationships, and any other assumptions is always necessary, especially
when the model is based on the modeler's guesses.
Dynamic models have an interesting interpretation in the world of dynamical
statistics. The entire dynamic model in STELLA might be considered as a single
regression equation, and it can be used that way in a statistical analysis of the best
parameter choice for optimization of a performance measure
1
, such as maximizing
the effectiveness of a vaccine by choosing the best fraction of the population to
be vaccinated. It replaces the arbitrary functional form of the regression equation
used in statistical analysis. Thinking of the dynamic model in this way lets one
imagine that because more actual system form and information is represented in the
1
Personal paper: Twenty-Five Years (Isolated) Behind Enemy Lines: Do economists and statis-
ticians have anything new to offer Systems Dynamics? George Backus, Policy Assessment Cor-
poration, 14604 West 62nd Place, Arvada, Colorado 80004, George Backus@Energy2020.com,
2003.
