DEN -- Heal thy self

[Kromkowski@aol.com]

Responding to Myron:

A learning opportunity has obviously presented itself. 

First about process behavior prediction methods (is that 
what Wheeler is calling this these days?) and second about 
the DEN.

First things first, control charting: Which I hope you will 
see that I anticipated by my comment: "[I prepared a chart]
using the first 35 data points (maybe you will want to use 
25 or 20 or 30 or 40 or whatever).

A. Myron used only 66 data points, when I posted 71.  A 
minor point which is expectable in using data. Mistakes happen.

B.  Deming suggests repeated in examples from OTC and TNE, 
the efficacy of "using available data" and making control 
chart as a beginning point of investigation. Control tells 
us whether the process producing the data is "statiscally 
stable" (my term), or in other words capable 
of "prediction".  They also help us sort and sift out 
special causes of variation from systemic/common causes 
of variation as and aid to deciding what kind of action, 
if any, to take.

Here there are 71 data points. We are faced with four basic
choices:

i. Construct a control chart using all of the first data 
points, and then use those control limits for entire data 
set.

ii. Construct a control chart using the first N data points,
and then use those conrol limits for the entire data set.

iii.  Construct a control chart using the last N data 
points, and then use those control limits for the entire
data set.

iv. Randomnly chose N data points (keep them in time order),
and then use those control limits for the entire data set.

Choice i. and iv. will probably not be so different.

Choices ii. to iv., present the addition issue of how many N.

Before the spread sheet and macros, or Jean-Marie's program (I have used it and also in the early 90s I wrote my own 
version is Superbase), Deming just used pencil and paper 
(so did and still do I).  His examples suggest that he used between 20 and 40 data points.  Mostly, I think that this is because there is a certain amount effort to hand calculation and above 40, there is usually not much information added. 

If we use method i. (which is essential what Myron did) 
and we use _all_ the data.  Yes, we will find three points
outside on the high end (the ones Myron note which are 
easily explanable:  Jim gets behind one month and then 
catches up so some things get pushed into the next month) 
and also using trend signals, we see this shift.  Putting aside the trend shift signal (7 consecutive points below 
average), we can say: Oh everything is between the limits.  The problem is that what we would need to say is that it 
is total predictable and acceptable that there might be as
few as 9 messages a month or as many as 300 a month! Feast
or famine may form a predictable and statistically stable
process, but then what should we do to about the sytemic
causes of such a wide variation.  (Method iv. produces same
kinds of questions.)

If we use method ii. (first N data) or method iii (last N data), there is question of what N.  Surely, there is some paper or research that will tells economically what N to use. And if we really cannot distinguish between the 
efficacy of the low N and the high N, then it would seem to me that we should use the smallest N that produces 
reasonably indistinguishable results. Using as few as
N=20, seldom will steer someone hopeless wrong.
You may do the math yourself, but you will see either a
signal beginning at point 38 by trend rules (by my data list, or point 34 in Myron's list), when using just the first 20 points to calculate the limits.

If you use the last 20 points (Again you may do the math
for yourself), you are faced with a system which is 
not statiscally stable (or "in control" or "predictable").
The prediction of a negative number of posts, I guess is a prediction that not merely are people not wanting to post but that they are "repulsed" by the idea of posting!
I note here that for some the notion of "predicting"
back in time seems silly or wrong. But theory is supposed
to explain what went before as well as what lies ahead.
See Deming on the rooster and the sunrise.

As to the learning opportunity about the DEN. I am sure
the archives have this suggestion which I previously made:

If the DEN hired Deming as a consultant what would he say.
I think that he would do some form of this control chart. 
And I think we would ask, what the heck is the purpose or
aim of the DEN. (yes we have something written down and it
gets posted from time to time, but does the structure and 
method of the DEN actually help accomplish the aim.)

He also might ask: who's in charge? Who is(are) the 
leader(s) with _the actual power to change things_?

And what are they doing or not doing in relationship to responsibity as leaders or leadership?

Finally, here is a very radical idea:

Close the DEN list down for six months. After six months, find out from the membership whether it was missed and 
whether it should be started again.

JDK
Message posting through the Clemson CQI Web Server.

[Moderator's Note:  The DEN was "down" for 10 weeks in the March-April-May timeframe.  There were e-mails and
phone calls asking why and when we restarted, messages began to flow once again.  I doubt a 6 month test would prove any different.