Abstract
The problem of controlling correlated data is considered. Data are modeled with common time series models. Both Shewhart and Exponentially Weighted Moving Average charts are placed on the data. It is shown that average run lengths are greatly changed when charts are placed on correlated data, assuming independent data. We discuss how to adjust the control limits to compensate for the correlation in the data and find average run lengths of the adjusted charts. The residuals from a properly modeled time series are independent, and thus are suitable for charting. Performance characteristics are found for Shewhart and EWMA charts on residuals. Performance of residual charts is poor when the data are highly positively correlated. Finally, a control chart based on the one-step ahead and two-step ahead forecast errors from the model is developed, and simulation is used to find its run length properties under various models.
Michelson, Diane Kay (1994). Statistical process control for correlated data. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -1551972.