RE: SPC on the rocks

[Rip Stauffer <rstauffer@bluefirepartners.com>]

J.D. Kromkowski wrote, "I submit that SPC may be applied to any process that
produces data."

I submit that this statement, while technically accurate, may be less than
useful. We can run charts on any process that produces data, certainly. The
results may contain (quoting Don Wheeler) "more math than meaning," like an
average phone number.

The data should be homogeneous (come from the same process), they should be
time-ordered, and they should be independent. Above all, they should be
selected to be representative of the process. Unless the data meet at least
these criteria, charts will be less than useful in studying a process of
interest.

I was asked to look over a control chart at a large industrial site several
years ago. When I looked at the chart, I asked, "what does this dot
represent?" I was told that it was "time taken to complete a job." I knew,
right then, that they were in trouble--this was a huge site, with hundreds
of diverse maintenance jobs completed each month. The center line was around
4 hours, with an upper and lower natural process limit at about plus or
minus 3.5 hours. 

They showed me one "special cause" that was of particular concern. This was
a 74-hour spike. I asked, "What does THIS dot represent?"

"That was a ship that pulled in, and needed to have a crane replaced."

"How long does it usually take to replace a crane?"

"About 3 days, if we work 'round the clock..."

"Then, why is this a special cause?"

"WE DON'T KNOW! That's why we called you!"

I explained that the point was not a special cause, did not signal a shift
in the process. The majority of their jobs were deliberately written to take
up smaller timeframes, for ease of scheduling. A 3-day job is going to take
about 72 hours, and should not be included in a chart for shorter jobs.

More importantly, we discussed what they were tracking. What question were
they trying to answer? They already knew how long jobs took to complete. As
it turned out, an important quality characteristic was meeting schedule. If
they estimated 3 hours for a job, they needed to know how often that job
actually took 3 hours. So they immediately began tracking deviations from
estimates. This gave them an idea of how good their estimation process was,
and pointed to opportunities for improvement in the estimation process. 

It's true, you can plug any set of numbers into a formula, generate a
centerline and some control limits, and produce a control chart. If the
chart doesn't answer questions of interest, or gives misleading answers, it
is at best an academic exercise and at worst misleading and harmful.

How does this apply to the number of messages in the DEN? Well, if we learn
anything from our chart exercise, it's that the data are not independent.
Interesting threads tend to breed clusters of related messages. That
discovery leads me to conclude that the number of messages posted to the DEN
is inherently unstable, therefore unpredictable. That's OK with me...I'm not
sure that stability is a desirable state. We could study the clusters to
find out what makes up interesting threads, I suppose, if that interests
someone. 

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
mailto:rstauffer@bluefirepartners.com
http://www.bluefirepartners.com/