Abstract
Normal approximation to the distribution of the estimated process capability index _??_pk is discussed. The direct normal approximation to _??_pk and an alternative approach are compared in terms of skewness and kurtosis. It is shown that the latter approach improves on the former one.