Comments on: VZ and Huybers Comment and Reply

By: forex exchange

forex exchange — Tue, 22 Jul 2008 10:02:45 +0000

Many thanks for your sincere efforts.

By: Mark T.

Mark T. — Wed, 28 Jun 2006 23:49:33 +0000

Particularly as it pertains to fang over here.

Mark

By: Mark T.

Mark T. — Wed, 28 Jun 2006 23:41:44 +0000

Mark was using the correlation matrix in the sense of engineering. Mark, in statistics, by the correlation matrix it is meant the matrix of correlation coefficients.

I was speaking of the same thing… At least, the correlation matrix I mention is E[XX.'] where .’ is the matrix transpose (I use that rather than ‘ as the latter is Hermitian in Matlab). The result is R = [r00 r01 r02...;r10 r11 r12...;r20 r21 r22...]. Unless you are referring to the ultimate correlation matrix of R = [1 r01;r10 1] relating the correlation, I assume, of the “signal” with temperature?

Yes, I am the same Mark starting on the PhD, btw. Picking a topic has muddled my brain so excuse me if I say something out of line enough to cause confusion. That is not my goal.

Mark

By: TCO

TCO — Wed, 28 Jun 2006 23:27:59 +0000

On 31, thanks for clearing up. I think impressicion is a problem. When people use terms like “true”.

By: TCO

TCO — Wed, 28 Jun 2006 23:23:56 +0000

I was not aware that his std division was for the calibration period as well.

But still, the thing that was bizarre was the off-centric nature (of either, mean or std deviation). Steve conflates issues and overstates the impact of the bizarre by not dividing by std deviation in his test case.

By: Jean S

Jean S — Wed, 28 Jun 2006 23:17:12 +0000

TCO, after this I’ll give up. I still don’t understand what’s your problem. Mann did not remove the mean. He removed the mean of the caliberation period. Mann did not divide my the std. He divided by the std of the caliberation period. So are you asking exactly which one of those is causing the hockey stick? Or are you asking which would have been a “correct” way to proceed, correlation/covariance PCA? What part exactly I should re-read from Huybers?

BTW, (re #31): the result applies to the covariance PCA.

By: TCO

TCO — Wed, 28 Jun 2006 22:37:04 +0000

1. Jean, I don’t want to overstate the impact of the (unprecedented) centering convention by mixing in a (precedented and very debatable) “dividing by standard deviation convention”. It’s just common sense when you’re testing the effect of variable A, not to simultaneously change variable B (along with A) in your test comparison.

2. Read the Huybers comment. He really lays this out nicely.

3. Look at Burger and Cubasch flavor work (as a framework) for how to do logical analysis of multiple issues.

By: TCO

TCO — Wed, 28 Jun 2006 21:54:46 +0000

I feel like the bulldog in WHITE FANG.

By: Jean S

Jean S — Wed, 28 Jun 2006 21:52:00 +0000

TCO, I really can’t follow what’s exactly your problem. Mann was not using either covariance or correlation based PCA, so why is this such an issue to you? Or am I missing something?

By: Jean S

Jean S — Wed, 28 Jun 2006 21:47:56 +0000

re #38/#32: Mark was using the correlation matrix in the sense of engineering. Mark, in statistics, by the correlation matrix it is meant the matrix of correlation coefficients.