Robert Bacal, responding to Mary Schaefer, first cites Mary:
> > Is there a moderator on this listserv to exercise communication control?
> > Some of us have actual jobs with professional responsiblities and don't
> > have e-mail capacity for 7th grade recess talk.
My own response to Mary's comment follows: I'm sorry, Mary, but I can't
make the connection to "7th grade recess talk." Does that tie to boys and
basketball or to a view of the conversation as immature or what?
Robert continues:
>Mary, I'm sorry you don't find the conversation interesting, and that
>apparently you aren't clear on the purpose of it. Not all
>conversations can be encompassed in a brief sound-byte.
>
>There's a couple of points to it (and I commend Fred for making the
>effort to play this out -- something most people can't be bothered to
>do). First, if people want to teach managers basic statistical
>concepts, we have just gone through a mini-process that can be
>used to help that.
I don't know that anyone else has learned a whole lot but I added to my
store of knowledge.
>Second, the issue (which I guess goes way over the head of some
>people) has to do with whether we educate managers on the use
>(or misuse) of statistical concepts as they apply to real life
>decisions. Jack Ring's contention that one can assume a normal
>distribution and make decisions that affect real life is one I hope
>nobody here would teach managers. Further, the distribution issue
>has a direct bearing on the use of forced rankings on a normal
>curve sometimes used for appraisals and even in layoffs.
I'll echo Robert on this score. I'll also extend the conversation and loop
it back to the original issue. I grabbed the NBA stats only because they
were handy. Frankly, I don't watch NBA games and don't plan to start. The
question I was exploring, via some pretty hard data, was whether or not a
selected sample might have a distribution approaching anything like a
normal distribution. In this case, the skewing is apparent but it's not so
extreme as to be totally unlike a normal distribution. This was in the
context of Jack Ring's comment about assuming a normal distribution and
Robert's admonition against doing that. I was testing the proposition, as
it were.
What's interesting (to me) about all this is that selecting people on the
basis of some physical attribute such as height is a very different matter
(in my view) from selecting them on the basis of some less unambiguously
measured attribute (e.g., intelligence, mechanical aptitude, and so on).
So, when we argue against assuming a normal distribution when the
population is a selected one, I can see how such arguments can be put to
the test when the selection is based on something that is reliably measured
(e.g., height) but I don't see how that same argument can be applied when
the population is selected on the basis of an unreliable measurement or
assessment of something like attitude, character, performance, potential
and so forth. It seems to me that if we don't have a reliable and
unambiguous measurement as the basis for selecting the population in
question then any arguments about the appropriateness of assuming or not
assuming a normal distribution are theoretical and what some might term
academic. In short, it can't be put to the test.
Anyone care to pick up that thread? (That invitation extends to you, Mary.)
--
Fred Nickols
The Distance Consulting Company
"Assistance at A Distance"
http://home.att.net/~nickols/distance.htm
nickols@worldnet.att.net
(609) 490-0095