ABSTRACT
This chapter describes methods for estimating the capability of a process from variable data when the data do not come from a normal distribution. Three methods are included: seeking a transformation of the data to a metric in which the transformed values are normally distributed, fitting a large number of nonnormal distributions and selecting the one that fits the observed data best, and selecting a Johnson curve that matches the first four moments of the data. In each case, capability indices are constructed that have the same relationship to the proportion of nonconforming items as when the data come from a normal distribution. Confidence intervals are constructed for the capability indices using bootstrapping.
