Bishop Hill Keenan on climate statistics
Apr 4, 2011 Climate: Sceptics Climate: Statistics Climate: Surface 
Doug Keenan has just scored something of a coup by getting an article about statistical significance in temperature records published in the Wall Street Journal. It's a wonderful piece of popular science writing, explaining complex scientific concepts in clear simple English.
For years, some researchers have argued that the evidence for global warming is not nearly as strong as has been officially claimed. The details of the arguments are often technical. As a result, policy makers and other people outside the debate have relied on the pronouncements of a group of climate scientists. I think that is unnecessary. I believe that what is arguably the most important reason to doubt global warming can be explained in terms that most people can understand.
Bishop Hill
Doug tells me that if you Google it you should get free access - Google has an agreement with WSJ. I can't therefore provide a direct link. Try clicking here and following the link to Doug's article.
Reader Comments (51)
Unfortunately, that article is paywalled.
I had trouble with the paywall going in direct, but coming throught Google news it gave me a freebie.
A quick 'Google' provided this readable link
Correction - the link gave an 'Article Free Pass' with the full text but is blocked when I retry!
R2
Strangely, that article is paywalled for me.
This one worked for me
http://online.wsj.com/article/SB10001424052748704615504576171863463697564.html
Nope; that didn't work either. Anybody any good at copy and paste? I hope so after this teaser.
What worked for me: I searched in google for "keenan wsj". The article was the top hit. Clicking on it gave me a free pass.
Similar to Turning Tide I just google searched for the title and clicked through. Worked fine.
Try a web search on google for "how scientific is climate science", the first link returned by Google takes me straight to a readable article. If I cut and paste the link, as I did above, it hits the paywall.
Great explanation.
Figs, 6 + 7 really hammer home the point
Thanks Turning Tide, Google worked well.
Well written article, but will any politician understand it?
PPE does not seem to be a sensible education for anyone running a country whether UK, Europe, Australia or Europe.
Worth checking out the more detailed stuff at
www.informath.org/media/a41/b8.pdf
as well
'That would be gross negligence for a field claiming to be scientific to commit.'
Very telling
Good effort, Douglas.
And plenty of good stuff in the Comments, as well:
Very good lawyerly stuf. Not a word about CO2. This kind of stuff can be argued forever, and will be.
The wold needs a Feynman to collapse this CO2 garbage.
Who needs thermometers when you have satellites,satellite altimetry and land-ice?
Now we need a comparable article that deconstructs the GCMs. Any volunteers out there?
Excellent essay, well worth the wait. And good to see the WSJ not cow-towing to the Soros-funded groupthink. Time to post a link to my favourite graph of the year so far (courtesy of NikfromNYC):
http://oi49.tinypic.com/rc93fa.jpg
Since seeing Nik's graph, It has always mystified me how anyone could say that the slight temperature increase of the late 20th century was of any statistical significance. Now Douglas has explained it; there isn't.
I think the final graph highlighted by Doug, showing the changes in ice volume against solar intensity over millenia (from Gerard Roe's "In Defense of Milankovitch") is also up there.
But will any scientists within the IPCC change tack after seeing this damning exposition of their statistical incompetence? Or will they continue to let the alarmist politicians, activists and bureaucrats dictate the IPCC output? I think the short answer is no, but I would love to be proven wrong.
I accessed the article easily however the graphs would not load, "file corrupted".
I've given up on this, much as I would like to read it. Kept on hitting the subscribe now paywall.
Life is too short.
Going back to sorting out my bird photograph folders.
PW
Great article, Doug! Well done!
A bit OT, on the opposite side, Krugman has a very poor piece of Propaganda full of straw men in the IHT today. Quite amussingly he describes the UEA emails as "innocuous". Well Mr. Krugman, if you believe that you can believe anything.
@retephslaw (and anyone else who is having trouble accessing this article)
The Hockey Schtick blog has published it in full:
http://hockeyschtick.blogspot.com/
Doug Keenan's article is superb and takes apart the IPCC's self-promoted myth of it's own 'good science'.
The article takes some intensive reading for those such as I whose understanding of statistical methodology is somewhat tenuous, but his explanations are clear, succinct and lucid. Well worth the time invested in reading and understanding it.
My brain jumped unbidden to Donna L's 'No Concensus' article about the number of hats worn by the inner cabal (Coven?) of editors/reviewers/lead authors/etc at IPCC and the odours emanating from that August body seem to be increasing in strength and nastiness.
Very good and carefully worked out article (allthough the WSJ figures mgmt could be better)
This would make the perfect preamble to base a TSA course on.
The seminal BV-VS unit roots blog is now more than a year old , and our scientist Bart is still pruning it on a daily basis to make it all better look to his debased but rather incompetent satisfaction :)
The standard non scientific objection against t-series differencing as suggested by VS will be "why go through all that effort when the hadcrut is built up by such a vast amount nr of data(and coin throws)"
But you cannot do that.
You do not make data more thrustworthy by averaging more numbers from more sites etc.
Each site is a TS on itself . Merging data involves TSA techniques.
Only thing Goddard and UEA CRU and their many minions (calling themselves scientists) hv been doing all these years is averaging without a clue and making "regression lines" on paper "because they look good" that way.
Haavelmo !!
SPN, Sir Paul Nurse, the queendom's pippettrist, shld be asked for his opinion ?
And I do not mean this in any pejorative way. The chap seems intelligent and is approachable, JD will agree on that, and although enshrined in the clergy, capable of independent thought, as well.
He might make a Monbiot volte face..
I had trouble loading the WSJ page, clicked the first link before doing the google thing. I needed to delete my cookies file to get it to work after that. Hope this helps someone.
Turning Tide
Thanks for the link to the reprint of the Keenan article at the Hockey Schtick blog. Permalink here:
http://hockeyschtick.blogspot.com/2011/04/how-scientific-is-climate-science.htmlerma
phinnie
I was not aware that Bart was scrubbing the VS thread. On a regular basis?
You can look at this thread too, where seems to have performed a similar 'function'.
Roe, G. (2006), In defense of Milankovitch, Geophys. Res. Lett., 33, L24703, doi:10.1029/2006GL027817.
http://earthweb.ess.washington.edu/roe/Publications/MilanDefense_GRL.pdf
Shub
Re comments to the BV post you link: isn't dehog a prize pillock?
The very model of a modern alarmist-general.
The article is very clear on the basics of time series analysis and possible pitfalls. It is potentially powerful, but did I miss something? What is Doug's aternative to AR1? Without knowing this, I found it difficult to understand what the alternative explanation is to the increase in CO2 over the last 50 years. I realize it has something to do with sunlight and some type of heat reservoir but what is it?
@ Bernie - you mean the alternative to the increased temperatures resulting from increased CO2 in the last 50 years?
If so, my money is on a combination of a range of factors: natural variation, long term oceanic and solar-magnetic cycles, and decreased cloud cover. Black carbon could contribute to a little local warming in the Arctic but only for about 6 weeks in the summer, the rest of the year air temperatures are well below freezing/melting point. Aircraft contrails may also be having an effect particularly over Europe and North America, but I haven't looked into that in any detail. I don't think that increasing atmospheric CO2 from 0.085% to 0.0385% has anything to do with the warming. (That's an extra two molecules in every 10,000 by the way). The existing CO2 is already reflecting back almost all of the little IR which water vapour and methane misses - so it is a case of the law of diminishing returns or Beer-Lambert Law to be more precise. No matter, the warming has also been exaggerated to say the least, the only evidence I have seen for it is about half a dozen mild winters in the 10 years prior to 2005. Big deal, same thing happened in the 1930s.
As I usually say in this situation, enjoy the Holocene while it lasts -
http://jonova.s3.amazonaws.com/graphs/lappi/gisp-last-10000-new.png
bernie, and others, I spent quite a lot of time on Bart V's celebrated VS thread. VS was somewhat of the "take no prisoners" school of explanation, and plainly never managed to get through to Bart, who kept bleating back "but the physics says there is warming" and not listening to VS. It is sad if he is censoring it now.
For what it is worth, this is what I believe Doug Keenan and VS are suggesting, some of it explicitly, some of it implicitly. For what it is worth, Steve McIntyre also made similar points longer ago, see e.g. this link. I'm far from understanding all of this, so if anyone can tell me what I've got wrong I'd be very grateful!
Step 1. Forget completely any bias you might have that "physics predicts warming". Completely. This is for various reasons: (i) so as not to bias the way you look at the data, and (ii) because this might make sense anyway, because the physics showing that that should occur is not quite so well established as some people claim (aerosols! ocean heat content! response time to forcing!), (iii) to avoid the vicious circle where your temperature observations support your GCM model but your GCM model also supports your interpretation of the temperature observation.
Step 2. Look at the data. Different models of what is causing this time series will lead to very different shapes to the time series. If there is a steady, slow process superimposed by a random set of jumps from year to year, as would be the case if you just had steady warming combined with short-term weather, you'll get one shape. If on the other hand, you have persistent processes that act on a timescale of decades or at least > 5 years, superimposed by (perhaps) a steady warming, superimposed by year-on-year noise, then you'll get another shape.
Step 3. Do statistical analysis to see which of these is more likely to be the case, given the data you have. Immediately conclude that it is more likely to be the second one.
Step 4. Realize that if that is the case, then detecting the (hypothetical) underlying slow, steady, process, is statistically near-impossible. There may not be any steady, slow process! It might all be other effects to do with the sun, ocean heat content, La Nina, and so on.
Step 5. Think about physics again. Realize that the observations now no longer corroborate the GCMs at all really. Worry that the aerosol stuff that have been spatch-cocked in so as to fit the post-war plateau in temperatures really looks very very ad hoc now. Realize that if the GCMs do not reproduce the 5-10-20 year excursions of the temperature, up and down, without the aid of such aerosols deus ex machina, the GCMs may be missing some important physics, as well as having assumed terms that might be wrong. Start wondering what sort of heat reservoir effects might be operating.
lapogus and j:
Thanks for the responses. I guess I was not clear. What is Keenan alternative statistical assumption - if it is not AR1 and what is the concommitant physical process?
bernie, I think he does not properly spell out his alternative. I think he does not seek to assign the physical process. For those who believe that there are no 'natural' forcings that can possibly lead to temperature variations of 0.2-0.6 degrees, this is enough to dismiss Keenan, VS and their ilk out of hand. This was part of Bart V's approach. For my part, I think that it is interesting to note that the observed temperature record may not be well explained at all by the "steady process + white noise" model, without going on to state what does explain observations well.
Bernie, there are more details at http://www.informath.org/media/a41/b8.pdf
He seems to be referring to the paper by Cohn and Lins which is discussed in some detail on the "Un-naturally Trendy" thread at Lucia's blackboard.
Keenan appears to be using a driftless ARIMA(3,1,0) model - see his Supplement, accessible via a link from the WSJ article. So this is a quasi-random walk process (with a unit root, and AR3: orders 1, 2 and 3 auto-regression coefficients).
Keenan doesn't discuss any physical process, but he does point out that AR1 is probably not an physically realistic assumption.
There is certainly evidence that various climatological time series exhibit long term memory - corresponding to a fractional root (Hurst exponent, H, between 0 and 1)
Since the seminal unti root thread came up here, and my role therein, let me clarify that I certainly listened carefully to VS' arguments. That attitude was unfortunately not returned, as any suggestion to come to some resolution that would incorporate his statististical analysis with a more physics based model (e.g. taking net forcing rather than only CO2 concentrations into account) were ignored.
Not sure what exactly people mean by pruning and scrubbing that thread, but I haven't deleted stuff there recently (except at the time for comments that were way over the line). My site is still open for people with different opinions to discuss things. That doesn't mean I don't have limits (as ptw can attest to), but they are fairly lenient.
Shame on Bart if he is pruning the unit root thread.
Has he given any justification for why this should be done?
Does anybody have a copy of the "full fat" original?
Anybody able/prepared to host it as an archve?
Bart - apologies, should have refreshed before posting.
bernie,
My take is that he is trying to make a basic, simple point about the incompetence underlying the IPCC. He uses the assumption made in AR1 [i.e. "that this year's temperature affects next year's, but that the temperature in previous years does not. Intuitively, that seems unrealistic."]
Checking that assumption is something which has to be done. It's recklessly negligent not to. Yet, the IPCC does not include any indication that it has been checked. And Keenan's checks indicate that the assumption is wrong.
It is my opinion that climate science is overwhelmed with examples of corruption and incompetence. This example provided by Keenan certainly qualifies.
BV-VS : well I had a post in there at the time warning about the risk of scrubbing to our scientific heritage. I guess that post was "over the line" then :)
At least the blog still exists.
stick to the fissix:
Haavelmo (the econometrist ) his contemporaries also could have remain stuck to the Navier Stokes equations, only.
After all , consumers behaviour is determined by their thinking, which is determined by blood in veins to their brains and acidity levels in neurons before they fire. So, it's all fixed by Navier Stokes. So lets stick to the fissix. Why pay for all the psychologists etc why not just have Navier Stokes scientists, only.
Instead Haavelmo considered an easier approach, and got along nicely with the observed data and creating simple models that fit that data. It brought something unheard of to the kind of characters that populate institutes like the IPCC or the Dutch dust-counting asylums . It brought verifiable RESULTS
Thinking about the data indeed implies using correlation . In fact that's what thinking is, mostly.
Courtillot;s recent presentation insisted on correlation as well.
Correlation was VSs next planned step and proved a bridge too far for certain scientists.
Is not to difficult to see whyso:
The day an idiot-savant, somewhere, starts to correlate GHG with temp, he'll probably find out something's aloof with the years 1998-2011 and the corresponding GHG emissions..
As Keenan notes in his article: intuitively AR1 is the wrong choice.
Intuitively, temp is a measure of earth's "warming", i.e. its calorific content (which is on itself a foggy thermodynamical extensive property , as well), at least that's how it's sold to us via the meejah.
well, that temp = calorific content = " heat energy" is known to be subject to lots of action and multi-year oscillations (sun, el nino , la nina etc)
So here is the real fissix knowledge , on the scale of the observeables, immediately chucked in the bin, with AR1.
This is no science. This can only be called science by a google team with too much time and money on their hands.
GHG(CO2) is an observeable , btw.
totally different level compared to mickeymouse GCM variables like "net-forcing"
totally different ball park, that.
Anyways, observables:
Courtillot's talk after minute 4 is revealing. What quality can we attribute to an observeable that is been ringfenced by the team?
His remark on not finding young people to work for him is of the same chilling tenure as Nigel Lawson's mentioned problems in getting "an appeal to reason" published.
It just proves the BBC and other leftwing doce vita houses geared up to spin only ,should be closed down. They are a threat to our civilisation.
I did check the additional material. It helps but does not really identify the assumption that Doug is championing. It looks like Doug is arguing for the existence of long term persistence as a better assumption of the underlying process rather than a simple process of a year over year increase in CO2 (or any other relevant forcing). I cannot follow all their math, but the Cohn and Lins conclusion is pretty emphatic:
"But could this warming be due to natural dynamics? Given what we know about the complexity, long-term persistence, and non-linearity of the climate system, it seems the answer might be yes. Finally, that reported trends are real yet insignificant indicates a worrisome possibility: natural climatic excursions may be much larger than we imagine. So large, perhaps, that they render insignificant the changes, human-induced or otherwise, observed during the past century."
Demetris Koutsoyiannis and Alberto Montanari(2007) make a more general argument - again the details of the math are beyond what I can work through - but my take away is that compared to assumptions of short term persistence, the assumption of long term persistence leads to greater uncertainty and less certainty as to the significance of any identified trend.
Demetris Koutsoyiannis and Alberto Montanari highlight and endorse another quotation from Cohn and Lins which nicey sums up the point Doug is making: "From a practical standpoint . . . it may be preferable to acknowledge that the concept of statistical significance is meaningless when discussing poorly understood systems.’’
Keenan, Cohn amd Lins, and Koutsoyiannis and Montanari are a making the same point - climate scientists like Jones, Mann, etc., have assumed a model of how climate works that is both unvalidated and leads to a potentially large underestimation of the statistical uncertainty attached to any trend that they identify.
Please let me know if I have misunderstood the key issues.
One thing still puzzling me,
Keenan:
"For the global temperature series, it means that this year's temperature affects next year's, but that the temperature in previous years does not. Intuitively, that seems unrealistic."
I agree with that, but the IPCC report seems to have a 'proof':
IPCC Table 3.2. :
"The Durbin Watson D-statistic (not shown) for the residuals, after allowing for first-order serial correlation, never indicates significant positive serial correlation."
Are these D-stat calculations available somewhere?
Nic Lewis, that is all correct.
bernie, your second-last paragraph is right.
The article does not propose using any statistical assumption. The alternative assumption, a driftless ARIMA(3,1,0), is used solely for comparison with the IPCC assumption: “We don't know whether the alternative assumption itself is reasonable—other assumptions might be even better …”. The ARIMA(3,1,0) is “so much better that we can conclude that the IPCC's assumption has no support”.
UC, the DW statistic is solely a test for first-order autocorrelation, nothing else. The IPCC applied the test after removing the first-order autocorrelation (via their AR1 modeling). An analogy is a test for a mean of 0 being applied after subtracting the mean from the data. They obviously did not understand what they were doing.
There is now a copy of the article on my web site:
http://www.informath.org/media/a41.htm
The figures there are slightly different than those published by WSJ. I submitted the article with (essentially) the figures at the above link. WSJ redrew the figures. WSJ apparently redraws all graphs: they require authors to submit the data, and then they draw the graphs themselves. Doing so is a quality control: it guarantees that they have the data for the graphs. Too bad science journals do not do that! Because the redrawn figures are copyrighted by WSJ, I used the originals on my site.