Date: Sun, 15 Sep 2002 20:39:32 GMT
Reply-To: Arthur Tabachneck <atabachneck@ROGERS.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Arthur Tabachneck <atabachneck@ROGERS.COM>
Subject: Re: Comparing regression coefficients to a known value
Hopefully, corporate HR and executives will eventually realize the precise
point you've made. All too often, especially with data mining having been
adopted as one of the more recent panaceas, non-statistical literates are
likely achieving many invalid conclusions.
Fortunately, I'm starting to see a trend in the opposite direction, where at
least some companies are starting to realize that they've just wasted a few
million (or more) on the wrong answers that spit out.
There may yet be hope but, alas, paradigms do take a long time to change.
"Dale McLerran" <stringplayer_2@YAHOO.COM> wrote in message
> Well, there is considerable concern that the original poster is
> swimming in water way over his head when he still needs his water
> wings. I don't usually put it so, but the original question was
> "How do I test whether the confidence limits for a regression
> coefficient contain a certain hypothesized value?". That is the
> most elementary of statistical questions that should have been
> covered in STAT 101. Now, the poster did not know the answer to
> this elementary question. In fact, the poster did not even phrase
> the question as it is phrased above, so there is even more concern
> that his statistical knowlege is not even rudimentary. And yet,
> he is fitting a model in which the intercept is forced through a
> certain value. As discussed here, this can be extremely misguided.
> Truly, it is not the person who is getting slammed, but his approach
> to statistical analysis which is seriously in question. We have
> heard concern expressed by almost everyone who has weighed in on
> the matter that this is of great concern. While the original poster
> did not solicit advice on the subject of forcing the regression
> through a certain point, the collective wisdom of the list has
> raised this as a big red flag.
> Given languages like SAS which make it quite easy for anyone to
> fit all manner of models, it is easy to believe that statistics is
> not a subject matter which requires considerable understanding.
> All that we need to do is plug our data into proc reg/proc glm/etc.
> and let the computer spit out the estimates. This is naive and
> often leads to questionable statistical practice, as in the present
> --- "Meldrum, Sean" <Sean_Meldrum@URMC.ROCHESTER.EDU> wrote:
> > Sorry to interject, but it kinda seems like someone is getting
> > slammed just
> > for asking a question. I think that a lot of people (most, perhaps?)
> > use
> > this list to ask questions about the mechanics of SAS, while others
> > may use
> > it hoping for a statistical consult (even though that isn't the
> > advertised
> > mission of the list.) It wasn't clear to me that statistical advice
> > was
> > requested or warranted.
> > On the other hand, I do enjoy the perennial intercepts discussion - I
> > think
> > good arguments can be made both ways. Perhaps this discussion could
> > be
> > moved to a thread independent of the question that spawned this
> > incarnation.
> > Please pardon me if I'm speaking out of turn.
> > Sean
> Dale McLerran
> Fred Hutchinson Cancer Research Center
> mailto: email@example.com
> Ph: (206) 667-2926
> Fax: (206) 667-5977
> Do you Yahoo!?
> Yahoo! News - Today's headlines