00418 �56a13f3cdc41e5c794440043a03715

422

Pignatiello & Ramberg

2 N =    1000

Midpoint Count

Figurę 6. Histogram of Samples from a Stable Process with C_p = 1.00.

For the process with the true value of C._p= 1.33, the simulation study found

these values to be 1.095 and 1.705 which indicates an interval width of 0.61.

Plots of such interval widths (obtained in a similar manner) versus sample sizes are given in Figures 7 and 8. Theses figures show that for normal processes whose true C_p value eąuals 1.33, sample sizes of roughly 300 would

be needed to obtain a width of about 0.20. Altematively, the estimated standard errors given above can be used to determine approximate sample sizes for normal processes. For example, one could flnd the sample size n so that the interval width (approximated by, say, four times the standard error) is sufficiently smali.

It should be noted that Figures 7 and 8 show that for processes that are not normal, substantially larger sample sizes would be needed to achieve the