Quality Progress - August 2015 - 50
BY LYNNE B. HARE
Are the Data the Data?
What you see in production line data isn't always what you get
AN ARGUMENT BROKE out over a
data? And what might be the real message
where samples had been removed from
large set of production data. Can you
in the data?
the stream of cookie dough. The blank
spots were all on the close side of the
imagine such a thing? At issue were some
large financial implications. Who you gon-
Are all data created equal?
stream. The bakers and I had missed this
na call? A statistician, of course-namely,
For starters, we might wonder how the
important aspect of sampling. Some quick
me. The team consisted of subject matter
data were created. Are all the potential
inspection showed cookie dough weights
experts, production experts and manage-
sources of variation represented in the
to be 10% higher on the unsampled side
data? If not, all bets are off.
of the dough stream, and the sampling
Subject matter and production experts
Often, statisticians will recommend
problem was quickly resolved.
talked about the large mass of production
random sampling or systematic sampling.
line data accumulated over the years. Of
If sampling is truly random, every unit
course, they didn't have time to clean up
produced has an equal chance of appear-
Next, we would want to know what
the data before making it available to the
ing in the sample. Systematic sampling is
happens to the data between the time
rest of us, they said.
usually more convenient in manufactur-
an observation is made and when it is
ing. There, sampling frequency is tied
recorded. That sounds like precious little
to the clock; every 30 or 60 minutes, for
room for error. What could possibly go
instance, with representative samples
wrong? Sometimes the errors are subtle,
taken at those times.
but no less present.
That's as it should be, the bosses said.
The data are the data.
That comment got me wondering.
Are the data really the data? Perhaps
from the management perspective, they
I recall that once, in a short course on
Fat fingers: One kind of error ap-
constitute the actual record. Therefore,
process variation reduction for cookie
pears in data entry, especially, but not
they are the data. But my questions were:
bakers in Spain, we discussed the need
exclusively, during manual data entry.
Do they represent what really went on in
to accumulate representative samples
Sometimes referred to as the "fat finger
the manufacturing process? Do they accu-
during each sampling to be sure that all
problem," the wrong keys get pressed
rately characterize what eventually ended
sources of variation were included in the
and extreme values are entered into the
up in the possession of the consumer?
sample. The bakers assured me that they
If so, the bosses are right. If not, what
might be reasons why the data aren't the
followed that sampling principle.
Later, during a factory tour, I saw
In the case of extremes, data elimination may be beneficial. For example, a
value of 812 is obviously wrong, given that
Histogram-flinching at a
specification / FIGURE 1
production values are centered near 7.9
with a standard deviation of 0.13. Right?
Maybe it was meant to be 8.12 or 8.20 or
some other number, but we'll never know.
It is probably best to discard it.
If these extremes are left in the data
set, are the data still the data?
Flinching: Another kind of error occurs as a result of "flinching." It is very
human to edit the data mentally as we enter them. The tendency may be to discard
a value and take another sample if we see
an eyebrow-raising number. After all, one
50 QP * www.qualityprogress.com
sample is as good as another, right?
Table of Contents for the Digital Edition of Quality Progress - August 2015
Mr. Pareto Head
Creative by Design
All the Right Moves
3.4 per Million
Quality in the First Person
One Good Idea
Back to Basics
Quality Progress - August 2015