Measurement Problem

[Jonathan Siegel <jmsiegel@yahoo.com>]

I realize this doesn't answer all your questions, but
hope it's a start.


The first thing I would do is try to find out if
you're dealing with apples that vary in size and
occassionally get bigger than you want, or if instead
you have a bunch of cherries and every now and then a 
watermelon pops out. I'd plot out the
time-to-completion data you have and see if it forms
more-or-less one cluster, or if there's one group
centered on a few days and another group that takes
really long. If there's a clear separation into two
groups it should be clear on the plot (I'd use both a
bar chart/histogram and a plot over time for this.)

If there's only one group, then it will make sense to
track and manage all the orders together and make
changes to the whole process to get smaller and more
consistent apples. Since you don't care about lower
limits, a Shewhart chart should do the trick. 

If you're dealing with cherries and watermelons, the
difference should be unmistakable from your chart. If
this is happening, it would be best to assign each
outcome to a hogh/low category and chart three things:
the incidence of high values (percent of orders with
high values) as a p-chart, and separate Shewhart
charts of the high and low values. A single chart
Mixing cherries and watermelons would only make your
situation appear out of control when it may well not
be, and cloud your ability to manage effectively.
You'll need to set a cut-off for the high values. When
you do this, it's very important to base it on the
historical data as it has historically been (as it
appears in your initial chart), not as your boss would
like it to be. For example, if one group goes from 1-4
days and another generally goes from 1 week to a
month, the cutoff point should be 5 or 6 days, not two
weeks.

If your high and low values are each in approximate
statistical control, the managerial goal would be to
make fewer watermelons and more cherries. To do this,
you'll need to learn how to tell a watermelon from a
cherry while they're still seedlings, maybe before
they're planted. You should expect to be able to
examine the high-turnaround work orders as a group and
find some sort of systematic difference between them
and the low turnaround-values: it should be there
waiting to be found. (type of job, customer, operator,
etc.) 

On the other hand, If there's only one group (all
apples), trying to make changes based on comparing the
high values to the low values likely won't help you
and may hurt. They're all "really" the same -- all
based on the same process, and it's only "chance" that
makes some high and others low. Never good to tamper
with chance. Instead, it might be better to simply not
worry about the high values, and chart and manage the
process as a whole. Effective change will only come
from steps that either reduce the overall turn-around
time, or make it less variable. The high values will
improve along with everything else.

Hope this helps. 

Jonathan Siegel