Multiple Components of Variation

Note: See Multiple Components of Variation in the SAS/QC Sample Library.

In the preceding section, the excessive variation in the conventional Shewhart chart in Figure 17.190 is the result of positive autocorrelation in the data. The variation is excessive not because it is due to special causes of variation, but because the Shewhart model is inappropriate. This section considers another form of departure from the Shewhart model; here, measurements are independent from one subgroup sample to the next, but there are multiple components of variation for each measurement. This is illustrated with an example involving two components.[50]

A company that manufactures polyethylene film monitors the statistical control of an extrusion process that produces a continuous sheet of film. At periodic intervals of time, samples are taken at four locations (referred to as lanes) along a cross section of the sheet, and a test measurement is made of each sample. The test values are saved in a SAS data set named Film. A partial listing of Film is shown in Figure 17.197.

Figure 17.197: Polyethylene Sheet Measurements in the Data Set Film

Sample Lane Testval
1 A 93
1 B 87
1 C 92
1 D 78
2 A 87




[50] Also refer to Chapter 5 of Wheeler and Chambers (1986) for an explanation of the effects of subgrouping and sources of variation on control charts.