110 thoughts on “Q/As in Parliament: Keenan Kerfuffle.”

1. lucia,

   All you have to do is calculate the 95% envelope for DK’s ARIMA(3,1,0) using a Monte Carlo method and you’ll see why it’s ridiculous to use even if it weren’t physically unrealistic by being I(1). If you make your noise model complicated enough, you’ll never find a significant trend because the variability is literally wall to wall with random walks.

   Or try calculating your position by double integrating accelerometer readings from your smart phone. Unless you have a nuclear submarine/ICBM grade inertial measurement device, the calculated position deviates from the true position by meters within seconds.

2. DeWitt–
   Yep. That’s what diffusive systems do. It’s ok in diffusion problems where things diffuse. But it’s silly to suggest the earth’s temperature is “diffusive”. As the Met office says in the first question where Lord D mentions the I(1) type models:
   “often lack physical realism in the case of temperature data”.

   This answer was given to Lord D. He appears to simply ignore that part of the answer. My experience is that Doug Keenan essentially does the same thing.

3. Well I took the HADCRUT4 global data set. Then I took the rate of temperature change over a 31 year period, then I plotted the data.

   http://i179.photobucket.com/albums/w318/DocMartyn/HADCRUT4Global_zps5c19cb37.jpg

   The average rate is 0.714 degrees per century, with a standard deviation of 1.01 degrees.
   However, the lineshape of the slope is not noise. the line shape is quite clearly information rich. I could not claim that the rate is a function that can be understood or modeled into the future.

4. This exchange between Lord D and Baroness V is so uncontroversial, limited and dull it’s hard to see why anybody would want to spin it into something more.

5. from your first exchange:
   “whether those figures suggest there has been any significant global warming since (1) 1998, and (2) 1880”

   when the government replies with a number (0.8C), that needs to answer the question of whether there has been significant global warming. Answering with a number, which isn’t significant, is to fail to answer the specific question that was asked.

6. HR: They did spin it though; did they not?

7. per–
   Significant has multiple meanings in English.

8. Here is how Doug Keenan described the process:
   “Lord Donoughue then tabled a Parliamentary Question asking HM Government for their assessment of the likelihood of the trending autoregressive model relative to the driftless model. HM Government did not answer. Lord Donoughue asked a second time. They did not answer. He asked a third time. Again they did not answer. He then asked a fourth time.”

   Sure enough folks at WUWT and BH wanted Dr Slingo sacked for refusing to answer questions.

   Doug Keenan is not to be believed.

   Incidentally, Lord Donoughue has given his version at WUWT.

9. Now…see the way I look at it
   4 January 2013 Lord D asked about the ARIMA. The government gave a perfectly respectable answer. Admittedly, one needs to understand a little statistics and a little physics to recognize it’s a perfectly soudn answer, but it’s a perfectly sound answer.

   5 Feb 2013 He asked again, and they elaborated further, and suggested he look at their previous answer. Given the fact that he reposed it, and his new version did not even seem to recognize the importance of the information in the previous answer, I suspect Lord D did not consult with anyone with any background in the physical sciences. Otherwise, he might have understood what “often lack physical realism in the case of temperature data.” meant.
   In particular ARIMA(3,1,0) lacks physical realism for temperature data and should be excluded for that reason.

   Lord D asked again, the government elaborated further and suggested he look at their previous answer.

   It seems to me that if Lord D did not know what “lacks physical realism for temperature data” meant, he should have asked someone until he could discover what it might possibly mean. Then he either could have (a) accepted the reason the Met office gave or (b) asked another question like “Do you think ARIMA(3,1,0) lacks physical realism for temperature data and if so, why do you think that?”

10. Oops, I just realized I asked a rhetorical question above. Sorry. ‘Won’t do it again.

11. I posted a response to Lord D at WUWT. By accident I first posted it here; I’ve tried to delete it, but I’m not sure if that worked. Apologies.

12. Do you mean:
    Nick Stokes (Comment #114219)

    It still shows for me. But that could be cache? Hmmm… no it’s probably there. Do you want me to delete it?

13. “Significant has multiple meanings in English.”
    okay. The numbers being compared are complex, susceptible to statistical analysis and have been statistically analysed. So if these numbers are not statistically significantly different, and you know that, in what way can these numbers be “significantly” different ?

14. Thanks for this posting about an arcane tempest in a teapot, full of sound and fury, and as you point out signifying almost nothing. With apologies to the Bard.

    Meanwhile Cooks 97% nonsense garners global MSM attention and Presidential tweets, yet rebuttals mainly in the sceptical blogosphere. AR1 versus ARIMA, no matter how right or wrong technically, simply is not of the same magnitude of political impact in what is increasing a political rather than scientific debate.

    What is relevant is that the science isn’t settled, the GCM’s are being statistically invalidated, ECS is something like half of what was thought, and we can observationally begin to explain why.
    Those are the sorts of points that need to be hammered home again and again. Falsification of GCMs by the temperature pause would seem a good platform for you and your most ardent and statistically knowledgeable community.

15. Lucia,
    Yes, thanks. What happened was that I pressed delete in good time, but I was switching between browser windows, and it threw up a dialog asking for confirmation, which was buried. By the time I confirmed, it must have been too late.

16. per
    Significant doesn’t necessarily imply “statistically”. It can just mean “are they sufficiently large to be of concern”. Sometimes it can just mean “Is it big enough to detect”, though that usage would be a bit unusual.

    The numbers being compared are complex, susceptible to statistical analysis and have been statistically analysed.

    Your previous objection seemed to be that they were not analyzed statistically in the first answer to the question (October 2012) when they should have been. But they need not have been because the word “significant” doesn’t necessarily imply “statistically”.

    Later as questions were reworded, and represented, statistical answers were provided.

17. per

    So if these numbers are not statistically significantly different, and you know that, in what way can these numbers be “significantly” different ?

    BTW: Despite your inserign “you know that”, in the sentences, I don’t know which numbers you are discussing here nor what is supposed to be statistically significantly different.

18. RUd Istvan (Comment #114234)
    I’m puzzled. Are you complaining I haven’t posted on Cooks stuff? I think I’ve probably posted more than any other blogger.

    Feel free to hammer home anything you want to hammer home at your blog. But, honestly, comments that sound like you are giving me marching orders about what topics to discuss at my blog are not especially welcome.

19. Lucia – here is Lord Donougue’s comment at WUWT:
    ++++++++++++++++
    Lord Donoughue of Ashton says:
    May 29, 2013 at 10:31 am

    I am the Labour Peer who embarked on this exercise. I am not a scientist nor statistician, though with some background in economics. Became engaged for three reasons. First I was appalled by the terrible energy costs inflicted on the poor. Second, that I realised when talking to Green alarmists that I was dealing with a faith cult not interested in facts. Third, that in Parliament and the rest of Europe we had launched a huge energy strategy, with enormous implications for our economic competitiveness, which needed stronger justification than I had encountered. At first I believed the ministerial assurances that there was a total scientific consensus for the alarmist predictions and when, with the help of sites such as this I saw this was not true, I became more concerned. As a former adviser to the tv series Yes Prime Minister, I am aware but wary of political schemes to con the public, especially where some of the proponents, as those conducting Parliaments Climate Change Committees, may have financial interests.
    My worries grew as I asked my Question (there was no confused discussion as one of your bloggers suggest – just a straight question and a series of waffle responses, not an Answer as is required by Lords custom). In 28 years in Parliament I do not recall such obfuscation. The fact that most of the non answers came from the Met Office and not the responsible Department of Business, raised my concerns.

    One blogger suggests that heads should roll. That perhaps applies to the Met Office , but not to the relevant minister, Baroness Verma. She signed responses to a technical Question. Ministers are usually, as I was, generalists, not technical experts and have to trust their officials to give proper and honest replies, as the Ministerial Code requires. When it became clear that they were ducking the issue, she ensured that the sixth Question received the reply which the MO clearly, and for understandable reasons, exposed in this blog, did not want to give.

    I began with the short Question whether the claimed rise in temperatures was significant because that seemed the right place to start. It published their assumption. Then it was possible to probe the basis for that – with a model which might or might not be adequate to support the huge economic and social implications which governments of both parties have accepted and imposed.
    I shall start from the basis that this is the greatest transfer of wealth from the poor to the rich (including some of my friends) since the enclosures of the eighteenth century. It beggars belief that my Labour Party introduced it, with such unquestioning good will).

    Whether Parliament or Government will follow this issue up and question the sandy basis on which their huge superstructure of energy reconstruction and massive financial imposts on industry and the community is built, remains to be seen. It will not be easy , since questioning a self righteous faith is never easy. But I propose to persist, despite obfuscation from departments and moral disapproval from political colleagues. In this I am greatly helped by brilliant Doug Keenan (who I have never yet met, but will) and the GWPF. Suggested Lords Questions, brief and factual, which might extract helpful factual information, on this site, will be helpful. My general policy position was stated in the Lords Queens Speech on the Tuesday of that debate. Warming alarmists have already consigned me, a Catholic, to Hell, so correspondence will find me at that address.l
    Bernard Donoughue
    ++++++++++++++++++
    http://wattsupwiththat.com/2013/05/27/uh-oh-the-met-office-has-set-the-cat-amongst-the-pigeons/#comment-1320662

20. my take is that the questions were badly posed and that neither the person who claimed to ask them nor the spokeperson who had to reply knew what they were saying. oddly Lord D does not say who phrased the questioms that were put into his “mouth”.

    At the end of this weird dysfunctional exchange of questions and answers, is anyone further forward in understanding?

21. Curiously it’s a Conservative peer (rightwinger) defending the consensus against a ‘skeptical’ Labour peer (leftwinger).

22. Lucia,

    I still honestly do not understand your stance here.
    The critical question was posed on Nov 8th and it did NOT ask whether the temperature rise was “statistically significant”.

    To ask Her Majesty’s Government, further to the Written Answer by Baroness Verma on 30 October (WA 114-5), whether they consider a rise in global temperature of 0.8 degrees celsius since 1880 to be significant.[HL3050]

    The Met Office had every opportunity to interpret “significant” as “meaningful”. If the question HAD raised the issue of statistical significance, then the correct answer would have been the one suggested by Doug McNeall (long after the event) explaining that it is not a question that can be validly answered. Or the Met Office could have voluntarily made the distinction clear.

    Instead, someone in the Met Office volunteered/ that the temperature rise was statistically significant (my bold below):

    Statistical (linear trend) analysis of the HadCRUT4 global near surface temperature dataset compiled by the Met Office and Climatic Research Unit (table 1) shows that the temperature rise since about 1880 is statistically significant.

    All of the Met Office’s problems stem from that answer.
    Try as I might I am unable to put any gloss over this. It lends itself to only one interpretation as far as I can see.

    Suppose I were to approach you professionally with a time series of acoustic amplitude of length 1100 seconds recorded by a point micro-seismometer. I tell you that we can see that the amplitude increases irregularly from 1000 seconds to 1100 secs by about 0.8 units. Because of a monitored seismic event some distance away which occurred at 1000secs, it is important for us to know whether that increase in amplitude is significant. I want your best technical opinion. You run a Fourrier analysis and identify two clear oscillatory signals throughout the length of the record, and you learn that there were two machines operating nearby with the capability to produce those signals. You filter out the signals and find that the residual is AR(1) noise. You report that you see no evidence of any statistically significant change in the residual over the target interval. Conclusion: the increase in amplitude is not statistically significant.

    Alternatively, you fit an AR(1) linear trend model to the data from 1000 to 1100 secs. You note in passing that the model has low information content and the residuals show clear evidence of mis-specification based on a number of tests. However, you are excited by the significance levels. The gradient is clearly positive and is significantly different from zero. Conclusion: the increase in amplitude is highly significant.

    Which technical advice is more valuable?

23. Lucia, I lack the cyberspace skill you have to respond specifically to your reply above. Sorry, have too many other responsibilities. We have a failure to communicate.
    What was meant to say was simple, not complicated or derogatory:

    Take it to them where they live, not repeating trivia to us where we already do.

    Co- author on a separate topic on a separate blog, might I hope you agree? Brandon, you know who/where you are?
    No contest even intended. I will phrase better if any next response.
    Regards

24. Paul_K #114253
    “The critical question was posed on Nov 8th and it did NOT ask whether the temperature rise was “statistically significant”.”
    Didn’t they? Beats me. But here’s what Doug Keenan said about that answer:
    “A proper Answer to the above Parliamentary Question must not only state Yes or No, it must also specify what statistical model was used to determine significance.”
    That’s what they are up against.

    “Instead, someone in the Met Office volunteered/ that the temperature rise was statistically significant (my bold below):”
    After a lot of badgering see above. Eg:
    “Lord Donoughue then sent a strongly-worded letter to Under Secretary Verma, citing the section on Ministerial Responsibility, and adding “I trust we will not reach that point since you are clearly not behind the wilful refusal to answer the Question”.”

    That’s how it goes. Ask an essentially meaningless question. Monster the scientists with a House of Lords. Get the Met to echo a crank calculation. And then, gotcha!:
    “Met Office admits claims of significant temperature rise untenable”

25. PaulK–
    Why are you extracting a snippet from the full 3 paragraph question to talk about what they could or could not do? They had to explain their basis for “upward trend” based on the earlier paragraphs. They did that by volunteering the descriptive statistics they use. Using that descriptive rule, they define “flat”, “upward”, “downward”. Of course the ‘volunteered’ this information– in the sense that thats the rule the use and that’s their basis for answering simple queries. That doesn’t mean it’s what they use to decree a “signal of AGW”.

    Which technical advice is more valuable?

    You are assuming Lord D question is assumed to be requesting techincal advice. If you read it, at that point in time he seems to just be asking them to describe the method they use to make announcements. It’s a method which they apply generally.

    Later, in the Kerfuffle, it is being treated as some sort of “diagnosis” of AGW from the signal. That’s not necessarily what the Met office thought they were being asked.

26. RUd Istvan

    What was meant to say was simple, not complicated or derogatory:

    Ok. I’m not meaning to snap at you– But I guess I don’t know what that simple thing you meant to say is. It sounds like you are telling me I shouldn’t discuss this topic. Did you mean something else?

27. Nick–
    It would be interesting to interlace dated letters in here. But I don’t have those letters nor the dates.

28. lucia (Comment #114259)
    “Nick–It would be interesting to interlace dated letters in here.”

    I’m very reluctant to rely on Doug Keenan for anything factual. But he says the strongly worded letter came between the fifth and sixth answers. He describes some other correspondence where he is remonstrating with the Met director. It’s all about getting them to echo his 0.001 number so he can write his gotcha headline.

29. Rud Istvan:

    Take it to them where they live, not repeating trivia to us where we already do.

    Co- author on a separate topic on a separate blog, might I hope you agree? Brandon, you know who/where you are?

    I don’t agree this is what lucia ought to be doing. This is her blog, and she should do whatever she wants with it. Your advice only applies if lucia’s purpose is to advance a cause. If that’s not her goal, she has no reason to do as you say.

    Your strategy is correct. The problem is your strategy is designed for one particular goal, and it isn’t lucia’s goal.

30. Lucia
    You have failed to understand Lord Donoughue’s question that kicked the whole thing off.
    .
    Do you really believe Lord D wanted the Met Office to calculate linear trend magnitudes for him? Wouldn’t the Lord feel his intelligence insulted if the Met Office put this sort of a Stokesian spin on his question?
    .
    The question is: “What is the basis…?”

31. Lucia
    A second thing. In the whole of Keenan’s main point, there is nothing about ‘anthropogenic’ in there. It is about detecting the ‘global warming’ signal.

32. Hi Lucia,
    “Why are you extracting a snippet from the full 3 paragraph question to talk about what they could or could not do? They had to explain their basis for “upward trend” based on the earlier paragraphs. ”

    Only because I am trying to identify exactly where the Met Office crossed the line into a truly unscientific response. You are writing as though the Met Office were somehow led to respond in this manner because of some cunningly contrived process of entrapment. Please note that these three written questions which talk about trends, the “basis for” and “significant” were presented simultaneously. There is no arm-twisting to make the Met Office use the term “statistically significant” in their response. They decided to walk into the quagmire of their own free will.

    I repeat what I have said before – I think that Doug McNeall was perfectly correct when he stated that there is no valid answer to the question “is the temperature rise statistically significant?”
    You appear to be suggesting that there is a valid answer or am I misinterpreting your stance?

    Can I ask you to set out rather more pedantically:-
    (a) what specific question do you believe was being tested in this significance test applied by the Met Office, given that it was already asked and answered that temperatures were about 0.8 degrees higher now than in 1880
    (b) what do you think was the null hypothesis in this test if not covered in (a)
    (c) what is the structural justification for the choice of the AR(1) model in this test if not covered in (a) and (b)?
    Thanks, Paul

33. Nick Stokes (Comment #114256)
    May 29th, 2013 at 8:16 pm

    Nick,
    You are confounding two separate things here to make it look as though the Met Office was somehow badgered into bad science. They were not.

    According to both Lord Donahue and Keenan, Lord Donahue formulated the initial suite of questions himself. It was this suite of questions that led the Met Office to describe the temperature rise since 1880 as “statistically significant”.

    The “badgering” as you call it related not to the initial suite of questions but to the later question, which was posed and re-posed multiple times and which was not answered. The question was about the relative likelihood of the driftless ARIMA(3,1,0) vs the linear trend AR(1), and again there seems to be no controversy about the fact that this question was posed after dialogue between Donahue and Keenan, and as a direct result of such dialogue.

    As for the use of the term “badgering”, you should be aware that UK parliamentary questions are governed by a certain tradition well established over 400 years.
    The rules are:-
    – The question must always be answered or a clear reason given for why it cannot be answered.
    – The answer must be honest, but doesn’t necessarily have to be complete.
    – If the responsible Minister is caught out in a dishonest answer, knowing that it is a dishonest answer, then he is obliged to fall on his sword.
    The threatening letter from Lord Donahue (which you describe as badgering) was because he posed the same question (about relative likelihood between the models) multiple times and did not get an answer. This has not happened for three decades, as far as I am aware, although some parliamentary historian may prove me wrong.
    I also don’t think it is helpful to the discussion in any way to make snide innuendo about Keenan’s honesty, especially when in this instance the salient facts are largely in writing and there for anyone to see. I do not share Keenan’s views on several technical elements here, but I cannot see that as any justification for impugning his character.

34. Paul_K,
    First I have not impugned Keenan’s character, I have said that I am unable to believe him. The reason is that he has presented us with a headline that is clearly not true (“The Met Office has admitted…) and he has said that the Met did not answer questions that it is clear that they did, and Hirst says that they did. He apparently says this because he was dissatisfied with the answers. Why he does this I don’t know, but to me as a practical matter I can’t rely on what he says.

    The issue of not answering is not one of divulging facts; it is a reluctance of the Met Office to do a pet calculation of Keenan’s that they likely think is worthless. This does not come under the rules that you describe – it is an issue of resource management. Can Lords direct the Met how to allocate its time and tell it to do science it thinks is worthless?

    And then the final result – the Met buckles and does Keenan’s calc and agrees that C(3,1,0) gives a lower AIC criterion than C(1,0,0) with trend, and the gotcha – “Met Office admits claims of significant temperature rise untenable”. Do you really believe that is a correct characterisation of what the Met said?

35. Shub

    The question is: “What is the basis…?”

    I’ve interpreted it as starting with “What is the basis”. The specific words are “on what basis they assert that there has been a long-term upward trend”. You can’t just strip out that what they want the basis for asserting something. They gave that basis– and presumably the basis for asserting it in the context where the assert it. That context is descriptive.

36. PaulM

    You are writing as though the Met Office were somehow led to respond in this manner because of some cunningly contrived process of entrapment.

    No. I didn’t say they were entrapped.

    I’m saying they read the full 3 paragraphs and gave an answer to what those 3 paragraphs seemed to be asking– in context. But you are pulling out a bit and saying the full answer wouldn’t have been elicited by that bit. It wouldn’t have– because the parts of the answer that were elicited by the question come from other parts of the question.

    Anyway: I don’t think the Met crossed the line. I think that Parliamentary Q/A doesn’t have the appropriate structure for good communications on these things.

    There is no arm-twisting to make the Met Office use the term “statistically significant” in their response. They decided to walk into the quagmire of their own free will.

    I’ve never claimed arm twisting. I am saying that those statements were represented as “detecting a signal” or “proof of warming” but merely as the basis for communicating “upward” or “downward”. It is a descriptive statistic.
    Like it or not– descriptive statistics are not always big major claims of attribution. People can say “Our standard method for communicating with reporters is blah, blah, blah”. And their answer is statistically significant based on that model. It’s fine to say that.

    But it is now being treated as if they made an attributional claim. They weren’t. They weren’t asked to make a attributional claim.
    If Lord D wanted to know whether they thought the Red Noise model was appropriate to positively prove AGW in the signal, he should have asked that. He didn’t.

37. I repeat what I have said before – I think that Doug McNeall was perfectly correct when he stated that there is no valid answer to the question “is the temperature rise statistically significant?”
    You appear to be suggesting that there is a valid answer or am I misinterpreting your stance?

    You are misinterpreting. My stance is that back in Oct/Nov, they were not asked “is the temperature rise statistically significant?”

38. PaulK

    a) what specific question do you believe was being tested in this significance test applied by the Met Office, given that it was already asked and answered that temperatures were about 0.8 degrees higher now than in 1880

    The specific question is “Do we describe this upward, downward or sideways when having casual not-very-technical discussions?”. They were asked to explain their basis for saying “upward” “downward”. I see this as more like being asked to explain what they consider the “strike zone” in baseball.

    They gave that basis. The basis happens to be a simple “turn the crank” test and the description happens to involve a cut off point. Describing the cut-off point uses the word “statistical significance” using that particular model.

    But this was not an attributional claim.

    (b) what do you think was the null hypothesis in this test if not covered in

    There is no real “hypothesis”. This was like calls in baseball.

    (a)
    (c) what is the structural justification for the choice of the AR(1) model in this test if not covered in (a) and (b)?

    The justifications are:
    1) We do know there is strong correlation so we aren’t going to use least squares.
    2) This is going to be a simple-rule-of-thumb baseball strikezone like call.
    3) AR(1) is the simplest model, discussed in many text books. People aren’t going to think we are doing anything sophisticated or making any attributional claim.
    4) Other models could be unphysical and there is no good way to pick one.
    5) We can apply this simple model quickly under generic circumstances because we aren’t making any attributional claim.

39. Paul_K

    The threatening letter from Lord Donahue (which you describe as badgering) was because he posed the same question (about relative likelihood between the models) multiple times and did not get an answer.

    No. It was because he got a perfectly good answer to the question he actually asked and either
    (a) didn’t recognize it as an answer to the question (because he doesn’t understand physics).
    (b) didn’t know how to word his question to get the answer he wanted (because he doesn’t understand physics or statistics.) or
    (c) was badgering.
    There could be other options. I lean toward (a) or (b)

40. I am having real trouble following any of this, aside from the obvious conclusion (which was my first impression anyhow) of Not Much To See Here. Who cares what exactly the Met said or meant? It’s a soundbites world.

    But I am wondering how this relates, if at all, to the much more subtle kerfuffle a little while ago when a pseudonymic statistician began commenting (at SkS?) that the temperature record cannot be proven to be anything but a random walk, and various commenters trying to follow his reasoning and/or answer him. Or something like that. Does anyone remember this? Could they describe the issues there correctly? Is it the same issue?

41. MikeR–

    Who cares what exactly the Met said or meant? It’s a soundbites world.

    Well… yes. This story is, at it’s origin, about Keenan trying to make some sort of “proof” out of a rather confused sound bite.

    Does anyone remember this?

    That statistician was anonymous. But the d=1 issue is related to the “unit root” issue.
    http://rankexploits.com/musings/2010/joshs-unit-root-cartoon-haiku/
    

42. MikeR
    You may mean VS in this thread at Bart Verheggen’s. Discussing a paper on polynomial cointegration which was recently published in controversial circumstances. But VS said that it wasn’t a random walk.

43. MarkR–
    From that thread, here is eduoardo Zorita had to say about I(1) (which corresponds to d=1 in ARIMA)

    http://ourchangingclimate.wordpress.com/2010/03/01/global-average-temperature-increase-giss-hadcru-and-ncdc-compared/#comment-2703

    Fair enough, but an additional important question is how these random fluctuations can be described. VS is arguing that these fluctuations must (or should) be a I(1) process. However, a I(1) process cannot be physically justified, since it is not stationary, and its variance would grow unbounded over time. This is not observed. Independently of the period of Earth history one chooses to look at last (thousand years, last ten thousand, last one million, it fluctuates, but within bounds. The variance of the temperature is always bounded. I would tend to give much more weight this observations than to any statistical test performed on a limited noisy sample. I would agree that the structure of the those random fluctuations could be complex, much more than it is thought, but it should be a stationary process. Other people have suggested a fractional-differenced processes. But then the recent trend becomes unusual, as far as I understood. So we have a conundrum here.

44. Re: Nick Stokes (May 30 07:34),

    One definition of a random walk is the integral of white noise over time. If VS did indeed say that an I(1) process wasn’t a random walk, he was (charitably) mistaken. But then he was also mistaken about B&R’s paper being a valid approach.

45. Shub:

    A second thing. In the whole of Keenan’s main point, there is nothing about ‘anthropogenic’ in there. It is about detecting the ‘global warming’ signal.

    You realize you can’t do attribution just from a temperature series, don’t you?

    I made the statement on another thread that there is (provably and almost trivially so) an infinite class of statistical models that can fit a given series of data.

    In order to weed this down to something usable, you have to apply the filter of “physically realizable”.

    What Keenan has done is flawed for two reasons, not recognizing that an uninterpretable statistical model has no physical validity and not recognizing that a fit to a model by itself tells you nothing about attribution.

    These things are so trivially basic it makes me wonder about his general level of aptitude.

46. DeWitt,
    I’m not sure where VS said it in Bart’s thread and it’s at odds with what Zorita said, but I was guided by this comment at Moyhu.

47. Nick, technically the integral of white noise is Brownian noise.

    Wiki it.

    VS is another one we will have to say hm…. about.

48. Nick–
    At Barts VS wrote

    Full text here:
    http://ourchangingclimate.wordpress.com/2010/03/01/global-average-temperature-increase-giss-hadcru-and-ncdc-compared/#comment-1216

    In other words, global temperature contains a stochastic rather than deterministic trend, and is statistically speaking, a random walk. Simply calculating OLS trends and claiming that there is a ‘clear increase’ is non-sense (non-science). According to what we observe therefore, temperatures might either increase or decrease in the following year (so no ‘trend’).

    I guess at your blog he wrote

    That random walk thing is nonsense (it simply served as a ‘strawman’), and in fact, I have statistically rejected the null-hypothesis of a random walk, in the proper thread.

    I guess it’s not too surprising people aren’t sure what he was claiming.

49. Carrick,
    Just for the record, neither Shub nor Keenan has suggested that the temperature record can be used for AGW attribution. And some time ago, I wrote an article explaining why it could not be used. Shub has tried to point out several times here and elsewhere that the issue is not about AGW attribution but about whether the temperature rise can be said to be “statistically significant”. The only context for me that makes any sense for such a statement is that the temperature rise is sui generis out of the ordinary i.e. that it represents a statistically meaningful departure from what is expected. In order to make such a statement, you need a credible (null) model to describe what is expected.

    My own position, which seems to be shared by Doug McNeall of the Met Office, is that such a credible null model does not exist as yet. Hence there is no science basis on which to state that the temperature rise is statistically significant. All you can say is that the temperature appears to be about 0.8 degrees higher now than it used to be in 1880. AGW attribution is a separate question, which so far has only been raised by Lucia as far as I can see.

50. Paul_K, are you really trying to argue that we can’t infer from this temperature data set that it has warmed since 1880?

    That doesn’t seem like a credible position.

    (Ah never mind, you did say “out of the ordinary.” Teach me to respond in the middle of a game of email soccer.)

51. Lucia:

    I guess it’s not too surprising people aren’t sure what he was claiming.

    Looks to me like he’s trying to walk away from some earlier claims. Or he’s senile like me, and gets to argue a new set of claims each day. 😉

52. Paul_K

    Shub has tried to point out several times here and elsewhere that the issue is not about AGW attribution

    Notwithstanding whatever point Shub tried to make about “the issue is not about AGW attribution” Keenan wrote this at BishopHIll

    To conclude, the primary basis for global-warming alarmism is unfounded. The Met Office has been making false claims about the significance of climatic changes to Parliament—as well as to the government, the media, and others — claims which have seriously affected both policies and opinions. When questioned about those claims in Parliament, the Met Office did everything feasible to avoid telling the truth.

    This claim by Keenan is a claim about attribution. Specifically: He wrote “the primary basis for global-warming alarmism is unfounded”. That is an attributional claim about the signal of global-warming in the temperature series analyzed. Specifically, it is a claim that we cannot attribute the temperature rise to global warming.

    Moreover: Keenan’s claim is simply wrong. A time series fit to temperature data since 1880 is not “the primary basis for global-warming alarmism”.

    So you may not think one can either prove or disprove an attributional claim based on this time series analysis and Shub may not think so, but Keenan sure as shootin’ made one. And people at BishopHill are understanding it that way because that’s the way Keenan wrote it.

53. Carrick,

    Paul_K, are you really trying to argue that we can’t infer from this temperature data set that it has warmed since 1880?

    mm. I read him as saying we can’t infer why it’s warmed, or perhaps that we can’t infer that there’s anything to it beyond natural variation.

    But I could be wrong, often am.

54. Carrick, no, Paul_K is not saying that.

55. Re:lucia (Comment #114279)
    May 30th, 2013 at 5:51 am

    Hi Lucia,
    The Met Office was not offering some loose language to reporters. They were responding formally to a parliamentary written question. This enters Hansard, and it is citable in parliamentary circles as a formal reference. I assure you that we will see the appearance of comments like “On the best scientific advice available, the rise in temperature from 1880 to the present day is statistically significant.[HL3050]” in parliamentary debate and in DECC sub-committees.
    And if you are correct in your assertion that the Met Office was actually talking about a baseball game rather than a formal hypothesis test, when they introduced the term “statistically significant”, then I think one problem is that this qualification will unfortunately be lost on UK parliamentarians, most of whom will have played cricket in their youth.

56. Paul_K:

    The only context for me that makes any sense for such a statement is that the temperature rise is sui generis out of the ordinary i.e. that it represents a statistically meaningful departure from what is expected.

    Having red-stamped my own previous comment, I agree we can’t tell from a limited data set whether it is “out of the ordinary” or not, but that’s not the basis by which you assign attribution.

    The warming from circa 1910-1950 is generally accepted as being dominantly natural (or putting it more technically, “you don’t need to invoke anthropogenic forcings to explain the temperature rise”), and nearly of the same magnitude as the warming from circa 1970-2010.

    I think Phil Jones suggested that they were statistically indistinguishable–sounds like a nice study for Lucia. 😉

    Attribution comes from first eliminating the impossible, and assigning what is left to be the cause. If you can explain the origin of the warming from 1910-1950 to natural causes, but not the warming from 1970-2010, then you could still attribute the warming from 1970-2010 to anthropogenic forcings.

    For the record I don’t personally think you can do that unambiguously, even though that is what the IPCC claims can be done.

    But whether attribution can or can’t be made unambiguiously, I don’t agree that you should look at whether a temperature variation is “ordinary” or not in order to eliminate a potentially novel cause.

    If you don’t know the forcings very well, or the model that predicts the climate response is too unreliable, then even if you saw a temperature response that was “out of the ordinary” that would be classified as a “wtf is that?” not an “agw!”

57. Mark & Paul M— I agree that Paul_K wasn’t saying that (see my edited comment).

58. Carrick,

    I suspected I was making a pointless comment the instant I hit submit, but pointless comments are working out so well for me why change a winning formula. 🙂

59. Carrick,

    The warming from circa 1910-1950 is generally accepted as being dominantly natural (or putting it more technically, “you don’t need to invoke anthropogenic forcings to explain the temperature rise”), and nearly of the same magnitude as the warming from circa 1970-2010.

    If you consider all GHG forcings (CO2, N2O, methane, ozone, CFC’s), the forcing circa 1950 relative to pre-industrial (say, compared to 1800) was close to 1 watt/M^2 (versus ~3.15 watts/M^2 today). So it seems to me likely that some, or maybe even even most, of the warming pre-1950 was due to GHG forcing. Since it took much longer for the GHG forcing to rise from zero to ~1 watt/M^2 (about 150 years) we could reasonably expect to see a larger fraction of GHG driven warming from that increase in forcing than we have seen from the more rapid increase in GHG forcing since 1950.

60. SteveF, you can’t just consider GHG forcings though, while neglecting other anthropogenic forcings though, nor ignore the natural forcings.

    Climate is going to respond to the sum of all forcings, not just to one “favored son”.

    Ideally you’d produce a curve like this:

    GISS Model E (Anthropogenic) Radiative Forcings, but include the error bars…

61. Paul_K

    Hi Lucia,
    The Met Office was not offering some loose language to reporters.

    I didn’t say they were.

    And if you are correct in your assertion that the Met Office was actually talking about a baseball game rather than a formal hypothesis test,

    I was using an analogy to explain this was more like explaining how they make “a call”. The question was more like asking “how do you make your call”.

    But yes, I think they were not providing a formal hypothesis test. Moreover, it’s not clear to me that the room full of Parliamentarians took them to be providing a formal hypothesis test. Has the House of Lords weighed in on what they thought the question and answer meant? If yes, we can look to that.

    What LordD thought, I don’t know. We do know that at some point, LordD was having side conversations with Doug Keenan. But if the purpose was to ask the Met office to discuss statistical significance, Doug could have suggested more useful questions to Lord D. Because the ones LordD did ask were not suited to discussing how one might assess statistical significance.

62. Hi lucia
    I commented earlier. I think the blog software chewed it up. Please see if you received it and help
    .
    Thanks

63. Lucia:

    Because the ones LordD did ask were not suited to discussing how one might assess statistical significance

    Based on Paul_K’s read, one might say instead “assess its statistical relevance” (which is a real term of art). It appears nobody is arguing over whether the data show it to be warming, the issue is rather “what that means, if anything”.

64. Carrick,
    .
    The problem with that graphic, aside from the lack of error bars, is that the historical aerosol effects (direct and indirect) are so uncertain as to be little better than an arm wave…. and are selected mostly to make the GISS model E look ‘right’. Indeed, the AR5 SOD suggests the best current estimate for the total of direct and indirect aerosol effects is in the range of -0.7 watt/M^2 (IIRC), while the GISS graphic says more like -2.5 watts… more than a bit of a discrepancy I’d say.
    .
    The uncertainty for black carbon is much the same… just an arm wave. At least there is pretty solid data for GHG’s. BTW, the GISS graphic starts in 1880; ice core records show that CO2 had already been rising for a long time by then.

65. shub (Comment #114319)
    Sorry. Whatever happened, it is totally gone.

66. SteveF, I realize that there’s big uncertainty, but that’s my point. If
    you could eliminate the other forcings, you would be justified in neglecting them. Because they have large uncertainties, they represent a big confound. You also have to be able to explain the entire history of the temperature increase, which includes 1950-1970 in addition to the other two periods I mentioned. I think it is this which drives larger aerosols. (Invoking a larger for the AMO may be an alternative though.)

    I realize Hansen’s model gets much of its sensitivity to GHG by “diluting” the ~ 3 W/m2 GHG forcings down to 1 W/m2. (This effectively triples the sensitivity of GHGs.)

    That said, if you look at Hansen’s GHG forcings only, here’s how they vary over time (see link):

    1910: 0.26 W/m2
    1950: 0.67 W/m2

    1950 – 1910: 0.41 W/m2

    1970: 1.21 W/m2
    2010: 2.99 W/m2

    2010-1970: 1.78 W/m2

    If we assume all of the warming from 1970 to 2010 was from GHGs, where the slope was about 0.17°C/decade, it would be hard to explain how you got 0.11°C/decade warming for 1910-1950 from less than 25% of the same forcing.

67. Lucia,
    (Sorry for the TRIPLE posting. Can you delete the posting under “P” and the previous attempt which lost the blockquotes, making it unintelligible. Definitely senility creeping in.)
    You wrote
    “But if the purpose was to ask the Met office to discuss statistical significance, Doug could have suggested more useful questions to Lord D.”
    I wholeheartedly agree with this.
    But let me go back to your previous comment, because I am starting to get an inkling of why we are interpreting things so differently.
    In your previous comment you quite forcefully argued that Keenan was presenting an argument about attribution.

    This claim by Keenan is a claim about attribution. Specifically: He wrote “the primary basis for global-warming alarmism is unfounded”. That is an attributional claim about the signal of global-warming in the temperature series analyzed.

    I had not interpreted anything that Keenan had said to be about attribution. His original op-ed piece was about statistical characterisation of the temperature series. He was clearly upset by the IPCC claiming that the recent temperature rises were statistically significant. He never mentioned attribution even once.

    My translation of Keenan’s remark above was as follows. The primary basis for global warming alarmism is that the temperature rise observed in recent times is abnormal; i.e. it exhibits a statistically significant departure from the normal (expected) temperature movement; this primary basis is unfounded.

68. Re: SteveF (May 30 12:06),

    At least there is pretty solid data for GHG’s. BTW, the GISS graphic starts in 1880; ice core records show that CO2 had already been rising for a long time by then.

    That depends on your definitions of long and rising. The 75 year smoothed Law Dome record says the concentration in 1880 was 291.8 ppmv. The average from 1010-1600 was 281.9 ppmv. It was 281.6 in 1790. If you ignore serial autocorrelation, which will underestimate the variability, it doesn’t get to two standard deviations from the mean until about 1850.

    The reason I cut the average off at 1600 is that it looks very much like the Little Ice Age caused a dip from about 1580-1790. Including that would lower the average but increase the variability. I’m betting that a proper analysis would show that CO2 didn’t start going outside the pre-industrial range until sometime in the 19th century, especially if you use log(CO2). That, IMO, is not “long before” 1880.

69. Re: Carrick (May 30 08:21),

    Nick, technically the integral of white noise is Brownian noise.

    Wiki it.

    VS is another one we will have to say hm…. about.

    And right there in the first sentence:

    hence its alternative name of random walk noise.[emphasis original]

70. Carrick,
    OK, I assumed 23% of the historical GHG forcing was offset by aerosols (similar to what AR5 SOD says about the present), and cooked up the following: http://i42.tinypic.com/33olf8p.png
    .
    I don’t have to squint too hard to imagine I see a ~60 year cycle superimposed on a secular trend which tracks GHG forcing pretty well. Most of the short term (few years) variation correlates with ENSO and large volcanoes. Could it be coincidence? Sure, but I don’t discount the possibility that the simplest explanation is the right one: there has been considerable net forcing all along and relatively low climate sensitivity.

71. DeWitt,

    I was considering N2O and methane as well, and IIRC, those do show a bit more fractional rate of rise than CO2 pre-1900.

72. SteveF, there are plenty of coincidences, and with noisy data like this, it’s easy to get one.

    I think you have to stick to credible physical models, and the idea that aerosol particles don’t affect climate isn’t credible. We have plenty of data to the contrary.

73. Anyway, how well does your model actually do?

    Actually not very well. Maybe even “terrible.”

    The only part it gets right is the part of the signal that is monotonically changing, and almost any quantity that monotonically changing will do as well.

    Trouble is there isn’t a separate data set that you can invoke to remove the AMO, you just have to assign all of the unexplained low-frequency variance not explained by your particular monotonically changing variable to that AMO.

    That hardly makes your model “the simplest one”.

74. PaulK

    The primary basis for global warming alarmism is that the temperature rise observed in recent times is abnormal; i.e. it exhibits a statistically significant departure from the normal (expected) temperature movement;

    If I take the first part of your sentence
    “The primary basis for global warming alarmism is that the temperature rise observed in recent times is abnormal;”

    That is equivalent to: “The primary component of attribution studies in ‘global warming alarmism’ is the rise in temperature in recent times relative to the rate that is considered normal”. This would be a claim about how those in the ‘global warming alarmism’ camp do attribution.

    The second ‘i.e.’ part ‘i.e. it exhibits a statistically significant departure from the normal (expected) temperature movement;’

    The example gets more specific and then claims that if the temperature trend exhibits a statistically significant departure from normal, that is used to deem the changes abnormal. Along with the first: that would mean that if the temperature change is abnormal, the changes are attributed to whatever “global warming alarmism” attributes changes that are abnormal. That would be GHGs– but whatever they might have picked, that would be an attribution claim.

    So: your paragraph appears to be a discussion describing how Global Warming Alarmism does attribution.

    He never mentioned attribution even once

    He didn’t use the word “attribution”. But his paragraphs convey the notion that the trend analysis on a time series since 1880 is used to by ‘global warming alarmi[sts]’ to diagnose whether warming is unusual and outside the range of normal and more over, it conveys the notion that this is the primary method. ‘Diagnosis’, ‘attribution’, ‘identifying a cause’,’ provide the basis for believing ‘… those are all different ways of wording “attribution”.

75. DeWitt,

    Page 9, Chapter 1 here: http://www.climate.unibe.ch/~spahni/papers/spahni06phd.pdf shows a graphic summarizing a host of ice core studies.

76. Carrick,
    “I think you have to stick to credible physical models, and the idea that aerosol particles don’t affect climate isn’t credible. ”
    .
    Of course it is not credible. I said I assumed that the influence of man-made aerosols has been about 23% of historical GHG forcing. What is not credible to me is a historical forcing which is clearly in conflict with the current best estimate of present day forcing (that is, about 23% of current GHG forcing according to AR5 SOD). It is the contrived nature of the GISS aerosol history that I object to, not the suggestion that aerosols reflect solar energy to space.
    .
    “Trouble is there isn’t a separate data set that you can invoke to remove the AMO,”
    Sure there is, the AMO index would do quite well. So would the long term lagging average of the PDO index that I wrote a post about some time ago. But I wasn’t suggesting a specific cause for an apparent oscillation, only noting that an oscillation near 60 years would appear to fit the temperature data reasonably well.

77. Re: SteveF (May 30 14:31),

    It still looks to me like CO2 doesn’t change significantly from the baseline until about the middle of the 19th century, especially if you convert to forcing.

78. SteveF:

    But I wasn’t suggesting a specific cause for an apparent oscillation, only noting that an oscillation near 60 years would appear to fit the temperature data reasonably well

    But that’s the problem …getting a monotonic behavior is easy and you’ve basically punted on the hard part. You’ve essentially included an infinite degree model to eat up the part of the variance your model can’t explain.

    You may not like the way they are explaining the observed variability with their model, but they do recognize the need that it should be explained, or Lucy we’ve got a problem.

79. DeWitt,

    I had the combined forcing at only about 0.1 Watt/M^2 in 1850, and zero in 1800. A large majority of the increase does indeed take place after 1850.

80. SteveF, if I were to characterize it, there is a large amount of uncertainty in the forcings, other than GHGs, so there needs to be big fat error bars in any estimate of ECS associated with that.

    Secondly, there are unanswered questions about natural variability (e.g.,AMO). Including a putative AMO does reduce the amount of variance in the system and also reduces how much warming you observe.

    see e.g. this.

    I did a least squares fit the a ~ 60 cycle, and we’re looking the residual. Most of this variability is explained by a nonlinear monotonic trend + ENSO… it pretty much agrees with your model.

    Moreover, it reduces the warming seen from 1970-2010 from 0.17°C/decade down to around 0.06°C/decade. If we combine that with your aerosol model, you aren’t even in lukewarmer ECS range anymore. I’ll let you do the math yourself. 😉

    So what’s not to “like” about this model? From my perspective, we don’t have any external data set to fit the “AMO” and there are too few periods for extracting it from the data set directly to be robust.

    In other words, It may just be a coincidence too. It’s a plausibility argument that it’s even there. We can’t discount that it’s present, so it just adds to the uncertainty (stretching the uncertainty distribution down towards lower ECS numbers).

    What I object to with Hansen is he consistently seems to pick the combination that yields the largest ECS. He might be right, but it’s easy to see why his models seem skewed relative to more recent ECS estimates.

    It’s also interesting watching the motivated CAGW group defend not including aerosol forcings (they like your curve too as “proof” of CAGW). Since I know what taking the aerosols out of the total forcings does, it is a source of amusement for me.

81. Carrick,

    You may not like the way they are explaining the observed variability with their model, but they do recognize the need that it should be explained

    “May not like” is a bit of an understatement. What they are doing is using aerosols as a kludge to make the model fit the historical temperature data… a kludge like that always makes things look swell. Of course, they never admit it is a kludge. I would like it a lot more if they just said, “We have no idea if our model is right, since we had to use an arbitrary aerosol adjustment, without a shred of supporting data, to make the model hindcast fit the historical data. Other modeling groups don’t use that adjustment… they use other aerosol adjustments.”
    .
    I think we have had this conversation more than once! :-0

82. SteveF, in my parlance, “they don’t do enough to emphasize the uncertainty in their numbers.” 😉

    It’s an interesting thing to revisit, because neither of us is stuck in the mud, so there is some evolution in thought that occurs over time.

83. By the way,

    Here’s a figure of HadCrut + (HadCrut-60yr).

    The result is pretty remarkable.

84. Carrick,
    ” From my perspective, we don’t have any external data set to fit the “AMO” and there are too few periods for extracting it from the data set directly to be robust.”
    The NOAA ESRL has an AMO series here. It’s in the data base for the climate plotter where you can do the regression. Here is a snapshot of Hadcrut4 regressed against 1+t+AMO. It’s the dark curve (also labelled Hadcrut4). Pretty good fit.

85. Nick, Thanks…

    But I’m looking for an index like the ENSO MEI that doesn’t exclusively use atmospheric or ocean surface temperature… do you know of one?

    The reason is we have very limited amount of data for a period that look, and it is very easy to “data mine” with so few periods.

    To illustrate my concern more concretely, models like Hansen’s suggest an increase in temperature till about 1950, then a flattening off. There are reasons to expect this to happen that don’t depend (post facto at least) on the observation of the temperature drop.

    Since this happens to be more or less in phase with the AMO, an OLS fit using the temperature series to extract the AMO amplitude & phase is going to pull part of this (putative) anthropogenic-in-origin warming phase into the AMO amplitude.

    That’s especially true because in practice I let the frequency be fit to as well as the amplitude. Anyway, it would be clearly a case of circular logic were I to apply an OLS filter that can chew up part of the anthropogenic signal (the portion associated with the interplay between GHGs and aerosols), and having done so, say “look! There’s no anthropogenic signal here!”

86. re:lucia (Comment #114344)
    May 30th, 2013 at 2:31 pm

    OK Lucia,
    Well at least we have found the reason we are talking past each other. Let me define two separate questions:
    a) Is there anything abnormal or unprecedented in the instrumental temperature series which we can test using inferential statistics that tells us that we have an unusual problem with global warming (NB with global warming not with AGW)?
    b) Irrespective of the answer to (a), to what (sundry) causes do we attribute the observed warming over the period of the instrumental temperature series?

    From your last response, it seems that you are combining these two questions into a single question of attribution – which you are of course entitled to do if you wish. However, I do not believe that Keenan is combining the two questions, and for my part I can say that I am most definitely not. My comments on this subject have been focused entirely on the first question.

    The correct answer to the first question is that we cannot apply inferential statistics to the instrumental series because as yet we do not have sufficient information to formulate a credible null hypothesis about what constitutes “normal” temperature fluctuations, or if you prefer, natural variation in the long term series. Both Keenan and Doug McNeall agree on this.

    I therefore think that it is scientifically irresponsible to declare that the warming from 1880 onwards is “statistically significant” based on a mis-specified model tested against a null model of a flat long-term temperature with AR(1) noise.

    I am also unimpressed by the defense that this is some readily pardonable, lightweight use of descriptive statistics to help explain the view that the world has warmed up since 1880. You don’t need an invalid model fit to help you reach that conclusion; a temperature graph with error estimates will do. On the other hand, this misapplication of statistics causes a lot of erroneous citations, and outright mis-statements by people who don’t understand the lack of validity of the test, or who perhaps willfully choose to ignore its limitations in order to be able to argue that inferential statistics have shown that global warming is “extremely unlikely” to be explained by natural variation.

    If there is an unusual signal in the temperature series (and I believe there should be some AGW signal in there), it is too well masked by our own ignorance of natural variation for us to identify it in the temperature series using inferential statistics.

87. Carrick,
    re AMO, yes, I see what you mean. AFAIK, AMO is only defined in terms of SST. NOAA has a FAQ here.

88. Paul_K,
    “we cannot apply inferential statistics to the instrumental series because as yet we do not have sufficient information to formulate a credible null hypothesis about what constitutes “normal” temperature fluctuations

    When do you think we will have sufficient information to apply inferential statistics? Will someone ring a bell?

    The MO gave the best answer they could to the question as asked – do you consider a rise of 0.8°C significant? They interpreted this to be asking for statistical significance, which since it was a repeated tequest for “significant” seems reasonable. And they said yes, se 0.13°C, and gave their model.

    Now Keenan later says there is an admission because he has a more complex model which fits better and might reduce the error range. But even with an imperfect linear, the statistic requested was highly significant.

    You might say that non-random effects should be identified and eliminated. But Keenan’s analysis has no bearing on that.

89. Paul_K
    I don’t know why you are asking me my answers to (a) and (b). My answer to (a) is yes provide we do no limit ourselves to the bag of tricks Keenan does. Can we get the answer exactly? no. But we can use statisitics, do statistical tests and make inferences. We could infer the temperature rise is unusual in several ways. To (b) well, that would depend, wouldn’t it? Different people can pick different things. I pick Leprechauns. 🙂

    However, I do not believe that Keenan is combining the two questions, and for my part I can say that I am most definitely not.

    I’m not understanding how talking about the ability to separate clarifies anything. I think Keenan said “The primary basis for global warming alarmism is that the temperature rise observed in recent times is abnormal” he is saying they make an attributional claim. Since “global warming alarmists” don’t attribute the temperature rise to Leprechauns, we can infer they are attributing it to “AGW”. I don’t think there is any other way to read this. And when he follows that with ““the primary basis for global-warming alarmism is unfounded””, he is specifically making a de-attribution of their claim. That means: He is discussing his analsis in light of attribution and trying to ‘de-attribute’ based on the statistical analysis. His words communicate an attributional claim is true even if one could do the analysis, and then not make any sort of claim leaving open the possibility that a trend– if found– could have been caused by Leprechauns, rebound from the little ice age or so on. But he didn’t leave it open– he specifically proclaimed a de-attribution of “global warming alarmism”.

    Beyond that: If all he had wanted to say was that you can’t do attribution that way– he could have said that. Moreover, it appears Doug Keenan tells us

    Your message also says this.

    If we are asked if there has been a statistically significant change in temperatures since the 1880s, we do not say “yes” or “no”, we say “that is not a valid question”.

    That surely is a valid question! There are other situations where the data is well enough understood to select a model, albeit not with absolute certainty. What we should say in this situation is “we are currently unable to determine if the increase is statistically significant”.

    So he is violently objecting to McNeall’s suggesting that asking about statitistical significance is not valid.

    I am also unimpressed by the defense that this is some readily pardonable, lightweight use of descriptive statistics to help explain the view that the world has warmed up since 1880.

    But you keep changing what I said. I didn’t say a “lgihtweight”. And I didn’t say “explain the view that the world has warmed up since 1880”.

    I said they were asked for their basis for decreeing the record “shows a long-term upward trend”: that is they were answering the question that was asked. They weren’t asked to explain “the view that the world has warmed up since 1880. ” For the latter, they would only need to say the temperature was higher now than in 1880. But for the former, they have to state a trend.

    They happened to give a descriptive statistic. Can this confuse? Sure. But that doesn’t prevent it from being used as a descriptive statistic.

    it is too well masked by our own ignorance of natural variation for us to identify it in the temperature series using inferential statistics.

    Is it? I don’t agree. Zorita did an analysis of the likely hood of the final points all being warmest. That’s inferential statistics. We can also draw inferences estimating variability from models. It might not be perfect– but that’s still inferential statistics. Just because we can’t do the problem by fitting a straight line and using ARIMA doesn’t mean we can’t do inferential statistics.

90. Lucia,
    I hate disagreeing with you (genuinely), but I think we are going to have to stay apart intellectually on this.

    Incidentally, Eduardo’s paper makes some honest qualifications related to the importance of the null model in arriving at his conclusions.

    Just for fun, if I can find the time, I will show you what statistical significance looks like when the modern temperature series is tested against the null of an Occam’s Razor Oscillatory Model.

91. Nick,
    “When do you think we will have sufficient information to apply inferential statistics? Will someone ring a bell?”

    I don’t know about someone ringing a bell, but when we have sufficient information, it will pass the “elephant test”.

    When I find some time, I will show you what a significance test on the temperature series looks like when it is modeled using the Occam’s Razor Oscillatory Model. It gives a structurally perfect match to the instrumental temperature series. You will then claim (correctly) that I don’t have sufficient information to support that null model, but that is because the result showing that the temperature series falls within perfectly normal limits does not fit your preconceptions.
    As long as the result does fit your preconceptions, you are content to accept a flawed null and a mis-specified statistical model fit.

92. Nick writes “When do you think we will have sufficient information to apply inferential statistics? Will someone ring a bell?”

    Hilarious Nick. Slightly off topic but I’m now wondering whether a bell will be involved when the millions are supposed to move away from the approaching creeping doom that is the sea. I imagine great lines of refugees with their possessions on donkeys and herding chickens and goats.

93. Paul_K

    Incidentally, Eduardo’s paper makes some honest qualifications related to the importance of the null model in arriving at his conclusions.

    The null chosen is always important in all cases where one does inferential statistics. That doesn’t preclude using statistics to support making inferences.

94. TimTheToolMan
    In all seriousness TimTheToolMan: How does Paul diagnose the point when we can make inferences and when we can’t?
    I’m seeing “Incidentally, Eduardo’s paper makes some honest qualifications related to the importance of the null model in arriving at his conclusions” and “Just for fun, if I can find the time, I will show you what statistical significance looks like when the modern temperature series is tested against the null of an Occam’s Razor Oscillatory Model.”

    But of course choice of nulls make a difference in what we infer. This is why
    (a) our null is always an assumption which we judge based on factors extraneous to statistical arguments.
    (b) the argument about whether d=1 is precluded by physics matters and cannot be resolved by doing a statistical test on the data itself. It needs to be resolved by physical reasoning and likely supplemented by statistical treatments of pseudo-data to see what artifacts popup out when you stuff data generated using various different processes into ‘statistics black box’.

    If you set up rules such that a null can never be rejected when it is wrong in the way that one thinks the null is wrong, then the statistics test is useless.

95. The Met Office has just posted a response:

    http://www.metoffice.gov.uk/research/news/statistical-models-and-temperature

96. An interesting conversation. I realize my understanding of the larger argument is limited by my naivety with respect to the statistical programs at issue. Perhaps some of you can enlighten me.

    1. I quite understand why the linear trend with 1st order autocorrelation statistical model for the T anomaly time series should not be expected to describe reality very well; if the forcings presumed responsible for T change do not vary linearly with time then neither should T.

    Lucia has argued supra, quite reasonably, that a useful statistical model ought to conform to known physical laws (at least I think that represents her position). So why did the IPCC put that statistical model out there in the first place? I assume one reason was so that the public might reasonably infer that the calculated linear rate of T change would continue into the future. At least many advocates have used the numbers in that way. And weren’t the statistical parameters of that model being used to imply “statistical significance” in the usual sense for the IPCC assertion of AGW?

97. Barry Elledge

    So why did the IPCC put that statistical model out there in the first place?

    I think it’s in an appendix caveated like heck. It’s not used for attribution. It’s really not used for anything.

    They like to have these “FAQ” sections that are supposed to “help”. I think the goal is to hand these to middle school teachers who aren’t going to think any more deeply than the pretty pictures.

    Quite likely, someone came up with this idea for a heuristic. (That is: an analysis that ‘points to’.) But heuristics can be very dangerous especially in statistics. If they’d written a section they called ‘heuristics’ and explicitly said “We don’t really use any of these”, that probably would have been wiser. They could have stuffed the stoooopid graph showing temperature trends “accelerating” in there too.

    to imply “statistical significance” in the usual sense for the IPCC assertion of AGW?

    Oh. Not necessarily even that.

98. 2. Keenan was apparently trying to show that a random-walk in T could account for the observed T rise as well as a trend could. Lucia has argued that a random-walk model is not consistent with physical principles. I don’t know enough about the model to appreciate why.

    My inference is that the argument goes like this: T varies with forcings; for a given magnitude of forcings, T should stabilize around an equilibrium value; a random walk to a new T independent of changes in forcings is equivalent to gaining (or losing) global energy without changing the energy inputs and outputs. Perhaps Lucia or another reader could verify if this is the basis for the argument.

    If this is a correct summary of the argument, I would reply that the random-walk is a way of accounting for our ignorance of all the factors that produce earth’s energy balance. To me this seems little different from introducing an autocorrelation term. Autocorrelation may seem reasonable because of lags in attaining an equilibrium response to a change in forcings; the thermal inertia of the oceans provides a rationale for allowing autocorrelation. However, the fitting of the autocorrelation parameters in the statistical model is done without recourse to any physical models about thermal inertia of the oceans (as far as I can tell anyway).

    Similarly a change in global T independent of changes in forcings is consistent with all the unknown factors influencing climate equilibrium T. “Forcings” are defined by the effects we know and understand well enough to measure and model. The effects we don’t understand and account for get treated as “noise” even though they in general represent real processes rather than measurement errors. Some of these “noisy” changes may involve moving the earth to a new meta-stable state; for example, a change in ocean circulation patterns. The changes between meta-stable states might be modeled pretty well by random walks.

    3. If you are willing to contemplate a random-walk process as consistent with thermodynamics, what would be an appropriate statistical model? I can appreciate that a model with 3 auto-correlation terms is a bit fancy. But what would be a better test of the random-walk possibility?

99. There is a big controversy in economics about stock prices and whether they are a random walk. It’s just like climate in that in principle stock prices should respond to “forcing” otherwise known as fundamentals.

100. TimTheToolMan (Comment #114409)
     “I’m now wondering whether a bell will be involved when the millions are supposed to move away from the approaching creeping doom that is the sea.”
     Quite possibly. Here is how it used to be done.

101. Re: Barry Elledge (May 31 16:00),

     If a driftless random walk model for temperature were actually correct, it would mean that it’s equally likely that the temperature will rise or fall in the future. In fact, it could fall in the next 130 years a lot more than it rose in the last 130 years.

     But in fact, an empirical statistical model is entirely useless for projecting the future. At best it puts bounds on the likely range of temperatures. But the bounds for an ARIMA(3,1,0) model are so wide as to be entirely uninformative.

     Consider what was done in the Unit Root fiasco. Atmospheric CO2 was claimed to be I(2). Again, that would mean that even with an underlying trend, it would still be somewhat likely that atmospheric CO2 would decrease rather than increase in the near future. That’s what happens when you treat an almost entirely deterministic process as if it were entirely random.

102. DeWitt Payne:
     “If a driftless random walk model for temperature were actually correct, it would mean that it’s equally likely that the temperature will rise or fall in the future.”
     This is only a quibble, but in the short term the above statement is not true; the AR(3) part of the model ensures some “momentum” in the random walk. But in the long term — and you mentioned a century-long period — you are correct.

103. lucia writes “If you set up rules such that a null can never be rejected when it is wrong in the way that one thinks the null is wrong, then the statistics test is useless.”

     I’m not a statistician…. Nevertheless it seems to me the question of the null is paramount. IMHO the null for this question of significance is that temperatures were at equilibrium at the start of the industrial revolution.

     If they weren’t then we absolutely need to know precisely why they weren’t.

     The problem with that definition of the null, of course, is that we have no firm temperature data to support that idea. Everything back then is a proxy.

     And there are plenty of examples of considerable regional differences. For example snow on the Thames or from earlier, artefacts under the permafrost in Greenland.

     So the question is can regions have different and sustained over the longer term climates?

     If not then they must be reflected by the global temperature and that destroys the idea of the null as I defined it above and hence the idea of significance with this particular temperature excursion.

     I’ve never personally looked into regional climatic studies.

104. *To Dewitt Payne at 5/31 4:50 pm:

     Thanks for your reply. I think your point is a good argument against random-walk as a descriptor of climate change over long periods. If a random walk were the only process, T would be expected to vary all over the lot on long time scales with no periodicity. Instead we observe numerous strongly periodic variations dominated by the circa 100,000 year glacial cycle, and constrained within limits.

     You state, “That’s what happens when you treat an almost entirely deterministic process as if it were random.” I largely agree with you. In my earlier comment I mention that a “random-walk is a way of accounting for our ignorance of all the factors that produce earth’s energy balance.” The stuff that isn’t considered in the usual forcing models may very well be just as deterministic as the stuff that is. Since it isn’t quantified and incorporated into calculations it ends up being treated as noise. Some of the stuff that isn’t well treated in the models includes low level clouds, ocean cycles like the PDO and AMO which are strongly periodic, and ocean cycles like ENSO which are recurring but have no obvious periodicity. I strongly suspect these are deterministic effects, although I haven’t seen anyone who claims to understand the factors which drive the AMO and PDO (correct me if I’m mistaken). Clouds by contrast seem to enjoy an embarrassment of putative explanations.

     However, I wonder if there are not a few longer-term climate phenomena which would appear like random walks even if they were better understood. I had earlier mentioned ocean circulations and currents as possible candidates. Currents seem to shift fairly abruptly and unpredictably. Once a new current becomes established, it tends to persist. This seems like a potentially chaotic process which could appear like a random walk.

105. Lucia: The Met Office has responded publicly to Keenan’s post at BH. See: http://www.bishop-hill.net/blog/2013/5/31/met-office-responds-to-keenan.html

     I was unable to find any passages in this new material that support your contention that the AR1 model was used purely for descriptive purposes and not for drawing inferences. A key passage says: “There is very high confidence (using the IPCC’s definition) that the global average net effect of human activities since 1850 has been one of warming. The basis for this claim is not, and never has been, the SOLE use of statistical models to emulate a global temperature trend.” Nowhere do they say (like Doug McNeal and most commenters here) that the statistical significance of the AR1 model shouldn’t be used to draw inferences about whether the observed warming could be due to natural variability. They seem to be implying that the AR1 model is a viable PART of the evidence that attributes warming to human activities.

106. Frank (#114619):
     “I was unable to find any passages in this new material that support your contention that the AR1 model was used purely for descriptive purposes and not for drawing inferences.”
     Look at the fuller response here. In the conclusion, they write: “As the Met Office makes its assessments of global climate change on a wider evidence base and performs detailed detection and attribution studies based firmly in the physics of the climate system, the comparisons in this paper have no bearing on Met Office statements on climate change.”
     .
     [It is amusing to note that the head of the Met Office apparently is unable to spell Doug McNeall’s name correctly.]

107. Harold: Thanks for prompting me to read more closely. The material they released is schizophrenic. They seem to want people to believe that some evidence of human causation can be inferred from the significance of some statistical models (because those model agrees with D&A studies?), but that other statistical models can’t be trusted.

     The Executive Summary to Slingo’s paper (and blog post) say: “It should be noted that the Met Office does not rely SOLELY on statistical models in its detection and attribution of climate change.”

     Page 8: “Thus, the Met Office does NOT use one of these statistical models to assess global temperature change in relation to natural variability. In fact, work undertaken at the Met Office on the detection of climate change in observational data is predominantly based on the application of formal detection and attribution methods.”

     Page 13: “Notably, the relative likelihoods of the two [ARIMA] models that do allow for drift (trend) versus the driftless model proposed in the Parliamentary Questions, do not provide evidence against the existence of a trend in the data.” Wouldn’t a normal null hypothesis be that there is no drift and the inference be that the observations can’t reject this null hypothesis? And who in their right mind would cite likelihoods for two ARIMA models as evidence of anything when they state elsewhere that ARIMA models aren’t consistent with the physics of the atmosphere.

     The attitude seems to be: The null hypothesis is that the consensus is correct. Since you can’t demonstrate (statistically or otherwise) that the null hypothesis is inconsistent with experiment, the consensus stands. The statistical significant of the linear AR1 model supports the conclusion attribution obtained from D&A studies, because Keenan can’t prove the AR1 model is wrong. Trenberth does the same thing when claiming that anthropogenic GHGs contribute to all weather disasters unless proven otherwise. The hockey stick is right because McIntyre can’t prove that improperly-centered PC’s produced a wrong answer. And AR5 will probably say that ECS is still 2-4.5 and mostly likely about 3 because recent observational studies are not statistically inconsistent with this estimate. (Perhaps some studies are inconsistent with 3.0, but a null hypothesis of 2.9 or 2.8 or 2.7 can’t be rejected and these values are “around 3”.)

108. The Met Office emphasizes attribution through D&A studies rather than the observed temperature record. Figure 9.5 of AR4 WG I compares model simulations with the temperature record and the discussion says: “The fact that climate models are only able to reproduce observed global mean temperature changes over the 20th century when they include anthropogenic forcings, and that they fail to do so when they exclude anthropogenic forcings, is evidence for the influence of humans on global climate.”

     From a statistical perspective, how does the IPCC conclude that the observed temperature record can’t be reproduced without anthropogenic forcings? The nature of the variability in the observed temperature record prevents us from statistically concluding that ANY trend is present. Doesn’t the same problem prevent us from rejecting the model without anthropogenic forcings? That model looks trendless.

     The model with anthropogenic forcings is clearly better. That doesn’t mean that I can automatically reject the model without those forcings. Can’t both models be viable?

109. Noticed this comment in the post about McKitrick’s conference:
     http://climateaudit.org/2013/06/03/econometric-applications-in-climatology/#comment-422116
     “Will VS be there?” – “Yes”
     Maybe someone can find out his exact point of view.

110. I see that there is a speaker named Vid Stimac.

Comments are closed.