Richard Ulrich showed PROOF of his BUNGLES in Multivariate Regression
Reef Fish
Regression, Richard Ulrich bragged:
RU> Univariate multiple regression makes sense in contrast with
RU> multivariate multiple regression. I've been using the latter
RU> term for years -- Google shows me back to 1997.
RF> BUt did you use those terms CORRECTLY? If so, why would
RF> you be bumbling and stumbling all over the place over those
RF> standard terms NOW?
Indeed, I found a thread in 1996 on "Multivariate Regression" in which
Richard Ulrich participated.
http://groups.google.com/group/sci.stat.math/msg/c39a9c7f0d2c5a47?hl=en&
Of course there was NOTHING Multivariate about those regressions.
They were all UNIVARIATE multiple regressions, MISlabelled as
Multivariate Regression.
All that was discussed in the mislabelled post was variable selection
methods in UNIVARIATE multiple regression, such as the stepwise
procesures and all subsets stuff.
So, my comment
RF> BUt did you use those terms CORRECTLY? If so, why would
RF> you be bumbling and stumbling all over the place over those
RF> standard terms NOW?
was prophetic -- I hit the nail on the head about Richard Ulrich's
bumbling and stumbling.
I actually have an unpublished paper on sequential variable
selction method (algorithm) for Multivariate Regression (of
two or more Dependent Variables).
Instead of using the SSE as the selection criterion for bringing
in (or out) the independent variables, the selection criterion
is the use of the determinant of the covariance matrix of residuals.
Because of the extremely efficient way of updating the covariance
matrix of the residuals (even if there are millions of them), it takes
only THREE arithmetic operations (as in two multiplications and
one division) to find out what the new SSE would be IF any
particular X that is NOT in the model is brought into the equation.
That is accomplished via the use of the SWEEP operator.
I used Al Beaton's SWP, from his 1964 Harvard dissertation,
which I learned from Marty Schatzoff (those and other Beaton
operators). Jim Goodnight of SAS uses a slightly different
version of SWEEP in his variable selection, general linear
models and the computation of pseudoinverses of matrices.
To this day, I am not aware of any existing variable selection
software routine for Multivariate Regression.
One reason is of course not many people even KNOW about
Multivariate Regresssion -- that they SHOULD use, instead
of regression each Y on the same set of X's, and do it m
times for m dependenat variables.
The GAIN in doing a Multivarite Regression over multiple
univariate multiple regressions (which seem intuitive) is
that one gains information and precision on the OFF
DIAGNONAL elements of the covariance matrix of the
residuals!!
Beyond that, there's the Generalized Mulrivariate Regression
in which each dependent variable Y may depend on a
different set of independent variables -- and that's where
the notion of two-stage and three-stage least squares
and other more complicated methods enter.
The folks in THESE newsgroups haven't even scratched
the surface of the REGRESSION problem.
And Richard Ulrich, who has been practicing his Quackery
for YEARS in these groups, don't even know how to tell one
kind of regression from another.
It is the SAD, SAD, SAD, SAD, and MAD, MAD, MAD, MAD
world of sci.stat.*
-- Reef Fish Bob.
Richard Ulrich
<Large_Nass...@yahoo.com> wrote:
> In his lame defense of his ignorance about Multivariate
> Regression, Richard Ulrich bragged:
>
> RU> Univariate multiple regression makes sense in contrast with
> RU> multivariate multiple regression. I've been using the latter
> RU> term for years -- Google shows me back to 1997.
>
> RF> BUt did you use those terms CORRECTLY? If so, why would
> RF> you be bumbling and stumbling all over the place over those
> RF> standard terms NOW?
>
> Indeed, I found a thread in 1996 on "Multivariate Regression" in which
> Richard Ulrich participated.
>
> http://groups.google.com/group/sci.stat.math/msg/c39a9c7f0d2c5a47?hl=en&
>
> Of course there was NOTHING Multivariate about those regressions.
> They were all UNIVARIATE multiple regressions, MISlabelled as
> Multivariate Regression.
Which is this?
Bob's poor reading skills, or Bob's eagerness to attack me?
Combined, I call this Bob's "bad reading".
The Subject line -- from an OP -- was Multivariate regression.
My post did not include those words, nor was I writing about it,
since the subject matter was simpler.
Nor did this 1996 post show up in my count of my posts "back to 1997".
[snip - Bob's personal attack. Also, some interesting
comments on "multivariate regression."]
--
Rich Ulrich, wpi...@pitt.edu
http://www.pitt.edu/~wpilib/index.html
Reef Fish
Richard Ulrich wrote:
> On 23 Sep 2006 22:36:32 -0700, "Reef Fish"
> <Large_Nass...@yahoo.com> wrote:
>
> > In his lame defense of his ignorance about Multivariate
> > Regression, Richard Ulrich bragged:
> >
> > RU> Univariate multiple regression makes sense in contrast with
> > RU> multivariate multiple regression. I've been using the latter
> > RU> term for years -- Google shows me back to 1997.
> >
> > RF> BUt did you use those terms CORRECTLY? If so, why would
> > RF> you be bumbling and stumbling all over the place over those
> > RF> standard terms NOW?
> >
> > Indeed, I found a thread in 1996 on "Multivariate Regression" in which
> > Richard Ulrich participated.
> >
> > http://groups.google.com/group/sci.stat.math/msg/c39a9c7f0d2c5a47?hl=en&
> >
> > Of course there was NOTHING Multivariate about those regressions.
> > They were all UNIVARIATE multiple regressions, MISlabelled as
> > Multivariate Regression.
>
> Which is this?
> Bob's poor reading skills, or Bob's eagerness to attack me?
> Combined, I call this Bob's "bad reading".
>
> The Subject line -- from an OP -- was Multivariate regression.
>
> My post did not include those words, nor was I writing about it,
> since the subject matter was simpler.
>
> Nor did this 1996 post show up in my count of my posts "back to 1997".
Then why don't you just CITE one of those threads in which you
knew the definition of "Multivariate Regression" then, Richard?
Would it be so much simplier than your incessant arguing WITHOUT
coming up with one textbook definition of "Multivariate Regression"?
> --
> Rich Ulrich, wpi...@pitt.edu
> http://www.pitt.edu/~wpilib/index.html
-- Reef Fish Bob.