Reproducibility

By default, each simulation performed in @RISK will return a different set of random results. However, by specifying a fixed random number “seed” it is possible, assuming no structural changes have been made to the model, to force @RISK to return the same random numbers each time it is run. See Sampling Settings for more information on setting a fixed initial seed.

Setting a fixed seed value helps to isolate the effects of changes to the distributions from purely random variation. For example, one might want to simulate the same model twice, changing only the argument values for a single distribution function. By setting a fixed seed, the same values will be sampled from all distribution functions, except the modified one, during each iteration. Therefore, any differences in the results between the two runs will be due to the new argument values of the single modified distribution function. Please note, if the model’s overall structure (number of distribution functions, their relative positions, etc.) has changed, this type of analysis is not possible.

When running more than one simulation in a single run with a fixed seed, there is a choice between using the same seed for each simulation in the run, or using a different seed for each simulation in the run (see Sampling Settings for more information). The proper choice depends on the reason for running multiple simulations. If the goal is to isolate the variation between the two simulations, using the same seed makes sense. If the interest is more in determining how stable a set of results are, using a different seed each simulation makes more sense.