John Williams writes:
Here is an example of climate scientists pointing out that something they discovered doesn’t exist; seems like a good example for your blog.
He’s linking to a post by Michael Mann, who writes:
Two decades ago, in an interview with science journalist Richard Kerr for the journal Science, I coined the term the “Atlantic Multidecadal Oscillation” (AMO) to describe an internal oscillation in the climate system resulting from interactions between North Atlantic ocean currents and wind patterns. These interactions were thought to lead to alternating decades-long intervals of warming and cooling centered in the extra-tropical North Atlantic that play out on 40-60 year timescales (hence the name). Think of the purported AMO as a much slower relative of the El Niño/Southern Oscillation (ENSO), with a longer timescale of oscillation (multidecadal rather than interannual) and centered in a different region (the North Atlantic rather than the tropical Pacific).
Today, in a research article published in the same journal Science, my colleagues and I have provided what we consider to be the most definitive evidence yet that the AMO doesn’t actually exist.
Mann then provides the background:
Back in the 1980s and 1990s, a number of articles pointed to a pattern of North Atlantic warming during the 1930s-1950s, subsequent cooling in the 1960s and 1970s, and warming thereafter, which seemed to resemble a natural oscillation in the climate system. I co-authored an article (Mann et al, 1995) in the Nature demonstrating the apparent persistence of these multidecadal oscillations several centuries back in time based on the analysis of paleoclimate proxy data . . .
In 2000, in an article that led to Kerr’s commentary in Science, my collaborator from the climate modeling group at the Princeton Geophysical Fluid Dynamics Laboratory (GFDL) Tom Delworth and I argued that this observational evidence for an AMO-like climate oscillation was supported by an analysis of an extended (thousand+ year long) control simulation of GFDL’s state-of-the-art (at the time) coupled ocean-atmosphere model. Since it was a control simulation with no external “forcing” (no greenhouse gas changes, no variations in solar output, no volcanic eruptions, etc.), any oscillation that was produced has to be internally generated. . . .
About five years later, analysis of an extended simulation of yet another climate model—the coupled ocean-atmosphere model run by the Hadley Centre within the UK Meteorological Office, yielded evidence for a similar oscillation, albeit with a longer (roughly 100 year) period, and a more global signature . . .
But it turned out that these simulated patterns do not show up with newer models. There’s a real pattern out there, but the appealing internal explanation does not seem to work.
Mann concludes:
There are several lessons in this tale. One is that scientists must always be open to revising past thinking. . . . Two decades ago there seemed to be both observational evidence and modeling evidence (if rather limited) for the existence of a multidecadal AMO in the climate system. My own work supported that interpretation, and indeed it was I who gave this beast a name. The scientific community ran with the concept, and numerous scientists—even some at our leading research laboratories like the aforementioned GFDL—continued to misapply it in a way that downplays some critical climate change impacts like the warming of the North Atlantic and the increase in Atlantic hurricane activity associated with it.
Now we have come full circle. My collaborators and I, over the past decade, have continued to investigate the origins of the putative AMO signal and have been led inescapably to the conclusion that the AMO (unlike, say, R.O.U.S.) doesn’t actually exist. It’s an artifact, during the historical era, of competing anthropogenic (greenhouse warming and sulphate aerosol cooling) drivers and, during the earlier period, an artifact of the fact that volcanic forcing happens to have displayed a roughly multidecadal pacing in past centuries. . . .
Unfortunately for all of us, my colleagues and I weren’t wrong about the unprecedented warming revealed by the now iconic “Hockey Stick” curve, despite the unrelenting attacks on it by climate change deniers over the past two decades. But I was wrong about the existence an internal AMO oscillation when I coined the term twenty years ago.
Interesting. I know essentially nothing about the substance here, but I’m sympathetic to the general story of learning from mistakes.
Mann wrote:
“It’s an artifact, during the historical era, of competing anthropogenic (greenhouse warming and sulphate aerosol cooling) drivers and, during the earlier period, an artifact of the fact that volcanic forcing happens to have displayed a roughly multidecadal pacing in past centuries. . . .”
If you have never worked from a control system perspective on a complex, real-world system, it is unlikely that you would be able to recognize just how breathtakingly naive this statement is. He doesn’t have the slightest idea what it would take to be able to adequately validate a statement like this about past causation of a system as complex as the climate, and yet he states it as fact. He has just made up stuff like this his entire professional career.
First of all, in complex systems, you never, ever decide that because something doesn’t show up in your model, the real-world data must be wrong.
AGW theory is facing some huge obstacles right now, but you cannot be blamed if you don’t know about them. The biggest right now is the longwave/shortwave energy imbalances measured by satellites, which directly refute the most basic claim by Arrhenius about CO2 warming. Arms have been waved on that one, for sure. This post mentions another one, the surface of the North Atlantic is warming but with a better understanding of the evidence we now see a complex mechanism that is inconsistent with pretty much every claim made in the past as to why it is warming.
Speaking of Michael Mann, has this blog ever taken a look at his famous hockey stick, one of the most controversial (if one is willing to look at how the sausage was actually made) and consequential uses of statistics in our lifetimes? It is frankly a little weird to read about Wansink et al so many times with Mann’s work never getting a mention despite being far more prominent. Now that Mann got a chance to toot his horn on the hockey stick here, maybe it’s time?
I think you’ve got this one backwards. He’s not saying the data is wrong. The data do display the temperature oscillation—he’s saying climate reconstructions indicate there’s no reason said oscillations are continuing now.
If you click through to the link
https://www.science.org/doi/10.1126/science.abc5810
It explains that the reason is that with new reconstructions, the timing of the historical temperature oscillations match up very well with historic periods of volcanic activity. The source for the timing of historic activity appears to be measurements of sulphate concentrations in ice cores
https://essd.copernicus.org/articles/5/187/2013/
https://climate.envsci.rutgers.edu/IVI2/
In addition, the signal appears to be strongest near where all the erupting volcanoes were.
So broadly speaking:
1. There was an apparent periodic oscillation in the temperature timeseries
2. Physical evidence was used to reconstruct the timing of volcanic activity
3. The reconstructed timing and spatial distribution of volcanic activity seems to match up with the aforementioned temperature data
Of course, post hoc, ergo, propter hoc, but there’s a very well established mechanism by which volcanoes can have a cooling effect. It’s not bulletproof, but I’m not really sure what exactly your problem is with the reasoning above. It’s certainly doesn’t seem like making things up.
Don’t know what the hockey stick is
If volcanic activity is periodic, how is that an artifact?
Reminds me of when people filter out el ninos as if they are some kind of rare event rather than regularly occurring part of climate dynamics. Or model an Earth with no atmosphere but same albedo, when the albedo depends on the atmosphere. Makes no sense.
The hockey stick, i.e., Mann, Bradley, Hughes(1999) was an early multiproxy reconstruction of NH temperatures over last 1000 years.
https://en.wikipedia.org/wiki/Hockey_stick_graph_(global_temperature)
It was a fairly good approximation (with wide error bars) to the most recent IPCC(2021) PAGES2K reconstructions,
Fig SPM.1(a):
https://www.ipcc.ch/report/ar6/wg1/figures/summary-for-policymakers/figure-spm-1/
Most of the “blade” of the hockey stick is modern instrumental (land/ocean) temperatures.
Instead of taking potshots at what I wrote, everyone – statisticians in particular – should read the two original hockey stick papers:
https://www.meteo.psu.edu/holocene/public_html/shared/articles/mbh98.pdf
https://agupubs.onlinelibrary.wiley.com/doi/epdf/10.1029/1999GL900070
…and then read this:
https://climateaudit.files.wordpress.com/2005/09/mcintyre.mckitrick.2003.pdf
From the abstract:
“The particular “hockey stick” shape derived in the MBH98 proxy construction – a temperature index that decreases slightly between the early 15th century and early 20th century and then increases dramatically up to 1980 — is primarily an artefact of poor data handling, obsolete data and incorrect calculation of principal components.”
MBH98 is the penultimate zombie paper. Because the M&M03 paper was written despite Mann refusing to share methods and data, much had to be guessed. There was an even more damning paper two years later once some of the withheld information was available:
https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2004GL021750
Pretty much every subtopic discussed at this blog related to academic integrity comes up at some point in this sordid tale.
(Response to Mr. Skaggs’ reply)
As I vaguely recall from extensive blog coverage at the time, The Mann principal-component analysis was formally incorrect because it didn’t use the average of the whole (mostly reconstructed) data but instead used an average of known data prior to about 1950 as a baseline. (An ad-hoc procedure that appealed to me as a typical engineering approximation, but which annoyed some mathematicians.) That produced a hockey-stick principal component which was of higher order than the one which resulted from using all the data (#3 vs. #7, if I recall correctly). However, the plotted data itself was not affected, and shows the hockey stick (as Mr. Mashey notes, above).
Speaking as an engineer, much flakier, approximate methods are frequently used in engineering analyses. The main criteria is whether the part then works in service. I feel that ultimately it is all trial and error, all the way down–with the need to learn from the errors, of course.
Jim,
Yes, that’s my understanding too. I think that the way they normalised the data (using only the modern period) might have slightly amplified the hockey stick, but it didn’t create hockey stick shapes from data that didn’t have them in the first place. Also, the claim that MM03 was some kind of debunk of MBH98 is simply wrong. You often see claims that MM03 demonstrated that the method generated hockey sticks from random data, but their random data had persistence. Hence, the hockey stick-like shapes that emerged from their analysis was actually from data sets that had (by chance) a hockey-stick-like shape anyway.
As “AndTheThere’sPhysics” indicates above the sentiment in that abstract quote is not correct. I haven’t engaged with this stuff for years but did a quick recapture yesterday. The McKitrick/McIntyre (MM) critique you refer to was shown to be largely unfounded (even if they did uncover some errors in the original analysis) – and it seems to me they massively trumped this with some pretty dismal analyses themselves (especially due to a lack of validation in their obviously wrong “reconstructions” and removing datasets that they seemed not to like):
I expct this blog doesn’t appreciate loads of links to papers so I’ll just put the paper titles that address the MM critique, and the papers can be found by Googling if anyone cares 20 years down the line:
– Reply to: “Global-scale temperature patterns and climate forcings over the past six centuries: A comment.” By S. McIntyre and R. McKitrick (2004)
– Comment on ‘‘Hockey sticks, principal components, and spurious significance’’ by S. McIntyre and R. McKitrick (2005)
– Robustness of proxy-based climate field reconstruction methods (2007)
– Robustness of the Mann, Bradley, Hughes reconstruction of Northern Hemisphere surface temperatures: Examination of criticisms based on the nature and processing of proxy climate evidence (2007)
Fast forward away from that messy period 15-25 years ago and the results of the Mann et al reconstructions have been pretty uniformly reproduced as much larger datasets become available allowing global scale warming and much extend time periods to be assessed (as I discovered ystdy). e.g.:
Proxy-based reconstructions of hemispheric and global surface temperature variations over the past two millennia (2008)
Holocene global mean surface temperature, a multi-method reconstruction approach (2020)
Consistent multidecadal variability in global temperature reconstructions and simulations over the Common Era (2019)
Not sure how you consider the original hockey stick paper a “zombie” paper since while it helped to introduce a method of interpreting past temperatures/climate from proxy data, it’s been massively superceded by subsequent work. Perhaps it’s the McKitrick/McIntyre critique that’s the “zombie” since it only seems to get wheeled out to support the notion that climate science should be “frozen” in the 1998 state of uncertainty with Dr. Mann as a figure of dubious provenance!
…and Then There is Physics wrote:
“the hockey stick-like shapes that emerged from their analysis was actually from data sets that had (by chance) a hockey-stick-like shape anyway.”
Nope, that is just flat wrong. First of all, have you looked at the raw data itself, the actual measurement data that Mann used? Step one in good science is the initial eyeball of the data, which is immediately a problem with Mann’s hockey stick. It is impossible to put the ring width graphs side by side and see anything but random noise (which is not the same as saying trees don’t record temps, some do, but an unbiased average of those trees did not give the desired signal).
The M&M papers used autocorrelated red noise to simulate tree ring data, which has been a topic of debate just because something had to be challenged. But any data that contains trends, no hockey stick required, will create a hockey stick output. It’s frankly hard to believe that folks are still getting this stuff wrong.
My last post on this topic here, sorry Andrew.
Matt,
The biggest right now is the longwave/shortwave energy imbalances measured by satellites, which directly refute the most basic claim by Arrhenius about CO2 warming. Arms have been waved on that one, for sure.
If you’re referring to the observation of increasing outgoing LW and reducing outgoing SW (which implies increasing absorbed solar radiation) then that doesn’t actually challenge CO2-driven warming (it might challenge what Arrhenius suggested, but we’ve learnt a lot in the last 127 years). It’s long been known that much of the actual warming is due to increased absorbed solar radiation, but this is because there are both longwave and shortwave feedbacks to a change in external forcing. So, this doesn’t contradict our basic understanding of CO2-driven global warming.
You could try reading these papers.
https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2009GL037527
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4250165/
“this doesn’t contradict our basic understanding of CO2-driven global warming.
If the physical measurement evidence proves – finally after all these years of model tweaking! – that the most basic argument made by Arrhenius, that outgoing longwave radiation would be diminished by a deeper CO2 layer, is wrong, what COULD “contradict” our “basic understanding?”
Fact is, as currently implemented, climate models can be tweaked to be consistent with anything. That’s because they are almost purely ad hoc assemblages of parameterizations set to accurately hindcast while predicting future warming. Lots of complexity, almost no sophistication. Being able to simply shrug off a major problem with the underlying physics (BTW, do your colleagues know that you are willing to throw Arrhenius under the bus?) with a few parameterization tweaks would be a bug to anyone who is familiar with complex systems models, not a feature.
All the papers you and Chris cited would be of no value whatsoever if enough people capable of understanding what they are reading were to actually read the papers I listed. Full stop. It really is a fascinating story built entirely around what was at the time a novel statistical approach.
Mann’s approach to criticism is identical to Amy Cuddy and the others who think in terms of a “methodological Stasi.” Mann’s work is no better than hers, yet his attacks on his critics somehow always find a sympathetic ear.
The key finding from the M&M papers was that you can feed any data at all – the noisier the better – into his method and get a hockey stick. There never was any refutation of that.
The notion that PAGES2K was somehow a validation of Mann’s hockey stick is just more of the same. In fact, it is literally the same, it uses the same favorite proxies, systematically culls out all conflicting data in the name of quality control, and just returns the favored input as output. There is lots of good stuff out there on these topics, you just won’t find Mann’s name at the bottom of any of it.
That doesn’t seem too scientific to me.
For example:
That’s an unsupported assertion – what do you actually mean? What’s your evidence for that assertion? Of course a deeper CO2 layer doesn’t “diminish outgoing longwave radiation”. The outgoing radiation, once equilibrium is reestablished, is unchanged – but it occurs on average from a higher altitude and so the Earth system warms up right down to the surface to bring the new average height of emission back to the (blackbody) temperature required to balance incoming solar energy.
Seems to me the evidence supports the expectation that increased greenhouse gas concentrations affects outgoing longwave radiation. e.g.:
Trends in spectrally resolved outgoing longwave radiation from 10 years of satellite measurements (2021)
https://www.nature.com/articles/s41612-021-00205-7
So you need to explain what you are referring to.
That’s a pretty unpleasant slur. It’s also incorrect in my reading. Mann’s approach to critique is to address the critique and respond to this (did you bother to look at the papers I cited?). Secondly he seems to have continuously readdressed his methods, acquired new proxy data and so on so that the original analysis (MBH 1998) is extended and presumably improved (e.g. the Mann et al 2008 paper in my post above).
Not really – have a look at the papers I referred to in my post. Of course if you do a half-assed analysis you can easily produce a pigs ear!
Not really. The proxies are vastly extended compared to the original Mann paper. No one says that PAGES2K is “a validation of Mann’s hockey stick”; that’s silly. However it is a useful bit of science since it shows, along with quite a few other analyses, that the use of a variety of proxies using different methods of analyses come to broadly consistent interpretations of Earth surface temp over last several thousand years.
Can you clarify this? You seem to be equating average height of emission with temperature, ie the units of km can be seen as units of K. What is the data the average emission height is based on? The tropopause?
I’d guess it eventually comes back to the “constant atmospheric mass” and “constant relative humidity” assumptions.
Matt,
If the physical measurement evidence proves – finally after all these years of model tweaking! – that the most basic argument made by Arrhenius, that outgoing longwave radiation would be diminished by a deeper CO2 layer, is wrong, what COULD “contradict” our “basic understanding?”
No, that’s not what I’m suggesting. It is indeed the case the adding CO2 to the atmosphere reduces the outgoing longwave flux, as expected. However, there are both long and shortwave feedbacks that mean that what happens next isn’t simply the outgoing longwave recovering to what it was before as the system warms. It actually tends to a slightly higher outgoing longwave flux because of the shortwave feedbacks (mostly clouds) that also lead to increased absorbed solar radiation.
“…the most basic argument made by Arrhenius, that outgoing longwave radiation would be diminished by a deeper CO2 layer, is wrong…”
I just want to note that Arrhenius performed his calculation of the magnitude of equilibrium greenhouse warming under the assumption that the total amount of outgoing longwave radiation would be entirely unaffected by a deeper CO2 layer. Not only was “diminishment” not his most basic argument, he assumed that it was not so. He also correctly noted that his assumption would be violated if there was a change in albedo, with less snow cover, for example if caused by CO2-induced greenhouse warming, leading to greater outgoing longwave radiation, as observed.
The observations are consistent with Arrhenius’s most basic argument and his most basic caveat to that argument.
For those who want a rigorous scientist’s public education take on the nuanced physics of the greenhouse effect, you can’t do better than Sabine Hossenfelder. Matt Skaggs is not correct.
https://www.youtube.com/watch?v=oqu5DjzOBF8
Enjoy!
Well, she repeats that 255K without an atmosphere number (in the form of saying -18 C without an atmosphere). That is an obvious averaging artifact so I don’t think she has looked into this as much as she could have.
But that isn’t what I am concerned about. Rather, it is that if there is warming the volume and mass of the atmosphere should also increase proportionally. If that requires dropping the (convenient but baseless) assumption of constant relative humidity so be it.
You might search this blog for “Wegman”. The M+M rebuttal to Mann et alia was a fraud. Please learn from that.
JimV: M&M’s Principal Components claim vs MBH98/99 was a manufactured red herring, just plausible enough not to be immediately rejected.
Doug Nychka (distinguished statistician,1996-2017 Director Institute of Mathematics Applied to Geosciences) suggested project to check, got $ for Eugene Wahl, who with Caspar Ammann did the analysis to see whether or not centering over whole period versus cetenreing over 1902-1980 made a difference. Wahl&Amman thanked Nychkas and Claudi Tebaldi (another well-published staistician) for help. The result was no surprise: the choice moves the PCs around, but if one includes *enough* PCs in each case to cover the variance, the end result is not significantly different.
There were 2 papers, 1st on MBH98: Received: 11 May 2005 / Accepted: 1 March 2006 / Published online: 31 August 2007
https://cires.colorado.edu/outreach/sites/default/files/2020-03/Climate%20Resiliency%20HS4%20-%20Wahl%20ClimChange2007.pdf
See Fig.5a p.25.
2nd on MBH99: Received: 22 August 2000 (sic) / Accepted: 13 June 2007 / Published online: 24 August 2007
https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=dfb5f8891a4b38f6c6303c050a678ad03a45fbef
See Figure 2, p.84.
IPCC(2007) AR4 As usual, assessed recontructions, kept any plasuible ones including MBH99 in Fig.6.10(b), but explicitly rejected MM05(a, b) complaints:
Link to relevant Chapter
p.466: “McIntyre and McKitrick (2003) reported that they were unable to replicate the results of Mann et al. (1998). Wahl and Ammann (2007) showed that this was a consequence of differences in the way McIntyre and McKitrick (2003) had implemented the method of Mann et al. (1998) and that the original reconstruction could be closely duplicated using the original proxy data
McIntyre and McKitrick (2005a,b) raised further concerns about the details of the Mann et al. (1998) method, principally relating to the independent verification of the reconstruction against 19th-century instrumental temperature data and to the extraction of the dominant modes of variability present in a network of western North American tree ring chronologies, using Principal Components Analysis. The latter may have some theoretical foundation, but Wahl and Amman (2006) also show that the impact on the amplitude of the final reconstruction is very small (~0.05°C; for further discussion of these issues see also Huybers, 2005; McIntyre and McKitrick, 2005c,d; von Storch and Zorita, 2005)”.
(Huybers found problems with MM05(b, a), as did von Storch and Zorita, although they had earlier criticized MBH98/99.)
Again, there was nothing actually wrong with MBH’s PCA choice of centering, and to some extent, it even made more sense, because the 1902-1980 segment is different from earlier centuries in being “anchored” to modern instrumental temperatures, whereas earlier centuries are reconstructions from proxies. But anyway, at least for these sorts of datasets, it just doesn’t make significant difference, and in any case, PCA was only used as dimension-reduction for part of the data.
As it happens, one of the most eminent PCA scholars, Ian Jolliffe, and his colleague Jorge Cadima considered the math bases:
Jolliffe IT, Cadima J. 2016 Principal component analysis: a review and recent developments.
Phil. Trans. R. Soc. A 374: 20150202.
https://royalsocietypublishing.org/doi/pdf/10.1098/rsta.2015.0202
p.8 “(iii) Centrings
As was seen in §2, PCA amounts to an SVD of a column-centred data matrix. In some applications [15], centring the columns of the data matrix may be considered inappropriate. … This is often referred to as an uncentred PCA … it is not immediately intuitive that there should be similarities between both variants of PCA. Cadima & Jolliffe [15] have explored the relations between the standard (column-centred) PCA and uncentred PCA and found them to be closer than might be expected.”
Cadima & Jolliffe[15] is from 2009:
https://web.archive.org/web/20150512171220/https://www.pakjs.com/journals/25(4)/25(4)7.pdf
[12, 13, 14] cite MBH98 (although with typo) and M&M (2003, 2005).
Sadly, due to decreasing activity by IJ and increasing committments by JC, they didn’t further explore effects of different centerings.
ATTP & JimV
There were essentially two arguments:
1) That MBH99 made a BigMWP disappear, but:
1989 Jones et al, doubt, explicit concern that most of evidence for MWP was from Europe.
1990 IPCC “MWP may not have been global”
1991 Tucson workshop
1993 resulting papers submitted to Climatic Change
1994 Hughes&Diaz, summarizing: stop using sketches like IPCC(1990) Fig 7.1(c)
1995 IPCC uses Bradley&Jones(1993)
1998 MBH98, back to 1400AD
1998 Jones et al, back to 1000AD, no BigMWP, but like the later MBH99, modest MWP, downslope tO LIA
2) Idea that MBH98/99 would generate hockey stick blade from noise never made sense, given the 4 segments:
A) Reconstruction before modern instrumental temps
B) Validation of reconstruction vs early part of instrumentals
C) Later part of instrumental record used to calibrate reconstruction formulae
D) Instrumental record AFTER C)
Most of the “blade” is B+C+D, any reconstruction purporting to cover most of NH that had no blade would be seriously bogus.
Took me a while to find these:
Animation from PPT slides I happen to have, overlaying IPCC(2021) recosntruction & simulation atop IPCC(2001) hockey stick.
https://twitter.com/JohnMashey/status/1642658596696780800
and a 2minute GIF that shows the mainline evolution of temperature reconstructions from Lamb(1965) through IPCC(2021):
https://twitter.com/JohnMashey/status/1642677120953651202
Progress happened via:
1) More proxies, more types of proxies, increased geographical coverage => smoother curves, except of course for
the sharp dips from the right sorts of volcanoes. But some 2000s reconstructions (of limited geographies) had wild gyrations that didn’t make much sense from physics. Still, Jones et al(1998) got pretty good appoximations with 10 NH and 7 SH proxies, Fig 4, p.463. https://www.st-andrews.ac.uk/~rjsw/papers/Jonesetal-1998.pdf
2) Exploration and adoption of different statistical methods
These let them
3) Push reconstructions back, 1400AD -> 1000AD -> 1AD
4) Reduce error bars,as can be seen in the 1st GIF above. I.e., approximations improved over 20-30 years.
I recall Dr. Jolliffe showing up in the comments of the “Open Mind” blog (during the Mann vs. M&M controversy) to state his opinion that there just the one right way to do PCA and that was to use the average of all the data. Or at least that he didn’t have any reason to give credence to the results when a different centering was used. It sounds like he later did some work on the issue and amended his opinion somewhat.
I also recall the the M&M “random” hockey stick generation was seen as a transparent trick by the consensus of the old “Science Blogs” authors. As in, sure if you do PCA on enough random samples, sooner or later you will find some that have a hockey-stick component, because some of them will randomly have one which you can see in the data. At that point it began to sound like flat-earth stuff.
Not that my vague recollections amount to a hill of beans in this crazy world. I thank Dr. Mashey for his much better information.
JimV: Indded, some of the early discussions were confusing.
M&M in MM05b explained that half the hockey sticks were upside down, but generally, only showed the upwards ones.
Of course, their code
1) generated 10,000 curves
2) sorted in decreasing order by “hockey stick index”
3) Saved the top 100, i.e. a 1% cherry-pick to get the strongest upward blades
This was all found in 2010 by blogger Deep Climate, who investigated in great detail.
https://deepclimate.org/2010/11/16/replication-and-due-diligence-wegman-style/
Search for “SAVE A SELECTION” to see the explicit code, i.e. not a mistake.
Of course, the Wegman Report just ran M&M’s code…