Thread: Why The "Hockey Stick" Proves Nothing

Originally Posted by Windigo

PD you keep referring to this "real signal" what real signal is there in red noise?

The real signal is in the variable you're correlating to. The variable you're screening the data on. Once you select the data, it's no longer random!

Originally Posted by Windigo

Spurious correlation in random data is not a signal.

The correlation isn't spurious, it's real. Because the data isn't random any more.

Look at it this way. Suppose I roll a single die 600 times, and get 102 sixes. Then I select those 102 sixes and create a dataset with only sixes in it.

Is that dataset containing only sixes "random" in any sense of the word? Of course not. It's been selected and screened. Just like you selected and screened your random data, to arrive at non-random data.

Originally Posted by Poor Debater

The real signal is in the variable you're correlating to. The variable you're screening the data on. Once you select the data, it's no longer random!

The set of series you are using is no longer random. They all have one thing in common correlation to the dependent variable in the calibration period. But each series is still random and the correlation is still just as spurious.

The correlation isn't spurious, it's real. Because the data isn't random any more.

Of course the correlation is spurious the series is random red noise. It doesn't relate to the dependent variable at all. It is the ultimate spurious correlation.

Anyone who argues that correlation has to be real has really started to lose it. And since I've seen this argument popping up more and more in warmmonger circles they have clearly lost it.

Do you know that the statistics department at Dr. Mann's own university gives an annual spurious correlation award to whoever can identify the most spurious correlation? I dont really know why they let anyone in the world try and claim this award when they need look no further than their own earth science center.

Originally Posted by Windigo

The set of series you are using is no longer random. They all have one thing in common correlation to the dependent variable in the calibration period. But each series is still random and the correlation is still just as spurious.

And by using non-random series, you have introduced signal into the noise. That's the point. The one who has fallen for the screening fallacy is YOU.

Let's go back to rolling a die again, but this time let's create 1000 series of 10 rolls each. From those 1000 series, we find five series that start 6, 6, 6 and use only them, throwing out the rest. There is now a 100% real signal in the those selected series, even if the generation was random.

Originally Posted by Windigo

Of course the correlation is spurious the series is random red noise. It doesn't relate to the dependent variable at all. It is the ultimate spurious correlation.

Anyone who argues that correlation has to be real has really started to lose it. And since I've seen this argument popping up more and more in warmmonger circles they have clearly lost it.

And how does Mr. Statistics tell the difference between a real correlation with an r²=.9, and a spurious correlation with an r²=.9? There is no statistical difference. Any time I see deniers try to claim that A ≠ A, I know they've lost it.

Regarding this business about hockey-sticks being generated from "red noise", it should be noted that McIntyre screwed up his red-noise generation procedure.

I'll try to explain, in plain-English terms as much as possible, just how he screwed it up.

McIntyre used the R functions acf() and hosking.sim() to generate his red noise. He used those functions to (1)extract the tree-ring data autocorrelation structure, and (2)generate random noise with the same autocorrelation structure as the tree-ring data.

That's a nice idea in principle, but you have to be careful when you implement it. You must ensure that the data that you feed the acf() function contains only noise. If the data contains signal, then the "random noise" generated by acf()/hosking.sim() procedure will be contaminated with signal statistics. As a result, your "random noise" will be useless for evaluating "noise only" performance.

The data that McIntyre fed to the acf() function was tree-ring data that contained not only noise, but also the long-term "hockey-stick" climate signal. You can verify that for yourself by taking the tree-ring chronologies, normalizing them, and simply averaging them together. You will still see an underlying "hockey-stick" trend even when you don't regress the data against the instrumental temperature record. These "simple average" results are noisier, to be sure, but the underlying "hockey stick" signal is still visible.

So if you are going to use the tree-ring data as a model for your noise, you need to pre-process it to remove the "hockey stick" signal first! McIntyre failed to do that (as can be verified by inspecting his R code at http://www.people.fas.harvard.edu/~p...ckey/Rscript.R).

Because McIntyre failed to do so, his "random noise" was contaminated by long-term signal autocorrelation characteristics, and as a result, his "random noise" had an autocorrelation length that was a significant fraction of the total reconstruction duration. That can introduce "small sample size" issues that will increase the chance that a spurious trend will pop up.

Furthermore, the eigenvalues associated with the "noise hockey-stick" principal components that McIntyre generated were *much* smaller than the eigenvalue associated with Mann "hockey stick" leading principal component (even with the assistance of the "hockey-stick" signal contamination).

Even though the PC's will be reweighted by the regression step (and those weights will likely differ from the eigenvalues), the eigenvalue magnitudes can often tell you **whether it even makes sense to proceed with the regression step**. If your leading PC has a small eigenvalue magnitude (as McIntyre's "noise hockey sticks" did), then that's a good indicator that there may not be much of a coherent signal to extract, and that it may not even make sense to proceed with the regression step.

So in summary:

1) McIntyre's "red noise" was contaminated with hockey-stick signal characteristics.

2) Even with the "hockey stick" contamination, the McIntyre's "red noise" hockey sticks were much smaller in magnitude than Mann's tree-ring hockey stick.

As a result, McIntyre's claim that Mann's method will take random noise and generate a hockey-stick that could be mistaken for a real climate signal is completely invalid.

Originally Posted by caerbannog

As a result, McIntyre's claim that Mann's method will take random noise and generate a hockey-stick that could be mistaken for a real climate signal is completely invalid.

Ooops.

Originally Posted by Poor Debater

Mann's paper (which you clearly have not read) made no claim at all about the causes of global warming. The causal evidence is basic physics. And basic physics is something that climate deniers such as yourself (a) don't understand; and (b) don't want to talk about.

So just keep burying your head in the sand, FUD. The icecaps will continue to melt, and we will still be at fault.

I have been very busy at work, extracting energy from the earth so you can keep pounding away on your computer...even though you are convinced you are destroying the world.

However I have a minute now, and I would be delighted if you would take a few minutes to elaborate on how "basic physics" = "causal evidence".

Remember my name is listed on the former Chicago Climate Exchange so I think I will be able to follow you..... (getting popcorn....)

Why not read the paper you don't understand, and someone might explain it to you if you still don't get it? You are the one making the false assertions. Your turn.

Originally Posted by Colonel K

Why not read the paper you don't understand, and someone might explain it to you if you still don't get it? You are the one making the false assertions. Your turn.

I could say the same to your and caerbannog as caerbannog is once again arguing that McIntyre and McKintrick ever said you can created a hockey stick using the MBH98 method from data that doesn't contain a hockey stick. M&M have always insisted that there has to be a hockey stick somewhere in the data set for the MBH98 short centered PC method to produce a hockey stick. If there s no hockey stick in the data you will not get a hockey stick or if the method doesn't short center and you have a hockey in the series still you will not get a hockey stick.

Its complicated I know and it probably goes well over your and many warmmongers head but its simply comes down to the fundamental argument that one mistake justifies another. Simply put when warmmongers make mistakes they always tend towards creating hockey sticks.

The fact of the matter is that correct statistical analysis like what was done in MM03 will not produce a hockey stick.

However, most hockey sticks do not use the short centered PC analysis of MBH98 and instead use purely correlative models which will inherently produce hockey sticks from pure red noise instead of the pseudo red noise that MM05 used. One mistake doesn't justify another. The fact that there is more than one incorrect way to make a hockey stick doesn't mean that they are valid.

Originally Posted by Colonel K

Why not read the paper you don't understand, and someone might explain it to you if you still don't get it? You are the one making the false assertions. Your turn.

Forth grade reasoning....

Do you pray fervently to Allah 3 times daily?? Why not?? Since you haven't read the Koran, how can you claim you should not........


Doesn't mater anyway, you and yours simply ignore the fact that there have been thousands of "hockey sticks" in earth's past, and yet somehow your brainwashed minds decide THIS TIME....it is all due to SUV's and power plants.....

pretty weak conclusion don't you think??????

You are not fooling anyone.....well at least enough voters see through it now.....find another way to try bring on socialism to America

Originally Posted by Windigo

I could say the same to your and caerbannog as caerbannog is once again arguing that McIntyre and McKintrick ever said you can created a hockey stick using the MBH98 method from data that doesn't contain a hockey stick.

If that's the case there is no fallacy and MBH98 didn't make a mistake. Because the blade of the hockey stick is in the data: that's the upturn in temps that is known as part of the instrumental record. In other words, the hockey stick isn't broken.