In fact, the values represented come from measured end-to-end delays. This necessarily introduces a tail effect, as there is a clear limit on the possible values from the left side (i.e. time delays cannot be negative) but none on the right side. Looking at the range of absolute values, we see that with a mean sample value of 6.5ms very few messages achieved a delay less or equal than 2ms and the distance between the minimum value and the mean is approximately of 5ms. However, on the right side, this distance goes up to around 34ms, with a maximum value close to 40ms. Notice that the ® value performing the EWMA is responsible, in a certain way, of this effects. A lower ® acts as a filter for higher sampled values and hence, reduces the tail on the right side. However, this would also affect the sample variance s2 as values would get closer to each other. Thus, a side effect would be a distortion on the estimated distribution which would look thinner. On the other hand, higher values of ® would reduce the smoothing effect of the EWMA and estimate a better value for the sample variance. This would definitely reflect on the peak of the estimated distribution, although, at the same time, produce a thicker distribution shape.