RE: Prediction

["Prevette, Steven S" <Steven_S_Prevette@rl.gov>]

Question:  For some problems I use least squares methods (regression
analysis) for prediction and then SPC for monitoring the residuals.
Surely we are not using regression analysis in out brains 24/7, but is
there any kind of operational definition for 'prediction' that would
explain what Deming means when he uses this word in various contexts? 

There are a couple of contexts for "prediction" in the Deming thinking.
I myself am struggling through CI Lewis's Mind and the World Order for
about the past year.  For example, we have knowledge of apples.  We have
from experience categorized the shape and color of an apple (although
there are other varieties than red, most of us I believe think of red
apples).  If I held an apple now, I could predict the taste of the apple
if I were to bite into it in the future.  If it turned out to be sour, I
may need to modify my prediction (of at least that batch of apples
perhaps).

In terms of data, I use SPC to help with the prediction.  Yes, SPC on
residuals of a regression should also work, though you run into problems
in predicting within the data used to perform the regression.
Predictions toward the center of the data have a smaller prediction
interval than those extrapolated away from the data.  So we can run into
some conceptual problems there.

But, if my data are "in control" per SPC rules, then I can use the SPC
results to predict the future.  I can also use it as a risk analysis.
Can you live with the current results, centered on the average,
spreading to the control limits into the future?  If not, the process
must be changed.  If the process is not "in control", then I cannot
predict, except to say to expect uncertainty, especially if I do not
know the cause of said special cause variation.

Steve Prevette
Occupational Safety and Health
Fluor Hanford, A Fluor Government Group Project
ASQ Certified Quality Engineer
steven_s_prevette@rl.gov
509-373-9371