Raja's Question and Answer on R and MSE RE-VISITED


Raja wrote:
m00es and Reef,
From my own end I had closed this thread and came back to read it
today. I was taken aback by the rhetoric that has been going on here.

You CAN learn from my LATEST two posts, about the Preview to
Reef Fish Statistics for Dummies, and look at the numerical examples
that illustrated what ASSUMPTION VALIDATION is all about, which
is the one single topic m00es has erred in everything he had ever
posted - with no exception!


But I am sure that u guys are good friends actually and having some
sort of fun through this dialogue.

I am100% sure you are WRONG, just as I am 100% sure all the
wrong things m00es posted. :-)


I am not a statistician. I just like statistics and like to learn it by
intuition. Given this I cannot really argue about anything as well as I
am not even familiar with the jargon and terminology.

That is understandable. You asked a very good question about
whether MSE and R can both be "large" in a thread you started
and had since been thoroughly polluted by m00es.

In particular, m00es gave you the WRONG answer to your
intended question, citing formulas -- that's the only thing he knows.
He knows NOTHING about Applied Statistics and what one sees
and does in it.


Reefs, explanation helped me a bit. I couldnot understand him FULLY but
thats my fault as I dont have the right background. However, he made
his point and I got what I needed to know. Now what m00es said in his
formula baffles me again. It seems that if R^2 is high, mse should
always be low and vice versa.

That's what I mean by he knows ONLY formulas.


However, I have a numeber of datasets on
all of which I observed the converse. Some were multiple regression
problems and some were ordinary single variable regression problems.

You observed something contrary to that formula relation in DIFFERENT
regression problems. That was what I understood you to mean and
explained to you the reasons.

You can still review what I said in the SECOND post of this thread,
following your Opening Post:

http://groups.google.com/group/sci.stat.math/msg/193ac98990cad9a5

This example was given in my Preview Post to Validation:

This is the result:
X Y FITTED RESIDU
** *** ************ ******
1 1 -11 12
2 4 3.5527E-15 4
3 9 11 -2
4 16 22 -6
5 25 33 -8
6 36 44 -8
7 49 55 -6
8 64 66 -2
9 81 77 4
10 100 88 12

RF> It even has an R^2 that would have impressed Richard Ulrich. :-)

RF> R_SQ = .94976 R = .97456

RF> But the regression model is TOTALLY wrong,

I didn't mention the MSE of that example was SSE/9 = 58.667,
which was HIGH relative to the data. For X = 1, the fitted
error was 1,200%. For X = 10, the error was 120%,


This was quite strange to me and reef managed to help. m00es's equation
is confusing me again (although I still have to fully conceive the
equation).

m00es understood only that one equation within the same problem.
You can also review my comment on his post that misled you:

http://groups.google.com/group/sci.stat.math/msg/46372c717005b9ae

In that post, I even gave your the compliment you deserved:

RF> In that regard, the OP is infinitely more experienced and
RF> knowledgeable about APPLIED statistics than m00es ever will.

You noticed the phenomenon caused by "spurious correlations" and
you asked about it. I am sure I've pointed to the 1975 SPSS Manual
example in which the R's (in simple or multiple regression) were
ALL very high (in the high .9s) while the MSE's are also high
(in practical terms to render the fitted models USELESS).

That's the kind of experience m00es LACKS totally.

He has NO experience in applied statistics whatsoever.

He has some training in mathemtical statistics, and that's all he
knows, some formulas that are NOT applicable in application
contexts.

YOUR problem about MES and R is one of them. The simple
regression test and the validation of assumptions is the other.

m00es has scored a total of ZERO in those topics. He scored
minus infinity in the amount of noise and pollution he has
created.

Regards,
Adil Raja

-- Reef Fish Bob.

P.S. Raja's OP is in Post #1 of Google's thread:
"Relationship between pearson's correlation coeffcient and sigma"
My initial reply was Post #2. m00es's NOISE started at Post #6
of the same thread and lasted till Post #24. Raja's current post
is Post #25. My present reply should be Post #26.

.

Relevant Pages

  • m00ess NOISE explained because he is REPEATING the SAME ad infinitum
    ... You made nothing but NOISE for weeks because you did not ... Regression, before any statistical test or inference can be ... You are ASHAMED of the name m00es EXPOSED in relation to ... to issues of APPLIED statistics, even given the fact that he ...
    (sci.stat.math)
  • Re: Relationship between pearsons correlation coeffcient and sigma
    ... and a higher R value (and low mse and low R value)? ... This is Post #6 in the thread, the debut by m00es, ... in hundreds of posts) that m00es had NEVER done any regression ... statistical substance in this thread that Raja started and re-appeared ...
    (sci.stat.math)
  • Re: HELP: Non-linear regression problem
    ... over the standard regression methods and outputs. ... Jacobian to calculate the next estimate (assuming that's what you're ... In the Nonlinear Regression context, ... That was one of my "consulting projects" at Yale Statistics Department ...
    (sci.stat.math)
  • Re: Beyond simple penalized regression
    ... for the inappropriate reasons (wrong sign of regression coefficient ... OP> accuracy of the prior. ... to the fact that priors are NOT to be taken likely by a true Bayesian. ... as applied in _traditional_ statistics today. ...
    (sci.stat.math)
  • Reef Fish Statistics for Dummies: Applied Simple Regression
    ... Statistics to arrive at the "Simple Regression" topic, ... Yin the model comes from a Normal distribution with mean ... That was where "Lesson #1" was ...
    (sci.stat.math)