I am reviewing another engineers SimTraffic results. We generally take the average of 5 runs. However, while reviewing the report, I noted that the run numbers displayed were 1, 3, 5, 6, and 8. When asked about this, i was told that due to an occasional unrealistic result, they throw out the highest overall delay runs. Basically, choosing the best overall runs. This appears contrary to the intentions of the software. Could you give me some guidance on this matter.

I know a number of procedures that are followed by different companies, but we generally do runs 1-5, and summarize the five results. However, before summarizing, we watch each run to see if it was ok, and sometimes we throw out some of the runs. This is usually as a result of "stuck vehicles". Vehicles that do not know where to turn next, or other abnormal behaviors that lead to back ups and/or gridlock.
Some people choose to perform 10 runs, then throw out the highest and lowest.

What I've done in the past is to take the average of the first 10 successful runs. Sometimes the simulation locks up (lane drops, unusual lane configurations sometimes confuse simtraffic) and the run must be thrown out.
Recent training though has me using a different method, which is to examine the first 20 successful runs, and throw out the 5 lowest and 5 highest. Generally the MOE I look at is total network delay to make this determination.

Eric Royer
Caltrans

Eric and Orla made good comments.

I limit my simulation to five runs. So far, no reviewer has questioned that. There are two things that need to be considered:

1. In any run whether any unusual behavior is seen (for instance, a car is stopped for no apparent reason). Special care is needed when I am modelling ramps, roundabouts etc.

2. If it is urban condition I tend to change the driver parameters. I assume OD is always adjusted in synchro before running the simulation.

I do not agree with the concept of throwing out the runs with maximum and minimum delay. Throwing out the outliers (either way) is a good idea. Even if the outlier is possible to happen once in a year, that should not be base for any assessment.

I think the concept of multiple run is to address the traffic fluctuation through an average week, not for special case (special event/unreasonable driver/accidents etc)