In Statistical Process Control (SPC), to implement a control chart, process parameters have to be estimated from a sample that is assumed to be in control. This estimation is prone to be contaminated with special causes of variation. When using rationale subgroups, traditional approaches that uses within sample variation offer protection against shifts in the mean. However, if special causes affect the scale of observations, variance estimation becomes biased. To address this issue, subgroup variance is suggested to be evaluated with a nonparametric method called Modified Tukey Chart, a Phase I control chart created to deal with skewed distributions. By using this approach, perturbations can be detected and eliminated prior final estimation. Bias and standard error of the robust estimator are evaluated and compared against two parametric alternatives when perturbations in the scale of the observations are present. A numerical example using observations from a manufacturing company is also provided. Practitioners might find this procedure useful to reduce inspection costs by making robust estimations of variance, reducing the risk of having excess of false alarms or low power to detect changes during a Phase II of SPC.
