Re: Experienced Statistician to help decide whether a regression is legitimate
- From: "Reef Fish" <large_nassua_grouper@xxxxxxxxx>
- Date: 27 Oct 2006 06:36:21 -0700
Anahita wrote:
Indeed probably so...
however in the industry it is also the lack of time & hierarchy constraints (you cannot contradict your superior) that prevents (potentially not so untalented and even clever) people to take the time to formalize the problem they should tackle...or that pushes them to autocensure themselves..
That's one of the many excuses that was used by one of the discussants
in this group,
who had argued and argued and argued about his invalid use of the
correlation, and
finally attempted to flame his own errors on his superior. It's NOT a
valid excuse to
malpractice statistics.
Moreover distinguishing good from bad in applied work is time consuming.. Since most management is lay-men in stats-i.e. the rationale if it looks like stats it smells like it.. then it is stats-- is seldom "invalidated"...
There have been Executive Programs in Business Schools throughout the
country, in their MBA programs in which ONLY high level managers and
company CEOs are admitted. They are taught not only how to recognize
what is good and what is bad, but actually DO many of the analysis
themselves. My co-author and former colleague Harry V. Roberts of
the University of Chicago had been a strong moving force in that
program
at Chicago as well as in the movement to educate the Management. He
even taught principles management that are much more management than
statistics.
http://chronicle.uchicago.edu/041007/obit-roberts.shtml
#> Toward the end of his career, Roberts helped develop a curriculum
in
#> Total Quality Management at the GSB.
Harry even had succeess in pursuading several major companies to
REQUIRE their upper level management positions to be filled by only
those who had several advanced courses in statistics.
Applied statistics at least "seems" a field where "every(many)body feels "I" can do it provided "I" have a good number crunching software.." (less a problem in more theoretical branches where.... you have to prove something without a computer.)
That's the common ILL of statistics today. That's one reason I find
so
many incompetent statistics in this group and the TOLERANT attitude
of the readership. They are much more ready to criticize ME for
pointing out the malpractice than criticizing those who malpractice.
In contrast I've always felt applied work is difficult and that one needs a real talent and quite a bit of rigor to do that..
Of course it does. It is more difficult to be a top level APPLIED
statistician
than a brain surgeon. If a brain surgeon is obviously incompetent,
his
patients DIE and is easily observable. :-) When applied
statisticians
malpractice and have strong influence on CEOs of companies, the
effect is not as readily visible. That's why ECONOMISTS have been
practicing Quackery much longer than statisticians, and often using the
same malpractice techniques, such as misunderstanding the SIGNS of
multiple regression coefficients.
I believe that for some real problems(criminology, proving that price of assets have been manipulated, and many others..) "quality statistical treatment" can bring some light to the debate (while bad statistical doesn't, or only in the sense that it points to what should'nt be done)...so I do hope to get a chance to discuss with such people- as they exist and have expertise!
Nevertheless I think one should supersede the debate.. of who is good or bad...it is useless and brings nowhere..
Companies ALWAYS have the option to hire competent statisticians to
help do the job their own in-house statistics staff is incapble of
handling
properly or well.
In french there is a say: "l'art est difficile et la critique aisée".. art is difficult but critique is easy...
C'est très vrai. L'art des statistiques et de l'analyse de dta est
difficile.
There are and will be good people around everywhere, and their advice & knowledge is useful!
p.s. many books in statistics contradict each other which makes it pretty though if you are not marinating in the field all the time.
One of the things one learns is how to recognize a BAD book. One
of the most touted books by sociologists, and often quoted by
Richard Ulrich as if it were the bible of statistics, even made such
a BLUNDER as calling "Type II error" as a probability NUMBER.
The rest of what had been cited from that book were all statistical
GARBAGE. That explains in part why there are so many
malpracticing sociologists in statistics.
-- Reef Fish Bob.
To reply to some other poster:
-----------------------------
I cannot walk away from this mess (alas and even if I'd very much like to).
More I will have to propose something potentially improving (...sic).
I'll settle to "less (pseudo) sophisticated treatment" but more sensible....for instance a few well constructed descriptive statistics... at least it will oblige them to sit down and think and maybe then they will come up with something better....
.
- Follow-Ups:
- References:
- Prev by Date: Re: different priors (flat, uniform, etc)
- Next by Date: Re: p-value
- Previous by thread: Re: Experienced Statistician to help decide whether a regression is legitimate
- Next by thread: Re: Experienced Statistician to help decide whether a regression is legitim
- Index(es):
Relevant Pages
|
