RE: stable system?

[Rip Stauffer <rstauffer@bluefirepartners.com>]

Without having the sample data, it is difficult to say for sure, but it
appears that you have calculated a standard deviation for the raw means. You
should always use the average of k dispersion statistics for calculating
control limits. (See Wheeler & Chambers (1992), pp 56-60).

Try this: take a sample standard deviation for each subgroup. Find the
average of those sample standard deviations. Then multiply that figure by
the factor A3 (available in any SPC text). Add and subtract that result from
your grand mean. That will give you the correct control limits for your
means. You will also want to plot the sample standard deviation values, plot
the average sample standard deviation and calculate a control limit using
the factor B4. 

It's hard to tell from your description,  because you have used a single,
non-bias-adjusted dispersion statistic for the control limits; but it sounds
like you might have some significant within-subgroup variation. How were
these data collected? You need to have some rational basis for
subgrouping--you need to find data that are fairly homogenous
within-subgroup (consistent within-subgroup variation from subgroup to
subgroup) so you can cut the signal/noise ratio and let signals in the
between-subgroup variation come out. I don't know what others will say, but
I think you have to work the variation in the s-chart (or the R-chart, if
that's your wont) first, as it sets the noise level.

You should be suspicious if all your data are within one sigma of the
average. That is usually a signal of inappropriate subgrouping,
inappropriate calculations, or some other inappropriate use of the data in
the control chart.

Best regards to all,

Rip

Rip Stauffer, Senior Consultant
BlueFire Partners
1300 Fifth St. Towers, 150 So. Fifth St.
Minneapolis, MN 55402
612-344-1027
rstauffer@bluefirepartners.com