For the step shifts, the ARL curves behaved differently from what had been observed before (Fig. 7a). For shifts greater than 2%, the configuration with the smallest q value is the optimal (here q = 0.6), but for shifts of 1%, the ARL curve for q = 0.6 was the least efficient; the higher q values became more interesting. This behavior was a direct cause of the statistical inertia created by using of a q value close to 1 (or a λ value close to 0, in the EWMA case). When smoothing was significant, the control chart would take some time to register an instantaneous variation. In the case of step variations, the GWMA performed better than the EWMA for a 1% shift with an ARL of 39 days for q = 0.9. It might be complicated to decide what optimal value of the smoothing factor would be, given the fact that the optimal values changed according to the shift's amplitude. This kind of behavior was also observable for the EWMA with a step variation, but at much higher underperformance values. However, the focus should be on the performance of the control chart for small shifts. With the results provided by the validation bench, the difference in the ARL in function the smoothing parameter is bigger for small shifts (1–2%) than for relatively medium shifts (5–15%). So in a way, we could choose q = 0.9 as the optimal value. Various values of α were been tested from 0.6 to 1.2, but it appears that their influence on the ARL was small.