This is still another post from November of 1997 on sci.stat.consult
detailing my objections to over-interpreting any change in sign of
correlations (polarizations is what Chambers calls them) over the data
set.
Recovered via dejanews.
Paul
**** Begin included message ****
Byron L. Davis said on 10/30/97 10:34 AM:
>I believe that the polarization of iv's across their respective dv is a
>consequence of the model specification. It is indeed logical that when
one
>creates a model y=x1+x2 where the combination of x1 and x2 correlate
with
>y at approx. .71 and at the same time do not correlate with each other,
and,
>they are uniformly distributed, that when you sort the data by y and
look at
>the correlation between x1 and x2 in the upper and lower quartiles they
have
>opposite signs. This is a consequence of the fact that their overall
>correlation is zero so the negative correlation in one quartile
canciles out
>the positive correlation in the other quartile. So, I reiterate, this
seems
>a logicall consequence of the model specification and variable creation
and
>therefore is hardly profound.
I had previously said that I agreed with Chambers that polarization
occurs and asked him to explain why it
mattered that they occured. I want to modify that by saying that I agree
that we calculate polarization but it is so
misrepresentative of the data that I hesitate to make any claim about
the data based on that property. Even in
private email Chambers did not explain how/why this property inferred
causality. When I didn't agree with him, he
became insulting. Sad that this is the best he can do.
I don't find the correlation of the upper quartile is one sign and the
one in the lower quartile of the other sign and I
don't think Chambers intends this either.
I ran some simulations on Excel and had discussions via email with
Chambers. What Chambers says to do is create
the random variables x1 and x2 and calculate Y from adding them. Sort on
Y. Take the upper and lower quartiles
of Y. *Combine* these two quartiles of data and you get a positive
correlation between x1 and x2. The reason is,
of course, that you have a 'barbell' shaped distribution; two negatively
sloping groups lying on a line that is
positively sloped. The resulting correlation is positive but hardly
represents the shape of the scatterplot in any
reasonable way. The middle two quartiles is, of course, negatively
correlated. But this is only because we've
removed the upper quartile and lower quartile from the data which
effectively cuts off the upper right and lower
left triangles of data. It is like taking a rectangular piece of paper,
cutting off the upper right and lower left corner
and saying: Look! It was linear all along! I can cut up any distribution
using creative methods and make such
statments...
+=============================================================+
Nearly all men can stand adversity, but if you want to test
a man's character, give him power.
--Abraham Lincoln
Paul C. Bernhardt, M.S. in Social Psychology (non-clinical)
+=============================================================+
**** End included message ****
William Chambers wrote on 2/28/00 7:42 AM:
>Horst,
>
>Get your shit together, What do you think about the polarization effect in
>the model y=x1+x2.
>
>Bill Chambers
>
>Horst Kraemer wrote in message <[EMAIL PROTECTED]>...
>>On Sun, 27 Feb 2000 19:17:13 -0600, "William Chambers"
>><[EMAIL PROTECTED]> wrote:
>>
>>
>>> I am sure you feel almost like a real doctor (MD).
>>
>>
>>> Listen to me little man.
>>
>>
>>> Now grow up and have a conversation with me. How in the world do people
>>> like you get jobs in universities.
>>
>>
>>
>>Stop depositing your faeces in newsgroups. Go to a public toilet.
>>
>>
>>Tank you
>>Horst
>>
>
>
>
>
>===========================================================================
>This list is open to everyone. Occasionally, less thoughtful
>people send inappropriate messages. Please DO NOT COMPLAIN TO
>THE POSTMASTER about these messages because the postmaster has no
>way of controlling them, and excessive complaints will result in
>termination of the list.
>
>For information about this list, including information about the
>problem of inappropriate messages and information about how to
>unsubscribe, please see the web page at
>http://jse.stat.ncsu.edu/
>===========================================================================
>
===========================================================================
This list is open to everyone. Occasionally, less thoughtful
people send inappropriate messages. Please DO NOT COMPLAIN TO
THE POSTMASTER about these messages because the postmaster has no
way of controlling them, and excessive complaints will result in
termination of the list.
For information about this list, including information about the
problem of inappropriate messages and information about how to
unsubscribe, please see the web page at
http://jse.stat.ncsu.edu/
===========================================================================
