Finally, let us pay some attention to using statistics. It is clear that by the stochastic nature of EAs only statistical statements about behaviour are possible. Typically, averages and standard deviations supply the necessary basis for claims about (relative) performance of algorithms. In a few cases these can be considered as sufficient, but in principle it is possible that two (or more) series of runs deliver data that are statistically indistinguishable, i.e., may come from the same distribution, and the differences are due to random effects. This means that the two series, and the behaviour of the two EAs behind these series, should be considered as statistically identical, and claims about one being better are ill-founded.