Microsoft Office Tutorials and References
In Depth Information
Finally, question 4, regarding the sample size, is always relevant in Monte Carlo
simulation. It asks whether our simulation has a sufﬁciently large sample of model
operation (250 days in our case) to accurately describe behavior. As a modeler gains
experience, it becomes clear that relatively few sources of variation are sufﬁcient to
greatly affect results, even when data from a relatively large sample is collected. So
how do we deal with this question? The obvious answer to this question is to repeat
the experiments to produce results multiple times. In other words, simulate 250 days,
simulate another 250 days, and repeat the process numerous times. Then examine
the results of each simulation relative to the combined results of the simulations.
This is easily done in our model by recalculating the spreadsheet and noting the
changes with each recalculation.
In Exhibit 8.26 you can see Stkout data collected in 20 replications of the 250 day
experiment. Notice the range of Stkout values for Engine/electrical Diagnosis in
the ﬁrst column. Values range from 2 to 15 and average 9.9, while the range for
Mechanical and Oil are far less variable. A value of 2 appears to be a rare occur-
rence, but this example serves to suggest that some data elements will require
substantial sample size to insure a representative result, and it may be necessary
to increase sample size to deal with variation in Engine/electrical Diagnosis .
Exhibit 8.26 Multiple replications of 250 days simulation stock-out data