David A. Heiser wrote:
"Reef Fish" <large_nassua_grouper@yahoo.com> wrote in message
news:1165596313.275506.72380@n67g2000cwd.googlegroups.com...

Greg Heath wrote:
... So in summary,

1. Greg Heath claims

It can be useful to inspect the components of Cxx
and Cyx, submatrices of the all-variable correlation
coefficient matrix, to help understand the relationships
among variables.

The most useful time is when multicollinearities are present.

That is the LEAST useful time.

A simple eigenvalue
eigenvector analysis and the use of model-building
technique of eliminating the REDUNDUNT variable
in the equation is all that's necessary as the CURE.

++++++++++++++++++++++++++++++++++++
-----SNIP
Consequently it is a
matter of eyeball interpretation of a Scree plot. When there are many
variables (common in social research), and the Scree plot right tail is
generally decreasing without a defined break, there are different
intreptations by different people on where the cutoff point should be. See
Karyongo in Journal of Modern Applied Statistical Methods, Vol. 4, #1, page
120.

What's a "Scree plot"?

In the method of eliminating REDUNDANT variables in a multiple
regression, there are very well-defined and analytic methods of
examining the partial correlations in deciding which is the variable
to REMOVE that will least affect the FIT.

No one is talking about the cut off point or what value to use.

I think you are jumping into the middle of the SUMMARY Greg and
I were discussing about OUR points about multicollinearity and
its accommodation with something that is completely off the wall
on YOUR part.

If you don't use a covariance matriz, but the intial data matrix then you
may get complex eigenvalues, which actually are more informative (in polar
form), but can't be nicely plotted.as a Scree plot.

You are talking through your HAT, and I don't mean the HAT matrix
associated with a regression problem or any of the regression
estimates that are called the beta-hats either.

1. Greg Heath claims

It can be useful to inspect the components of Cxx
and Cyx, submatrices of the all-variable correlation
coefficient matrix, to help understand the relationships
among variables.
The most useful time is when multicollinearities are present.

If a simple explanation is needed, sometimes using
a display of [Cxx Cyx] is helpful.

++++++++++++++++++++++++++++++++++++
-----SNIP

Consequently it is a
matter of eyeball interpretation of a Scree plot. When there are many
variables (common in social research), and the Scree plot right tail is
generally decreasing without a defined break, there are different
intreptations by different people on where the cutoff point should be. See
Karyongo in Journal of Modern Applied Statistical Methods, Vol. 4, #1, page
120.

RF > What's a "Scree plot"?

GH> What you have termed an elbow plot: SSE/(n-p+1) vs p.

Some have chosen the cutoff w.r.t. where the plot has
maximum curvature.

In the method of eliminating REDUNDANT variables in a multiple
regression, there are very well-defined and analytic methods of
examining the partial correlations in deciding which is the variable
to REMOVE that will least affect the FIT.

I think David is referring to PCR and which PCs to remove.

RF> > No one is talking about the cut off point or what value to use.

GH> True.

I think you are jumping into the middle of the SUMMARY Greg and
I were discussing about OUR points about multicollinearity and
its accommodation with something that is completely off the wall
on YOUR part.

Lighten up Reef. You can't expect every replier to have a clear
picture of our months-long disagreement . He's a little OT. So
what?

If you don't use a covariance matriz, but the intial data matrix then you
may get complex eigenvalues, which actually are more informative (in polar
form), but can't be nicely plotted.as a Scree plot.

?? Eigenvalues of a nonsquare data matrix ??

You are talking through your HAT, and I don't mean the HAT matrix
associated with a regression problem or any of the regression
estimates that are called the beta-hats either.

Yes David, you seem to be way out in left field with that last
paragraph.

Greg

Greg Heath wrote:

GH> ... So in summary,

GH> 1. Greg Heath claims
see claim below
GH> 2. Reef Fish claims
see claim below
GH> Hope that ends this thread.

I AGREED that both claims were CORRECT, and made
a brief summary of why you were wrong. The rest are
ALL in the archives.

I thought for a minute you were back arguing again. But I see
you added a summary comment to repeat your claim, so I assume
that OUR thread of disagreement has indeed ENDED.

Each reader can go back to the archives and read what we each
had to say.

1. Greg Heath claims

It can be useful to inspect the components of Cxx
and Cyx, submatrices of the all-variable correlation
coefficient matrix, to help understand the relationships
among variables.
The most useful time is when multicollinearities are present.

I agreed that was your claim and merely explained WHY
RF > That is the LEAST useful time.

If a simple explanation is needed, sometimes using
a display of [Cxx Cyx] is helpful.

That is a repeat of the same claim.
++++++++++++++++++++++++++++++++++++
-----SNIP

The rest of your post was about David Heiser's unwarranted
comments when he jumped into the middle of our disagreement
without even knowing that we were discussing ordinary
multiple REGRESSION!

I said he was "talking through his HAT"

and you said he was "way out in left field"

so we AGREED at least on that part, and let's see David Heiser's
worthless as well as improper comments:

David Heiser wrote,
Consequently it is a
matter of eyeball interpretation of a Scree plot. When there are many
variables (common in social research), and the Scree plot right tail
is
generally decreasing without a defined break, there are different
intreptations by different people on where the cutoff point should
be. See
Karyongo in Journal of Modern Applied Statistical Methods, Vol. 4,
#1, page
120.

RF > What's a "Scree plot"?

GH> What you have termed an elbow plot: SSE/(n-p+1) vs p.

Then why not call it so? The term "Scree plot" was never used in any
previous discussion. Why introduce it at this point? Besides, my
"elbow plot" was the Standard deviation of residuals (sqrt of MSE)
vs p (the number of independent variables).

The MSE of a regression is SSE/(n-p-1).

Was the SSE/(n-p+1) your error/typo or Karyongo's/Heiser's error?

Some have chosen the cutoff w.r.t. where the plot has
maximum curvature.

Curvature is undefined. "small marginal decrease" and "increase"
are well-defined and well-understood terms.


In the method of eliminating REDUNDANT variables in a multiple
regression, there are very well-defined and analytic methods of
examining the partial correlations in deciding which is the variable
to REMOVE that will least affect the FIT.

I think David is referring to PCR and which PCs to remove.

PCR was never mentioned. WE were discussion ordinary multiple
regression and David Heiser jumped in with a completely nonsensical
comment:

DH> There are no p values associated with eigenvalues. Consequently
DH> it is a matter of eyeball interpretation of a Scree plot.

RF> > No one is talking about the cut off point or what value to use.

GH> True.

I think you are jumping into the middle of the SUMMARY Greg and
I were discussing about OUR points about multicollinearity and
its accommodation with something that is completely off the wall
on YOUR part.

Lighten up Reef. You can't expect every replier to have a clear
picture of our months-long disagreement . He's a little OT. So
what?

So what? I expect everyone who BUTTS into someone else's
disagreement to know at least what the PROBLEM was, let
alone what the disagreement was.

So. if he jumps into a month-long disagreement, he should at least
KNOW what the disagreement was about!

That is his RESPONSIBILITY as a responsible discussant.

So what? So, he should have kept his mouth SHUT.



If you don't use a covariance matriz, but the intial data matrix then
you
may get complex eigenvalues, which actually are more informative (in
polar
form), but can't be nicely plotted.as a Scree plot.

?? Eigenvalues of a nonsquare data matrix ??

You are talking through your HAT, and I don't mean the HAT matrix
associated with a regression problem or any of the regression
estimates that are called the beta-hats either.

Yes David, you seem to be way out in left field with that last
paragraph.

Greg

So, I take it that OUR disagreement STOOD and we agreed to have
no more argument about the disagreement and let the readers read
what is in the archives.

That's a MAJOR step toward progress. Richard Ulrich should take
very careful note!

-- Reef Fish Bob.
+++++++++++++++++++++++++++++++++++++++++++++++

"Reef Fish" <large_nassua_grouper@yahoo.com> wrote in message
news:1165758550.122579.111320@j72g2000cwa.googlegroups.com...

Greg Heath wrote:

GH> ... So in summary,

GH> 1. Greg Heath claims
see claim below
GH> 2. Reef Fish claims
see claim below
GH> Hope that ends this thread.

I AGREED that both claims were CORRECT, and made
a brief summary of why you were wrong. The rest are
ALL in the archives.

I thought for a minute you were back arguing again. But I see
you added a summary comment to repeat your claim, so I assume
that OUR thread of disagreement has indeed ENDED.

Each reader can go back to the archives and read what we each
had to say.

1. Greg Heath claims

It can be useful to inspect the components of Cxx
and Cyx, submatrices of the all-variable correlation
coefficient matrix, to help understand the relationships
among variables.
The most useful time is when multicollinearities are present.

I agreed that was your claim and merely explained WHY
RF > That is the LEAST useful time.

If a simple explanation is needed, sometimes using
a display of [Cxx Cyx] is helpful.

That is a repeat of the same claim.
++++++++++++++++++++++++++++++++++++
-----SNIP

The rest of your post was about David Heiser's unwarranted
comments when he jumped into the middle of our disagreement
without even knowing that we were discussing ordinary
multiple REGRESSION!

I said he was "talking through his HAT"

and you said he was "way out in left field"

so we AGREED at least on that part, and let's see David Heiser's
worthless as well as improper comments:

David Heiser wrote,
Consequently it is a
matter of eyeball interpretation of a Scree plot. When there are many
variables (common in social research), and the Scree plot right tail
is
generally decreasing without a defined break, there are different
intreptations by different people on where the cutoff point should
be. See
Karyongo in Journal of Modern Applied Statistical Methods, Vol. 4,
#1, page
120.

RF > What's a "Scree plot"?

GH> What you have termed an elbow plot: SSE/(n-p+1) vs p.

Then why not call it so? The term "Scree plot" was never used in any
previous discussion. Why introduce it at this point? Besides, my
"elbow plot" was the Standard deviation of residuals (sqrt of MSE)
vs p (the number of independent variables).

The MSE of a regression is SSE/(n-p-1).

Was the SSE/(n-p+1) your error/typo or Karyongo's/Heiser's error?

Some have chosen the cutoff w.r.t. where the plot has
maximum curvature.

Curvature is undefined. "small marginal decrease" and "increase"
are well-defined and well-understood terms.


In the method of eliminating REDUNDANT variables in a multiple
regression, there are very well-defined and analytic methods of
examining the partial correlations in deciding which is the variable
to REMOVE that will least affect the FIT.

I think David is referring to PCR and which PCs to remove.

PCR was never mentioned. WE were discussion ordinary multiple
regression and David Heiser jumped in with a completely nonsensical
comment:

DH> There are no p values associated with eigenvalues. Consequently
DH> it is a matter of eyeball interpretation of a Scree plot.

RF> > No one is talking about the cut off point or what value to use.

GH> True.

I think you are jumping into the middle of the SUMMARY Greg and
I were discussing about OUR points about multicollinearity and
its accommodation with something that is completely off the wall
on YOUR part.

Lighten up Reef. You can't expect every replier to have a clear
picture of our months-long disagreement . He's a little OT. So
what?

So what? I expect everyone who BUTTS into someone else's
disagreement to know at least what the PROBLEM was, let
alone what the disagreement was.

So. if he jumps into a month-long disagreement, he should at least
KNOW what the disagreement was about!

That is his RESPONSIBILITY as a responsible discussant.

So what? So, he should have kept his mouth SHUT.



If you don't use a covariance matriz, but the intial data matrix then
you
may get complex eigenvalues, which actually are more informative (in
polar
form), but can't be nicely plotted.as a Scree plot.

?? Eigenvalues of a nonsquare data matrix ??

You are talking through your HAT, and I don't mean the HAT matrix
associated with a regression problem or any of the regression
estimates that are called the beta-hats either.

Yes David, you seem to be way out in left field with that last
paragraph.

Greg

So, I take it that OUR disagreement STOOD and we agreed to have
no more argument about the disagreement and let the readers read
what is in the archives.

That's a MAJOR step toward progress. Richard Ulrich should take
very careful note!

-- Reef Fish Bob.
+++++++++++++++++++++++++++++++++++++++++++++++
I'm sorry that I didn't know that the sci.stat.* groups are only there to
provide a vehicle for Reef Fish, and is closed to everyone else.

Apparently
only those "selected" responders are allowed to debate.

The territory and
views are very narrow. No references outside of the "circle". No reading of
current ASA publications or of books outside of the "circle of Reef Fish's
memory".

David Heiser

Greg Heath wrote:

The rest of your post was about David Heiser's unwarranted
comments when he jumped into the middle of our disagreement
without even knowing that we were discussing ordinary
multiple REGRESSION!

I said he was "talking through his HAT"

and you said he was "way out in left field"

so we AGREED at least on that part, and let's see David Heiser's
worthless as well as improper comments:

David Heiser wrote,
Consequently it is a
matter of eyeball interpretation of a Scree plot. When there are many
variables (common in social research), and the Scree plot right tail is
generally decreasing without a defined break, there are different
intreptations by different people on where the cutoff point should be. See
Karyongo in Journal of Modern Applied Statistical Methods, Vol. 4, #1, page
120.

RF > What's a "Scree plot"?

GH> What you have termed an elbow plot: SSE/(n-p+1) vs p.

Then why not call it so? The term "Scree plot" was never used
in any previous discussion. Why introduce it at this point?

Besides, my "elbow plot" was the Standard deviation of residuals
(sqrt of MSE) vs p (the number of independent variables).

The MSE of a regression is SSE/(n-p-1).

RF > What's a "Scree plot"?

GH> The square of what you have termed an elbow plot: SSE/(n-p-1) vs p.

Was the SSE/(n-p+1) your error/typo or Karyongo's/Heiser's error?

Some have chosen the cutoff w.r.t. where the plot has
maximum curvature.

Curvature is undefined. "small marginal decrease" and "increase"
are well-defined and well-understood terms.

In the method of eliminating REDUNDANT variables in a multiple
regression, there are very well-defined and analytic methods of
examining the partial correlations in deciding which is the variable
to REMOVE that will least affect the FIT.

I think David is referring to PCR and which PCs to remove.

PCR was never mentioned. WE were discussion ordinary multiple
regression and David Heiser jumped in with a completely nonsensical
comment:

DH> There are no p values associated with eigenvalues. Consequently
DH> it is a matter of eyeball interpretation of a Scree plot.

RF> > No one is talking about the cut off point or what value to use.

GH> True.

I think you are jumping into the middle of the SUMMARY Greg and
I were discussing about OUR points about multicollinearity and
its accommodation with something that is completely off the wall
on YOUR part.

Lighten up Reef. You can't expect every replier to have a clear
picture of our months-long disagreement . He's a little OT.
So what?

So what? I expect everyone who BUTTS into someone else's
disagreement to know at least what the PROBLEM was, let
alone what the disagreement was.

So. if he jumps into a month-long disagreement, he should at least
KNOW what the disagreement was about!

That is his RESPONSIBILITY as a responsible discussant.

So what? So, he should have kept his mouth SHUT.

If you don't use a covariance matriz, but the intial data matrix then you
may get complex eigenvalues, which actually are more informative (in polar
form), but can't be nicely plotted.as a Scree plot.

?? Eigenvalues of a nonsquare data matrix ??

You are talking through your HAT, and I don't mean the HAT matrix
associated with a regression problem or any of the regression
estimates that are called the beta-hats either.

Yes David, you seem to be way out in left field with that last
paragraph.

Reef Fish wrote:
Greg Heath wrote:

-----SNIP

The rest of your post was about David Heiser's unwarranted
comments when he jumped into the middle of our disagreement
without even knowing that we were discussing ordinary
multiple REGRESSION!

I said he was "talking through his HAT"

and you said he was "way out in left field"

so we AGREED at least on that part, and let's see David Heiser's
worthless as well as improper comments:

David Heiser wrote,
Consequently it is a
matter of eyeball interpretation of a Scree plot. When there are many
variables (common in social research), and the Scree plot right tail is
generally decreasing without a defined break, there are different
intreptations by different people on where the cutoff point should be. See
Karyongo in Journal of Modern Applied Statistical Methods, Vol. 4, #1, page
120.

RF > What's a "Scree plot"?

GH> What you have termed an elbow plot: SSE/(n-p+1) vs p.

Corrected typo: SSE/(n-p-1) vs p.

Then why not call it so? The term "Scree plot" was never used
in any previous discussion. Why introduce it at this point?

1.Because PCA texts (Jackson and/or Jollife ?) call it a scree plot.

2. AFAIK the term "elbow" was coined by you

3. David probably never read your elbow post

Besides, my "elbow plot" was the Standard deviation of residuals
(sqrt of MSE) vs p (the number of independent variables).

The MSE of a regression is SSE/(n-p-1).

Sorry. Let's back up:

RF > What's a "Scree plot"?

GH> The square of what you have termed an elbow plot: SSE/(n-p-1) vs p.

Satisfied?

Was the SSE/(n-p+1) your error/typo or Karyongo's/Heiser's error?

My typo. It was never mentioned by Dave and I didn't look at the
K reference.

Some have chosen the cutoff w.r.t. where the plot has
maximum curvature.

Curvature is undefined. "small marginal decrease" and "increase"
are well-defined and well-understood terms.

Curvature of a smooth curve is well defined in 1st year calculus. To
use with data, estimate 2nd derivatives with 2nd differences.

Greg

3. David probably never read your elbow post

David never read 99% of the discussions related to multiple
regression in these groups. I even pointed him to read some of
the threads in the archives. He is just like Ulrich in that regard.

On 11 Dec 2006 14:56:27 -0800, "Reef Fish"
large_nassua_grouper@yahoo.com> wrote:

[snip]

3. David probably never read your elbow post

David never read 99% of the discussions related to multiple
regression in these groups. I even pointed him to read some of
the threads in the archives. He is just like Ulrich in that regard.

Ulrich, like everyone else, finds it painful to look back
at any old Reef Fish posts because Reef Fish

By the way, I recognized the idea of the elbow plot
immediately, though it was mainly familiar to me as a
scree plot. I also remembered an earlier discussion, which
I find in sci.stat.math, Googling groups for
group:sci.stat.* elbow scree >.

Jerry Dallal Replied to Bob on Feb. 10 --
=== JD, near the beginning of Reply to RF.
Kruskal's "Elbow Rule Plots" are called Scree Plots in some quarters,
"scree" being "A slope of loose rock debris at the base of a steep
incline or cliff." The plots are not uncommon. See, for example, p477
of Neter, Wasserman, and Kutner "Applied Linear Statistical Models,
3rd" for the plot, but not the name. ...
=== end of excerpt.

Reef Fish Bob's messages over the year
(and prior years) have stressed to fact that linear regression depends on
the predictor variables being independent.

Richard Ulrich wrote:
On 11 Dec 2006 14:56:27 -0800, "Reef Fish"
large_nassua_grouper@yahoo.com> wrote:

[snip]

3. David probably never read your elbow post

David never read 99% of the discussions related to multiple
regression in these groups. I even pointed him to read some of
the threads in the archives. He is just like Ulrich in that regard.

Ulrich, like everyone else, finds it painful to look back
at any old Reef Fish posts because Reef Fish

pointed out so many of Richard Ulrich's statistical ERRORS and
his malpractice of statistics.

You apparently need to go back to the start, and learn what
these tests do, and something about how they do it.
Pick up a textbook or two and browse.

Greg Heath wrote:
Reef Fish wrote:
Greg Heath wrote:

-----SNIP

The rest of your post was about David Heiser's unwarranted
comments when he jumped into the middle of our disagreement
without even knowing that we were discussing ordinary
multiple REGRESSION!

I said he was "talking through his HAT"

and you said he was "way out in left field"

so we AGREED at least on that part, and let's see David Heiser's
worthless as well as improper comments:

David Heiser wrote,
Consequently it is a
matter of eyeball interpretation of a Scree plot. When there are many
variables (common in social research), and the Scree plot right tail is
generally decreasing without a defined break, there are different
intreptations by different people on where the cutoff point should be. See
Karyongo in Journal of Modern Applied Statistical Methods, Vol. 4, #1, page
120.

RF > What's a "Scree plot"?

GH> What you have termed an elbow plot: SSE/(n-p+1) vs p.

Corrected typo: SSE/(n-p-1) vs p.

Then why not call it so? The term "Scree plot" was never used
in any previous discussion. Why introduce it at this point?

1.Because PCA texts (Jackson and/or Jollife ?) call it a scree plot.

I see you've been reading the wrong books too. Besides, nobody was
talking about PCA except David Heiser.

2. AFAIK the term "elbow" was coined by you

No. I borrowed that term from Joe Kruskal (or Multidimensional
Scaling fame) who recommended using the "Elbow Rule" to
decide on the number of dimensions to use, his his Stress plot
vs dimension. Obvious, the idea I used in choosing the number
of variables to use is the same (look at the marginal decrease to
see when the diminishing return is not worth an additional variable).

When I was talking about the elbow rule, several of the discussants
mentioned they saw that kind of plots in STATISTICAL textbooks,
as soon as I pointed out one. m00es (remember him?) said he
recognized the DATA immediately in the example I cited such a
plot in a textbook.

3. David probably never read your elbow post

David never read 99% of the discussions related to multiple
regression in these groups. I even pointed him to read some of
the threads in the archives. He is just like Ulrich in that regard.

Besides, my "elbow plot" was the Standard deviation of residuals
(sqrt of MSE) vs p (the number of independent variables).

The MSE of a regression is SSE/(n-p-1).

Sorry. Let's back up:

RF > What's a "Scree plot"?

GH> The square of what you have termed an elbow plot: SSE/(n-p-1) vs p.

Satisfied?

You already answered it in this post above. I asked the question only
ONCE. Are you just repeating the same clip and answered it twice, and
threw in your smart ass remark "Satistified?" the second time.

Was the SSE/(n-p+1) your error/typo or Karyongo's/Heiser's error?

My typo. It was never mentioned by Dave and I didn't look at the
K reference.

He also misspelled the name of the author -- not that it would make a
bit of different who that clown is, Koyongo or Karyongo.

Some have chosen the cutoff w.r.t. where the plot has
maximum curvature.

Curvature is undefined. "small marginal decrease" and "increase"
are well-defined and well-understood terms.

Curvature of a smooth curve is well defined in 1st year calculus. To
use with data, estimate 2nd derivatives with 2nd differences.

I see you're still at the 1st year calculus level? Never kind what
you
said is an ABSURD idea to use even if the curve IS continuouly
differentiable. The CURVE to which the "Elbow rule" (Kruskal's
or mind) is a piecewise linear function with joints (without
derivatives) at the dimensional points.

If David Heiser want to show his ignorance, let HIM -- I am suggesting
that to spare you of further nonsense such as the curvature and
second derivatives on piecewise linear functions.

Hmm, I thought I remember the term "scree" used in Kruskal's
MDS book also.

If David Heiser want to show his ignorance, let HIM -- I am suggesting
that to spare you of further nonsense such as the curvature and
second derivatives on piecewise linear functions.

Hey, don't shoot the messenger.

Greg

Hmm, I thought I remember the term "scree" used in Kruskal's
MDS book also.

I learned MDS from Joe Kruskal himself, in 1967, shortly after his
publication of his 1963 papers on MDS.

He used the "Elbow Rule" then.

What BOOK on MDS did he write?

"Reef Fish" <large_nassua_grouper@yahoo.com> wrote in news:1165937871.090284.265210
@f1g2000cwa.googlegroups.com:

Hmm, I thought I remember the term "scree" used in Kruskal's
MDS book also.

I learned MDS from Joe Kruskal himself, in 1967, shortly after his
publication of his 1963 papers on MDS.

He used the "Elbow Rule" then.

What BOOK on MDS did he write?

http://www.amazon.com/exec/obidos/ASIN/0803909403
http://www.amazon.co.uk/exec/obidos/ASIN/0803930682

David Winsemius wrote:
"Reef Fish" <large_nassua_grouper@yahoo.com> wrote in
news:1165937871.090284.265210 @f1g2000cwa.googlegroups.com:

Hmm, I thought I remember the term "scree" used in Kruskal's
MDS book also.

I learned MDS from Joe Kruskal himself, in 1967, shortly after his
publication of his 1963 papers on MDS.

He used the "Elbow Rule" then.

What BOOK on MDS did he write?

http://www.amazon.com/exec/obidos/ASIN/0803909403
http://www.amazon.co.uk/exec/obidos/ASIN/0803930682

David, you came to this thread quite a bit late. The use of "scree"
had been thoroughly identified in 5 or 6 different areas.

Go back up the Google tree of this thread (44 articles) and review
what had been discussed and clarified already.

If the word "scree" was ever used in the 1978 book co-authored by
Kruskal, I am SURE it was NOT used as a substitute (or even related
to Kruskal's use) of the "Elbow Rule" he used in his MDSCAL and MDS
published and unpublished articles.

"Reef Fish" <large_nassua_grouper@yahoo.com> wrote in
news:1166373175.068176.129360@73g2000cwn.googlegroups.com:


David Winsemius wrote:
"Reef Fish" <large_nassua_grouper@yahoo.com> wrote in
news:1165937871.090284.265210 @f1g2000cwa.googlegroups.com:

Hmm, I thought I remember the term "scree" used in Kruskal's
MDS book also.

I learned MDS from Joe Kruskal himself, in 1967, shortly after his
publication of his 1963 papers on MDS.

He used the "Elbow Rule" then.

What BOOK on MDS did he write?

http://www.amazon.com/exec/obidos/ASIN/0803909403
http://www.amazon.co.uk/exec/obidos/ASIN/0803930682

David, you came to this thread quite a bit late. The use of "scree"
had been thoroughly identified in 5 or 6 different areas.

Go back up the Google tree of this thread (44 articles) and review
what had been discussed and clarified already.

If the word "scree" was ever used in the 1978 book co-authored by
Kruskal, I am SURE it was NOT used as a substitute (or even related
to Kruskal's use) of the "Elbow Rule" he used in his MDSCAL and MDS
published and unpublished articles.


I wrote _absolutely_nothing_ regarding "elbows" or "scree". You were
responding to Greg Heath, remember? You asked what book on MDS Kruskal
had written. I answered. I made no statements about what might or might
not be in those books.

Searching on "Kruskal, book" in sci.stat.math, I see that one of the two
was referenced in a different thread.

------
We now return you to our regular failure to communicate.

--
David Winsemius