On the reliability of numerical studies of stochasticity