When estimating quantities in the Profiler's Simulator tool, we get a graphical rendition of the uncertainty. But in the numerical output both in the simulator and in the output provided if we make a data table (e.g., via Simulation Experiment), we just get a point estimate for the mean and standard deviation and the Out-Of-Specs.
Recently we attempted to run a simulation experiment with all of the factors locked at their profiler settings (e.g., unchecked the boxes for all the factors) to get an idea of the simulation-to-simulation variability. The basic idea was to try to figure out how many simulation runs were needed to control the error in the simulation estimates (primarily mean and OOS rate) to a certain maximum amount of error. This feels like quite a bit of trial and error, though.
If the simulator is keeping track of uncertainty (which I understand it might not be doing), it seems to me there might be a way to replace the number of runs criteria with a request to run enough runs until the estimated error is within some bound. Understandably this would still not be perfect since the error is partly dependent on the stochastics of the simulation itself, but I'm interested nonetheless in what might be possible. Any thoughts?
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />I have noted this in our request system and recorded the Defect ID above.
