The purpose of creating a discrete-event simulation is often to improve understanding of the underlying system or guide decisions about the underlying system. Numerical results gathered during simulation can be important tools. For example:
If you simulate the operation and maintenance of equipment on an assembly line, you might use the computed production and defect rates to help decide whether to change your maintenance schedule.
If you simulate a communication bus under varying bus loads, you might use computed average delays in high- or low-priority messages to help determine whether a proposed architecture is viable.
When you design the statistical measures that you use to learn about the system, consider these questions:
Which statistics are meaningful for your investigation or decision? For example, if you are trying to maximize efficiency, then what is an appropriate measure of efficiency in your system? As another example, does a mean give the best performance measure for your system, or is it also worthwhile to consider the proportion of samples in a given interval?
How can you compute the desired statistics? For example, do you need to ignore any transient effects, does the choice of initial conditions matter, and what stopping criteria are appropriate for the simulation?
To ensure sufficient confidence in the result, how many simulation runs do you need? One simulation run, no matter how long, is still a single sample and probably inadequate for valid statistical analysis.