Statistical quality control for correlated samples

SUMMARY

The paper is an extension of Neuhardt's work which questions the typical assumption that the subsamples used to establish statistical process control charts are uncorrelated. The work was prompted by Neuhardt's work and by checking the data used to determine statistical process charts of ten applications. In all ten situations the data was found to be correlated. Situations where the subsamples are multivariate normals with constant variance and an arbitrary correlation matrix are presented. The rules and equations to determine the control limits of x¯ charts are presented. Simulations are conducted to determine the effect of correlated subsamples on S, R, and S² charts. The output of user-friendly software that performs the necessary computations for establishing x¯, R, S and S² are presented and discussed.