Bishop Hill Batting back at Beenstock
Feb 6, 2013 Climate: Statistics Climate: Surface The Beenstock, Reingewertz, and Paldor paper that did the rounds before Christmas applied some whizzy statistical methods to the temperature and forcing trends and found that they were statistically independent of each other. There's a useful discussion of it here at David Stockwell's site. As Stockwell notes, this is not a debunking of AGW, but rather of the use of linear regression to "demonstrate" that something unusual is happening in the temperature records.
To that extent, Beenstock's paper should be entirely uncontroversial. I'm not sure that anyone really thinks linear regression is a suitable approach to apply to temperature records. Nevertheless there has been a rapid rebuttal posted to the journal in the shape of an article by two Oxford academics, D. F. Hendry and F. Pretis:
We demonstrate major flaws in the statistical analysis of Beenstock et al. (2012), discrediting their initial claims as to the different degrees of integrability of CO2 and temperature.
Reader Comments (26)
Scuse my ignorance but what are "integrability properties"?
-----------------
OK - I started to look at the Beenstock et al paper. They explain "cointegration" tests for detecting spurious correlation between nonstationary time series that, in reality, are independent. I imagine that is what Hendry and Pretis are referring to.
I am at a loss to see what this paper is driving at; it points out that the there is not enough accurate data to allow the use these statistical techniques. I think we all knew that anyways, but then goes on to say that a simple plot (of the poor data) matched by means and ranges, suggests the obvious: they are closely related. Carefully missing the point that it does not show a cause and effect releationship either way. I am often reminded of my statistics lecturer (yes I did study statistics at one point and found it extremely dry and boring) who would use the example of the increase in lamposts in Birmingham and the matching rise in the number of gay men living in the city. Not related in any obvious way but 'matched by mean and range'.
The concerns raised by Hendry and Pretis do not just affect the results of the Beenstock et al. paper. They affect most (all?) earlier papers too, including those papers that do find a relationship between greenhouse gas concentrations and temperature.
Would some kind soul please prepare a precis suitable for the educated layman about this controversy?
Somebody with a scientific background, a bent towards numeracy and writing skills would seem to be the obvious choice....unless pastoral and diocesean duties intrude?
@Latimer Alder
Hendry and Pretis argue that the data for CO2 etc are messed up. If you don't account for that, your statistical analysis is garbage.
Their temperature plot in figure 4 looks wrong to me. Where are the last 16 years of static temperatures, that I thought we all agreed on?
This is really interesting. Hendry and Pretis state:
When I posed a question on Science of Doom about how the logarithmic relation was derived, I was told that it came from fitting a curve to the results from numerical models - it was not derived analytically.
Are Hendry/Pretis bullshitting*? Or did Arrhenius really derive a log relation between CO² concentration and atmospheric temperature change that is considered valid today? If so, SOD need to be put right.
(S.L.B.T.M)
_____________________________________________________________________________
*Bullshitting = speaking with an air of authority about something you don't really understand.
Harry Frankfurt "On Bullshit"
"different degrees of integrability of CO2 and temperature."
Even assuming that's English, the recent divergence between rising CO2 and conspicuously-not-rising temperature would appear to suggest otherwise, vindicating Beenstock.
@richard tol
A while back I remember you saying something to the effect that the convention in academia is that you may not say that somebody's work is wrong unless you can also provide something better...and that is why so many academics let shoddy work in 'climatology' and other fields go by.
Yet here is a prime example where two guys don't present an alternative.. just say that the previous paper is crap - and seem to have good reason to do so. And they pull no punches...'Discredit'...'major flaws' 'none of their claimed conclusions has any evidential basis' is fighting talk.
Is it relevant that they are not climatologists - so presumably their funding comes from another stream and hence they are not subject to 'peer pressure' to conform to the received wisdom?
Or did I just totally misunderstand your point?
Latimer, didn't you see the name of Myles Allen in there? This appears to be a quick response from the other side to discredit a paper which casts doubt on the essence of warmism. Of course it needed to be rebutted fast. Note the short timescale to get it published. Note also the last graph of temps vs CO2. A sceptic would see at least three different slopes in there. Three different regimes, maybe. Just cos you can draw a line over a cherry-picked period does not make it true, or false, it's just a line.
This is the discussion "a while back" that Latimer refers to implicitly:
http://www.bishop-hill.net/blog/2013/1/25/uniform-priors-and-the-ipcc.html
Am I reading this right?
Hendry and Pretis spend some time explaining why the Beenstock etc al conslusions cannot be supported using the data presented, then finish with this:
Haven't they just demonstrated that this conclusion is impossible?
Hendry is a prominent econometrician. In 2006, I did a CA post (http://climateaudit.org/2006/07/10/alchemy-or-science/) on a 1980 article of his on spurious regression, in which he ironically analysed the properties of his new "theory" of UK inflation: he did a regression of UK money supply against cumulative UK rainfall, showing that it passed some standard regression tests though being a worthless theory.
Hendry is indeed heavy weight in the econometrics world and a pioneer in the area of cointegration and time series anlysis of non-stationary variables.
He would make some valid points in the criticism, but his introduction is poorly written and makes zero sense.
In that Hendry expounds on the radiation absorption properties of CO2 and the posited log relationship. It is that certainty that leads him to the conclusion that the analysis must be flawed and the only thing that can be done is to find the errors:
Thus, greater concentrations of greenhouse gases 15 increase the amount of absorption and hence re-radiation. To “establish” otherwise merely prompts the question “where are the errors in the Beenstock et al. analysis?”.
But of course Beenstock et al were only establishing what the whole system response was - i.e. including all the system feedback. Somethnig different entirely. In fact that is the reason why I don't think it a paradox at all to believe that CO2 can physically act as described, absorbing radiation in a log relationship, but that global tmeperatures may be weakly responsive to increases in CO2 in the atmosphere.
I am certainly flabberghasted that Hendry would make such an error, given his knowledge and work in macroeconometric models. An analogy would be Hendry arguing that a paper that found no relationship between wage rates and unemployment in natural data MUST be wrong, because demand is known to be a function of price.
But what is most discouraging is that this paper was "supported by grants from the Open Society Foundation". That is George Soros social policy playthnig for those who don't know.
I hasten to get in my CYA caveat that I have not read either paper through, and while I would very much like to, I have a long list of other tasks on my plate to do first.
So what does a casual observer, then, make of this tussle of probably greater minds than his?
Bemusement so far.
Hendry and Pretis [http://www.earth-syst-dynam-discuss.net/4/219/2013/esdd-4-219-2013.pdf] say this early on:
This strikes me as problematic right away. The Beenstock et al. analysis surely does not aspire to establishing anything contradicting the basics physics of radiation absorption and emission. This therefore looks bemusingly like something between a red herring and a straw man
All the Beenstock et al. paper has shown is that, in their type of analysis, the airborne CO2 series and global mean temperature series do not behave as they might if there were clear causal links between them. This does not mean that there are none. It merely means that if they exist they are relatively weak compared to variation from other sources.
Now in a complex system such as all agree the climate is, it is not at all strange that there can be found networks of causal relationships in which some will be more important that others, and some will tend to be harder to detect than others in the overall variation.
Beenstock et al. are merely saying that if time series analysis was all we had, we would not be able to derive convincing evidence for recent CO2-driven global warming as a mechanism from it.
Or am I missing something? (I quite possibly am - see first sentence - but couldn't resist sharing some first reactions).
What is the reference for figure 4? GISS?
Whoops, I see my cogitations took place while Geckko's post was already in place with the main fruit of them and more besides. Preparing comments can put you briefly into a little closed world, that you leave after rushing to post, and then later are able to look around a bit more.
John - see my comment at 11.26AM.
So far as I can see they are saying (my translation):
"A log relation between rise in atmospheric temperature and atmos CO2 was rigorously established in 1896. So someone whose analysis of observed data shows no relation between atmospheric temperature rise and CO2 has to be talking bollocks. All that is left is to explain exactly how they are talking bollocks".
Martin, There were other profound quotes from circa 1896: "everything that can be discovered in physics has been discovered", and, "heavier than air flight is impossible". (Or something like it). This was the scientific consensus.
Surely, you don't think thety were talking bollocks as well?
Interesting stuff, Martin A. Sometimes it looks like we have not progressed much beyond the simplistic calculations of Arrhenius. His 1896 paper describes some of the simplifications he had to make:
[Source: as quoted in Fleming's book 'Historical Perspectives on Climate Change']
Of course, to compute just about anything in the climate system, you have to simplify. His simplifications included no heat transfer horizontally in the atmosphere. Now this is the big task of the atmosphere every day of the year: to transfer heat from tropical to polar regions. It is the biggie, the basis of our climate and of our weather. Is it not plausible that the factors involved in this, which include an appreciable role for clouds and water vapour, will obscure the modest impact of CO2?
Contemporaries of Arrhenius who at first found his CO2 and temperature theory convincing, were later to change their mind. I am thinking in particular of Chamberlin who later decided that the role of CO2 in the system had been exaggerated (acc. to letter of his in 1922, referenced by Fleming, loc cit).
@Latimer
Indeed. Hendry and Pretis are just bitching about Beenstock's work. This is unusual, but then Hendry is a Giant in semi-retirement, so what does he care.
In fact, their critique applies just as much to all papers that do find a relationship between CO2 and temperature. They thrash the entire literature.
The Giant and the Beenstock. Where have I heard that before?
Richard Tol is correct that if all the caveats and sources of error Hendry and Pretis invoke were applied more broadly, we would have to conclude that there was no empirically derivable relationship among the relevant time series. But their framing seems obviously slanted, since the physics story they tell might be an example of precisely the "omitted variable" problem they later flag. What is at issue is whether their physics story is a complete account and if/how that story needs to be modified when other factors are included.
Methinks that Hendry doth protest too much.
All Beenstock et al. aim to show (and they succeed pretty well) is that cointegration analysis (which is a standard econometric technique for establishing the existence of relations between trending data series) does not show a relation (causal or otherwise) between CO2 and temperature. At no point do they claim no such relationship exists, only that this particular technique cannot establish it, because the orders of integration don't match.
Hendry and Pretis show that by cherry-picking a subsample you can get a different result. However, I think we need a theory of how the presumed break in the series happened when it did, before we can take that over-seriously. The point is that their Figure 4 is a classic illustration of spurious correlation. The two series both increase over the chosen sample period, and can be over-laid "after matching by means and ranges". Big deal! The power of cointegration is to be able to establish conclusively a relation between randomly changing trends ("random walks") by fitting the random changes together. Two straight lines always cointegrate, trivially, which tells us nothing.
Time series analysis has at best limited power to establish the existence of causal relationships. Spraying around words like "fallacy", "flaw", "discredit", and so forth is machine-gunning a straw man. Is there an agenda there somewhere?
In speaking of the co-relation of CO2 rise and temperature rise, I musta chalked 'last quarter of the last century' a hundred times @ the Blackboard, but bright as she was, Teach couldna see it.
======================
@James Davidson
There is a trend break in the CO2 and other greenhouse gas series around 1960. Before that, measurements are inaccurate and not equidistant. After that, measurement are fairly accurate and continuous.
Estrada and Perron have a paper, conditionally accepted for PLOS ONE, where they show that cointegration between CO2 and temperature comes apart when tested with the possibility of a trend break.