I made two simulation runs of a saturated network with different random seeds and found that the results (e.g. queue length) vary quite a lot. Why does that happen?
In a saturated network, minor changes may lead to big consequences. For instance, due to a slight variation of green time, the number of vehicles passing through may be one vehicle less per cycle. This vehicle might be the critical one which leads to a queue that builds up continuously during the simulation whereas in the other case, the green time was just sufficient to accommodate the entire demand. These effects can also be seen on the field, where normal day-to-day changes may lead to different traffic situations. A minor change (e.g. in lane change) can also lead to different results within the typical statistical boundaries. Generally speaking, a network which is not operating at capacity will react less to changes of the random seed.