The term Climateball has been gaining in popularity recently, with a number of people being labeled Climateballers. Like many newish terms, it’s not entirely clear how the word should be used. To help people out, I’m going to discuss an example of how one can be a Climateballer. You should begin by saying something like user KR said:
their ‘red-noise’ model is absurd, given that they didn’t detrend the proxies first (leaving hockey-stickiness in the ‘noise’), and having utterly absurd persistence in their noise, leading to non-stationary series. Fail.
Which was said more simply by user Kevin O’Neill:
M&M contaminated their “persistent red noise” by not detrending the proxy data.
Don’t let the terminology frighten you. The complaint is quite simple. Steve McIntyre and Ross McKitrick generated random series to use as input for a methodology to see what the results would be. Fully random series are called “white.” “Red” series are series where one data point affects the value of the next. That is, in red noise, how high your second point is depends partially on how high your first point is.
The complaint is regarding how “red” the test series should be. When generated red series, you have to decide how strong an effect each point should have on other, nearby points. MM decided to emulate the redness of a set of pre-existing series. The idea was to give their test series the same amount of noise as the series they were comparing to.
The complaint expressed by KR and Kevin O’Neill above is they feel you can’t measure the “redness” of a series which has a genuine signal in it without removing the signal first. They say attempting to results in your test series being “contaminated” by the signal.
Now, being a Climateballer takes far more than making a simple complaint like that. To understand, we’ll need to consider a second complaint posted by KR:
McIntyre and McKitrick did not apply standard significance tests, kept only two PCs, and threw the hockey-stick in the significant PC4 away. This is a basic error in PCA on their part.
This is about a somewhat different topic. The test series created by MM were created to show a methodology used by Michael Mann was biased toward cherry-picking hockey sticks. The idea was to show the methodology, a faulty implementation of principal component analysis (PCA), tended select hockey sticks as its PC1. PCs, or principal components, are the output of PCA. They are ranked according to how important they are, with PC1 being the most important.
MM argue when you use a correct version of PCA, rather than the faulty one used by Michael Mann, you don’t tend to get hockey sticks in PC1 (with the real or test data). KR accepts that but argues when using the correct version of PCA, MM ought to include PC4, which does have a hockey stick shape, even though it is only the fourth most important PC. He claims this is based upon “standard significance tests.”
Comparison of eigenvalue spectrum resulting from a Principal Components Analysis (PCA) of the 70 North American ITRDB data used by Mann et al (1998) back to AD 1400 based on Mann et al (1998) centering/normalization convention (blue circles) and MM centering/normalization convention (red crosses). Shown also is the null distribution based on Monte Carlo simulations with 70 independent red noise series of the same length and same lag-one autocorrelation structure as the actual ITRDB data using the respective centering and normalization conventions (blue curve for MBH98 convention, red curve for MM convention). In the former case, 2 (or perhaps 3) eigenvalues are distinct from the noise eigenvalue continuum. In the latter case, 5 (or perhaps 6) eigenvalues are distinct from the noise eigenvalue continuum.
KR apparently missed the point of this quote being provideddespite the sarcastic commentary which framed the quote:
Indeed. McIntyre and McKitrick should have done what Michael Mann did and used an objective rule to calculate how many PCs to keep. Then they would have had to post:
Where they’d be criticized for estimating red noise parameters from tree ring networks without detrending first.
To explain, the excellent reference KR likes so much shows Michael Mann performing Monte Carlo simulations where he created many red noise series with the same autocorrelation structure as the actual data. There’s no mention of him having detrended the actual data first. In other words, he did the same thing McIntyre and McKitrick did (though the exact details of his red noise models are somewhat different).
And that’s how KR shows himself to be a Climateballer. He sees a post showing Michael Mann did something, and he calls it excellent. He sees McIntyre and McKitrick doing the same thing, and he calls it “absurd.”
It gets better though. The paper KR criticizes was published in 2005. The post I quote Michael Mann from was published in 2004. That means Mann not only did the same thing McIntyre and McKitrick did, he did it first. In fact, contemporary commentary at the time shows McIntyre and McKitrick did what they did because Mann did it first. Referencing material in the excellent source KR is fond of, McIntyre said:
We had seen this diagram and calculation in August 2004 and had fully considered it in our GRL submission – in fact, it contributed to the approach taken in our GRL submission, which differs substantially from our previous Nature submission.
So Michael Mann creates test series by using the autocorrelation structure of the actual data without detrending. McIntyre and McKitrick respond by doing the same thing. KR thinks that means McIntyre and McKitrick are horrible people who don’t know what they’re doing. He doesn’t have a problem with Mann though.
And that is how he proves himself to be a Climateballer. It’s not hard. You can do the same. Many others have.
It’s worth pointing the link to that contemporary commentary also shows Michael Mann did not use the rule he claimed to have used. In fact, more recent work strongly suggests there is no “standard significance tests” which could possibly justify the decisions Mann made about how many PCs to use.
Even if that weren’t true, the rule Mann claims to have used, Preisendorfer’s Rule N, is only one of nearly a dozen rules Preisendorfer created. Other people have created more rules. There is no way to say one rule is “right.”
(See here for far more detail about principal component analysis and the selection rules for it than you would likely ever want.)