CONSIDER THE ADVANTAGES OF EXPERIMENTATION:

Consider a health service example, where the number of patients admitted to
the clinic is a set number arriving according to a variable timing profile. To
establish the weekly variability of utilizations of clinical staff, the model can
be run for either a single run of 50 weeks or for 50 runs of a single week. Each
of these runs will enable the variation of utilizations to be observed and confidence
intervals calculated. Multiple runs are important. You would not judge
the fairness of a coin by just one or two tosses!
It is true that the above model is simplistic, as indeed the profile of patients
in such a model would probably vary in ways that would require more extensive
modeling (e.g., the definition of a likely profile over a longer period to take into
account current observed fluctuations from week to week). However, the experimentation
options hold true: either a short period of experimentation can be
repeated many times (with different random number sampling), or a model can
be run for a much longer time and the range of random effects can be observed.
How many repeats and how long to run a model are questions that can be
answered by a combination of input and output data analysis. It is important
when running a model that it experience all the randomness of the input data
and that the output results �settle down� before conclusions are drawn.
At a second level, a user may wish to alter different parameters within a
model to observe the different effects. Again, with this type of experimentation
the range of results might also be important, once they are more accomplished
through elongated or repeated experiments.
At a third level, a user may wish to compare one model with another. For
example, in manufacturing there may be two investment options: production
layout A and production layout B. These may indeed be the only options,
although within each solution there may be parameter choices (e.g., containing
buffer storage level options). In addition, there is again the optional value of
establishing the potential range of results.
With all these levels of experimentation, there are a variety of experimental
designs that can be used � different numbers of replications and different types
of factorial experimental designs, ranging from full factorial to half factorial to
Latin square-type designs where different parameter combinations are chosen.
For different models, the model itself can simply be considered a different type
of parameter.