Sunday, September 28, 2014

Upsides Down and Backwards

Nick Stokes has his teeth into the hockey stick for a while.  Back in 2011 he explored Deep Climate's exposure of the Wegman hanky panky, Nick found that if you didn't do the cherry pick the results were much less hockey stick like for decentered PCs

Then this March Nick explaned how McIntyre and McKitrick effectively truncated the Gaspe cedar series by fifty years, leaving, well not very much or really not very much global data for 1400-1450.   MBH had padded out that series from its end in 1404 by persistence, but a Steve McIntyre relied on a narrow reading (and Steve McIntyre is famed for such) of MBH 98 to justify that step, except they were very legalistic in not clearly explaining what they had done, until Nick Stokes worked his way through the thicket.

Then this September, a festival at Moyhu.  Three posts on the manipulations necessary to be Steve McIntyre.  Evidently Kevin O'Neill had gotten to the creative nature of Steve who was compelled to come to the defense of the Wegmans.  Nick looked on with bemusement at how the vegetables were being manuvered

Brandon Shollenberger responded by trying to move the goal posts. The selection by HS index used by Wegman had the incidental effect of orienting the profiles. That's how DC noticed it; the profiles, even if Mann's algorithm did what Wegman claimed, should have given up and down shapes. Brandon demanded that I should, having removed the artificial selection, somehow tamper with the results to regenerate the uniformity of sign, even though many had no HS shape to base such a reorientation on. 
And so we see a pea-moving; it's now supposed to be all about how Wegman shifted the signs. It isn't; its all about how HS's were artificially selected. More recent stuff here. So now Steve McIntyre at CA is taking the same line. Bloggers are complaining about sign selection:"While I’ve started with O’Neill’s allegation of deception and “real fraud” related to sign selection,...". No, sign selection is the telltale giveaway. The issue is hockey-stick selection. 100 out of 10000, by HS index.
Nick earlier had provided a simple explanation why only looking at PC1, as McIntyre does, leaves a distorted picture (Eli has made this point in the past) and today, well today more on flipped curves from both McIntyre and Moyhu

Now Eli is a simple bunny, and likes to start with definitions.  McIntyre defines the hockey stick index as
. . "as the difference between the 1902-1980 mean (the “short centering” period of Mannian principal components) and the overall mean (1400-1980), divided by the standard deviation – a measure that we termed its “Hockey Stick Index (HSI)”.
If they had defined it as the absolute value of the difference BS and SM (well, Eli notices these things) might have a leg to stand on, but sorting on the defined HSI SM eliminated all negative going curves from their collection of 100.  Accident or incompetence?  Once again Eli notes that Steve McIntyre insists that he is incredibly precise.  Eli reports, you decide.


Jeffrey Davis said...

IIRC, didn't McIntyre include the graph of his pre-selected data as a kind of lagniappe, without comment, as SJ Perelman joked about Seth Pecksniff, as though butter wouldn't melt in his mouth? It's been so long ago.

David Appell said...

This all seems to getting almost mathematically theological.

One thing being missed is that creation in science is almost never precisely right in the first go. But that hardly makes it useless -- look at all the work that followed MBH since 1998: it led to a flowering of the field of paleoclimatology and analyses of the data with other methods that give essentially the same results.

Whether MBH98 & 99 were exactly correct in every detail isn't the point at all -- the point is that it showed the way for others to follow (and improve upon), which was its greatest accomplishment.

EliRabett said...

Exactly right. Gerry North nailed it in the NAS report and the Washington Post. Still, a bunny can have a great deal of fun agitating a bag of wind.

caerbannog said...

This whole fixation on PC1 is just silly; to capture the full climate signal in paleoclimate data, *multiple* PC's should typically be retained.

Wahl/Ammann demonstrated just that with an R-code package they released with their 2007(?) paper; their code, when run with Mann's NOAMER tree-ring data, shows that the "hockey-stick" PC1 produced from "short-centered" data can be reproduced from a linear combination of multiple PC's computed from "full-centered" data.

If you compute the number of PC's to retain with "full-centered" with the same algorithm that Mann used to compute the #PC's to retain from his "short-centered" data, your "full-centered" and "short-centered" results will be nearly identical.

IOW, if you apply Mann's PC selection criteria *consistently* for short-centered and full-centered data, you get the same answer (i.e. a "hockey-stick").

However, if you compute the number of PC's to retain for "short-centered" data (with Mann's selection algorithm) and then fail to recompute the number of PC's to retain when you switch to "full centering", you won't get a hockey-stick result.

That's not because Mann's "short-centering" mines data for phantom hockey-sticks; it's because you screwed up.

Anyway, this "discussion" has been going on for what, nearly a decade now? And over something as basic as computing the number of PC's you should retain before you proceed with your analysis?

I'm definitely not an expert at any of this; I just remember the basics of this stuff (from having applied SVD/PC techniques to speed up "production" processing of acoustic array data years ago).

BTW, full Wahl/Ammann code that demonstrates the above can be found here:

Anonymous said...

why doesn't Josh go over to climate audit and comment there?

Anonymous said...

In my opinion, the biggest mistake Mann et al ever made had nothing to do with PCA centering.

It was using PCA when they didn't need to.

J Bowers said...

Why doesn't McIntyre come over and comment here? It's his work being thrown in the dustbin of wrong, and he isn't a king (well, that's maybe not true in your eyes).

Anonymous said...

"This whole fixation on PC1 is just silly; to capture the full climate signal in paleoclimate data, *multiple* PC's should typically be retained."

And if you retain all the PC's, it's just like doing no PCA at all.

Actually, the whole fixation on PCA is silly. PCA is a data reduction technique. Nothing more.It's not even necessary (for this case or any other) and can actually be worse than doing nothing if one does not retain enough PC's for the case at hand.

Those who make it out to be something more don't really understand what it is (and what it is not.)

willard said...

Readers may also like:


Why the Auditor does not comment at Nick's?

Kevin O'Neill said...

Caption to the Wegman Report's Figure 4.4:
Figure 4.4: One of the most compelling illustrations that McIntyre and McKitrick have produced is created by feeding red noise [AR(1) with parameter = 0.2] into the MBH algorithm. The AR(1) process is a stationary process meaning that it should not exhibit any long-term trend. The MBH98 algorithm found ‘hockey stick’ trend in each of the independent replications.

The orientation issue in Wegman's Figure 4.4 is interesting for what the Auditor and others never want to admit: The orientation should have provided Wegman with the necessary clue to investigate the code. Investigate the code and discover the results in Fig. 4.4 are from the archived 100:1 cherry-pick file.

The Auditor and others claim that flipping the orientation so that they're all upward-sloping is for consistency. That point is moot regarding the Wegman Report. Wegman didn't flip anything.

We know Wegman didn't flip any images to build a figure with consistent orientation because the images are all from the archived file. In that file are the 100 results with the largest HSIndex - all with positive sign. He took twelve images the code grabbed from the archived file and it apparently never struck him as odd that all 12 were upward-sloping. The odds that a random sample would all have the same sign are astronomical.

The caption to Figure 4.4 is a multitude of errors. The noise model is incorrectly stated, the images all come from the 100:1 cherry-picked archive file, and the claim of independent replication turns out to be completely false.

Other than that I don't see any problems with it :)

Jeffrey Davis said...

I'm old and my memory isn't great, so grain of salt and all that, but I also remember an issue with the y-axis of McIntyre's graph. Wasn't it hugely out-of-scale with the graphs it was supposed to be debunking? that the "hockey-stick" it showed (from cherry-picked data) would just look like noise at the same scale as Mann's graphs?

EliRabett said...

Eli had tried to put a couple of comments over at the Auditor's lair, but alas they have not appeared although they were nice and calm.

To reiterate, by defining the Hockey Stick Index as McIntyre did and then selecting the 100 with the highest scores for the random draw, those pointing downwards at the end (which would have negative HSI) would not even be considered. Thus this whole nonsense about flipping the birds is just that.

Perhaps someotherbunnies could go post there. Eli tires of agitating a bag of wind.

Anonymous said...

'the claim of independent replication turns out to be completely false."

I think what he meant was 'duplication" because that's what Wegman did (and excels at), duplicated McIntyre's file of the top 1%
(that, and significant parts of Bradley's book)

Arnold is the Terminator and Wegman is the Duplicator. Maybe Xerox should hire him. Wonder how fast he is.

Marco said...

Actually, I don't think Wegman did much. He rightfully takes most of the blame, but I am quite certain Yasmin Said was put to the task, was overwhelmed, and took a short-cut.

So, Xerox should be hiring Said...

tonylearns said...

Am I the only one that loves Brandon? He somehow manages to get embroiled in all the most arcanely detailed rabbit holes ( sorry Eli) and manages to turn them into such complicated messes that even the most obvious reality gets totally confused.

tonylearns said...

speaking of flipping have you seen this?

( I hope this doesn't get me banned. I have to be so careful with you evil war mists)

Anonymous said...

Wegman blamed others, but that don't make it so.

Read Deep Climate and you will see that the Wegman report was not the first time. There is history.

John Mashey said...

Wegman (and Said): it was worse than we knew in 2010.
See FOIA Facts series from 2013.

Among other things they used Army and NIAAA grant $, claimed credit for Wegman Report on those.
So much for pro bono.

Wegman pretty much stopped producing relevant papers for his Army grant in 2006, while helping Said with alcohol papers and lots of attacks on climate science.

KR said...

Jeffrey Davis - Yep, in MM05 the sample PC1 extracted using MBH from synthetic data is almost an order of magnitude smaller than that from the proxy data (0.08 vs. 0.6).

That should have triggered a reality check right there.

Anonymous said...

McIntyre is simply trying to throw as much FUD up as he can at this point so that what Deep Climate pointed out long ago gets lost in the blizzard.

It's really rather pathetic.

Fergus Brown said...

Do these people have an explanation for Zhang et al? (see my blog). Tree rings. Temperature reconstruction. China. Big sample. Lo and Behold! a Hockey Stick!
I presume they would call this a coincidence, or a lie. Seems reasonably clear-cut from where I'm sitting.