Hello,

I am working for this project and SimTraffic 5 runs is neeeded. I did the 5 runs and I print out the result from the 5 runs using the "Multiple Runs" option. I also printout the result from each (run 1,2,3,4, and 5) of the 5 runs independently. The result from the 5 runs report is different from the average of run 1,2,3,4, and 5. From an example:

from the 5-run report (using the "Multiple Runs"), the queue for this movement is 1790 feet

from independent report, the same movement is 1840 feet (run 1), 1712 feet (run 2), 1674 feet (run 3), 1684 feet (run 4), and 1803 feet (run 5). When average these # (1840+1712+1674+1684+1803)/3 = 1743

How did SimTraffic calucated the average in the multiple runs report? Please Help!!

Here is what I got from John Albeck regarding my question.

"Thanks for the file. This is related to the email message below. Even when you are running 1 - 60 minute simulation, the 95th percentile queue is still based on intervals. Internally, SimTraffic stores the maximum queue every 2 minutes. For a single run, the average queue is the average of these 2 minute maximums. Then, a standard deviation of the 2 minute maximums is determined internally. To get the 95th, the 95th queue is then:



Avg Queue + 1.65 Std Dev



When you create a single run, the Standard Deviation is determined for 30 intervals (30 x 2 minutes = 60 minutes total). However, when you do the averaging of 5 runs, this standard deviation is actually 30 x 5 intervals, or 150 intervals. Using the same formula above, this will give you a different value for 1.65 Std. Dev.



Personally, I’m not a big fan of the 95th queue in the SimTraffic report. The reason is that it is NEVER really observed in the simulation. It is simply a statistical distribution on top of the average. In fact, since it is a simply calculation, it can lead to results such as what you are seeing. It remains in the reports since some users do want this value, but I could see where it is removed for the future.



I would suggest using just the average and the maximum queues. At least these are based on actual values that are observed.



Thanks,


John. "