Discrete-Event Simulation and Statistical Analysis
This domain tests simulation design and output analysis. Expect event scheduling, random-variate generation, input modeling, warm-up, replication, confidence intervals, common random numbers, verification, and validation.
Core rules
- A discrete-event simulation advances the clock from event to event, not by fixed time steps.
- Observations within one run are often autocorrelated; independent replications give cleaner confidence intervals.
- Delete warm-up data when initial conditions bias the steady-state estimate.
- Use the t interval when the population variance is unknown and replication means are approximately normal.
- Common random numbers reduce variance when comparing alternatives by using the same random streams.
Worked example
Ten independent replications of average waiting time give means 12, 14, 13, 15, 16, 11, 13, 14, 15, and 13 minutes. The sample mean is 13.6. The sample standard deviation is about 1.51. With 9 degrees of freedom and 95% confidence, t≈2.262, so the half-width is 2.262(1.51/√10)≈1.08. The interval is approximately 13.6±1.08 minutes. Using all individual customer waits as if independent would understate the standard error because they share run-level effects.
Traps
Treating autocorrelated observations as independent, ignoring warm-up, reporting a single run as a confidence interval, and validating only the computer code rather than the model against reality.
Key points
- Advance the clock by events.
- Replications support valid intervals.
- Remove warm-up bias.
- Use t intervals for replication means.
- Common random numbers reduce comparison variance.