Capability Indices for Non-Normal Distribution using Gini’s Mean Difference as Measure of Variability

Abstract

This paper investigates the efficiency of Gini's mean difference (GMD) as a measure of variability in two commonly used process capability indices (PCIs), i.e., Cp and Cpk. A comparison has been carried out to evaluate the performance of GMD-based PCIs and Pearn and Chen quantile-based PCIs under low, moderate, and high asymmetry using Weibull distribution. The simulation results, under low and moderate asymmetric condition, indicate that GMD-based PCIs are more close to target values than quantile approach. Beside point estimation, nonparametric bootstrap confidence intervals, such as standard, percentile, and bias corrected percentile with their coverage probabilities also have been calculated. Using quantile approach, bias corrected percentile (BCPB) method is more effective for both Cp and Cpk, where as in case of GMD, both BCPB and percentile bootstrap method can be used to estimate the confidence interval of Cp and Cpk, respectively.