Re: Linear regression

On 31 May 2006 09:58:58 -0700, "Jens" <jens.stolte@xxxxxxxxx> wrote:

Thanks Bob for your prompt answer.

[snip, some]

Bob >
<skip>
By "couplings" I think you mean either statistical or deterministic
relations some pairs of X's. Short answer, as long as the X's are linearly
independent, there's nothing to worry about.
Jens >
By couplings I mean some unknown relations between the columns. Unknown
in the sence that no-one have investigated such relations properly, but
based on "intelligent" guess some relations surely exists.


Does that matter? Maybe, maybe not.

Reef Fish Bob belongs to a "black-box" school of regression --
the statistician has no knowledge of control of any of the
many variables going into his equation, and no care for what
it may "mean". [And, in his ignorance, Bob is exceeding proud
that his way is the only right way.]

That is hardly ever the attitude in the sciences, but it is the
way business schools can approach the problem of predicting
who is the good insurance risk, etc. What is needed, mainly,
is a large dataset, and no concern for *wide* generalization
(that is, going beyond datasets that are rather similar).

If you want "meaningful" regressions -- an idea that is anathema
to Bob -- you will have a rather small set of variables, which
you might "reduce" even further before taking into the regression.


What is your purpose? Are you exploring, verifying, developing
on someone else's work?

What is your N? ... your number of predictors, your expected
R-squared? Do you have a choice of what variables to use,
or what functional forms of them? With Large-N and large-P
and no choice in the variables, you should follow Bob's
lead -- run the regression and study the diagnostics....

[snip]


--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.

Relevant Pages

  • Re: Contempt for Creationism
    ... snip ...] ... in friendly discours with our resident advocate for mass murder? ... I don't take Bob that seriously. ... to comment on an insightful criticism of one of my posts because Bob was ...
    (talk.origins)
  • Re: Barenboims CSO farewell
    ... Your remarks are an embarrassment to the CSO, ... passed to you as you left and orchestra which he has served for many, ... Do you think he would appreciate that comment, Bob? ...
    (rec.music.classical.recordings)
  • Re: note to Boon: High End vs. High Fidelity
    ... The Sansui 2000 might not be my preference, ... your opinions with my superior standing. ... I stand over you, bob. ... [snip adhominem slur] ...
    (rec.audio.opinion)
  • Re: Hamiltons Rule: light at the end of a LONG tunnel?
    ... > Bob O'Hara wrote: ... > "Hamilton requires the number of organism recipients ... > existing testable theory) for the 4 classes ...
    (sci.bio.evolution)
  • Re: Conversion factor for lab values
    ... didn't I just reply to "Bob" aka Robert1? ... No one knows what the optimal ratios are. ...
    (sci.med)