Personal attacks on Met Office scientists

I’ve been trying to stay out of the climate wars, but I really can’t resist commenting on this. There’s an interesting Greenpeace article discussing the UK government’s frustration with climate “sceptic” pressure (I added the inverted commas, as they seem necessary here). The thing I wanted to comment on were the alleged personal attacks on Met Office scientists

“In one email which was particularly concerning, [Keenan] accused Professor Slingo of deliberately misleading Parliament. Having taken legal advice, Professor Slingo took the decision no longer to engage personally with Mr Keenan.”

Bernard Silverman, Chief Scientific Adviser for the Treasury, had previously warned Professor that “attacks on her were being promulgated in the paper authored by Mr Keenan that he has sent to DECC and has been circulating internationally and has also placed on his website.”

So, who is this Mr Keenan? If you want to read his own site, it is here. I’ve discussed the problems with his ideas here, here, and here. Richard Telford has numerous posts about Doug Keenan and his tactics. I’ve even tried to explain the issue to Doug Keenan directly, twice.

My good friend Andrew Montford discusses this issue without – as far as I can see – mentioning that Doug Keenan is alleged to have personally attacked Met Office scientists. Doug Keenan appears in the comments, apparently also oblivious to this claim. If you want to get some idea of Doug Keenan’s general mindset, you really can’t do better than this

… the best time-series analysts tend to be in finance. Time-series analysts in finance generally get paid 5–25 times as much as those in academia; so analysts in finance do naturally tend to be more skillful than those in academia—though there are exceptions.

The only polite way that I can describe Doug Keenan’s views is not even wrong. We need a better class of climate “skeptic”.

Advertisements
This entry was posted in Climate change, ClimateBall, Global warming, Science and tagged , , , , , , . Bookmark the permalink.

233 Responses to Personal attacks on Met Office scientists

  1. Are you suggesting I’m being judgemental 🙂

  2. Rachel M says:

    Not at all. This is what I felt when I read Doug Keenan’s quote about analysts in finance getting paid more than academics and therefore being more skilful. No words could describe my reaction so I posted a giphy instead 🙂

  3. Rachel M says:

    Maybe this one is better:

  4. It was very slow, so all I saw initially was a picture of Judge Judy. It eventually illustrated your point, very nicely.

  5. Rachel M says:

    Ah, ok. They should be animated gifs and they’re pretty meaningless without the animation.

  6. It was just very slow, initially. Seems fine now.

  7. I trust a clever person like Mr Keenan made a fortune by predicting the financial crash of 2007?

  8. FLwolverine says:

    Rachel, the Judge Judy response was perfect.

  9. Whatever else you may think of Doug Keenan, the best time series analysts are indeed in finance.

  10. Richard,
    Two things. Firstly, how do you know? If your argument is the same as Doug’s (they’re paid more, therefore they’re better) then that’s particularly silly. Admittedly, I have only one example of someone who has worked on time series analysis in the finance sector, and that’s Doug Keenan, and that’s not particularly strong evidence in support of your claim. Secondly, who cares? This isn’t really simply a time series analysis issue, which is why the main thing Doug Keenan is illustrating is his ignorance.

  11. lord sidcup says:

    The name rang a couple of bells. This Real Climate post from 2010:

    http://www.realclimate.org/index.php/archives/2010/02/the-guardian-disappoints/

    “Keenan has made a cottage industry of accusing people of fraud whenever someone writes a paper of which he disapproves. He has attempted to get the FBI to investigate Mike Mann, pursued a vendetta against a Queen’s University Belfast researcher, and has harassed a French graduate student with fraud accusations based on completely legitimate choices in data handling. More recently Keenan, who contacted Wigley after having seen the email mentioned in the Pearce story, came to realise that Wigley was not in agreement with his unjustified allegations of ‘fraud’. In response, Keenan replied (in an email dated Jan 10, 2010) that:

    .. this has encouraged me to check a few of your publications: some are so incompetent that they seem to be criminally negligent. 
 Sincerely, Doug

    And this:

    http://www.theguardian.com/environment/2010/apr/20/climate-sceptic-wins-data-victory

    I wonder if Keenan ever did anything with the tree ring data he obtained. Its seems unlikely.

    Wish I had friends in the Lords who are willing to pursue my personal obsessions for me. What else are Lords for?

  12. lord sidcup,
    Yes, he does have rather a nasty history. I think he’s also involved in this whole silly saga with Nic Lewis.

  13. @Wotts
    How do I know? My initial specialization was in time series (taught by none other than Bierens). I’ve taught the stuff, published a number of papers, and supervised master’s and PhD theses. Colleagues have contributed to the theory. I’ve been the editor of a journal that publishes a lot of time series analysis. I’ve read and reviewed time series analyses in finance, economics, political science, biology, epidemiology and geophysics. The most advanced time series analysts are in finance. Geophysics is several decades behind the curve.

  14. Richard,
    Personal anecdote? That’s convincing. Just in case it wasn’t obvious, my main reason for highlighting that particular quote was the bizarre notion that being paid more made you better at something.

    Geophysics is several decades behind the curve.

    FWIW, given what I’ve seen you say publicly, you appear pretty clueless about the physical sciences. Just a personal view, mind you.

    All of this is rather irrelevant, though. Whether or not the best time series analysts are in finance, Doug Keenan is largely clueless about this topic, his views are simply wrong, and he appears to be a particularly nasty piece of work. Not surprised that you’re implicitly defending him.

  15. Joshua says:

    ==> “Whatever else you may think of Doug Keenan…”

    Whatever else you may think of Richard repeatedly making really bad arguments in blog comments…he has very impressive hair.

  16. Whatever else you may think of Richard repeatedly making really bad arguments in blog comments

    I’m still interested in knowing if Richard is just pissing about, and knows he’s making really bad arguments. It’s really hard to believe that he doesn’t realise.

  17. Sam taylor says:

    Any discipline which uses the hodrick-prescott filter in anger (hello economics) is, to my mind, less than clueless when it comes to time series analysis. Of course, as an academic Richard’s salary is probably 5-10 times lower than that of someone working in finance, so of course his opinion in such matters wouldn’t be worth a bucket of spit.

    [Mod : as amusing as this part might be, I’m going to discourage personal characterisations, even if somewhat mild.]

  18. Joshua says:

    “==> “Whatever else you may think of Doug Keenan…”

    Apart from that, Mrs. Lincoln, how did you enjoy the play?

    ==> “I’m still interested in knowing if Richard is just pissing about,…”

    Smart, knowledgeable people can make really bad arguments when they are emotionally and ideologically identified with the subject they’re discussing – particularly if they don’t make an explicit effort to control for their biases and to engage in good faith.

  19. @Wotts
    The context is whether or not there is a statistically significant trend in the temperature record. That is a time series question if there ever was one.

    One feature of labour markets around the world is that people who are better at something, get paid more. Lionel Messi, for instance, earns more than Leroy Fer, while no football club in the world is prepared to pay me anything.

    If geoscientists were good at time series analysis, you would see a steady stream of geoscientists into the city, where someone straight out of a PhD can demand a six figure salary. We observe large numbers of physicists and mathematicians move into the city (and the occasional astronomer, mostly on the strength of your numerics), but hardly any geoscientists.

  20. @Sam T
    My point exactly. Did you read any of the Rahmstorff-Vermeer papers?

  21. Richard,

    The context is whether or not there is a statistically significant trend in the temperature record. That is a time series question if there ever was one.

    Yes, I know. It’s not a particularly complicated analysis. What you fail to acknowledge is that this is not the context in which Keenan is using the term “significant”.

    One feature of labour markets around the world is that people who are better at something, get paid more.

    Really? Seems like rather a simplistic analysis. Can you prove it?

    If geoscientists were good at time series analysis, you would see a steady stream of geoscientists into the city, where someone straight out of a PhD can demand a six figure salary. We observe large numbers of physicists and mathematicians move into the city (and the occasional astronomer, mostly on the strength of your numerics), but hardly any geoscientists.

    Could it be – just brainstorming here – that geoscientists have another industry that snaps them up. I don’t know, maybe the oil and gas industry?

    Come on, this is both silly and irrelevant. Doug Keenan is wrong. Whether or not the best time series analysts are in the finance sector is irrelevant. Also being an expert time series analyst doesn’t mean that you have any understanding of the underlying physical processes associated with climate science – as aptly illustrated by Doug Keenan himself.

  22. Sam taylor says:

    Of course, Richard appears to be ignoring the fact that the entire oil and gas industry relies upon a pretty serious amount of advanced time series analysis in order to get basically any exploration work done whatsoever. Unless he thinks that the science underlying a process like 3DSRME or seismic spectral broadening is somehow childs play.

  23. @Wotts
    Indeed. Geoscientists who want to make money leave academia for oil & gas, where salaries are a lot lower than in finance.

    Keenan’s understanding of the physics is irrelevant, as the context is a statistical one. Keenan made enough money as a statistician to retire at a young age.

  24. Richard,

    Keenan’s understanding of the physics is irrelevant, as the context is a statistical one. Keenan made enough money as a statistician to retire at a young age.

    No, it really isn’t. This is bizarre. You actually think Keenan has a point? That’s amazing. It’s nonsense. He doesn’t know what he is talking about. That he earned enough money to retire at a young age is irrelevant. He clearly does not understand this topic sufficiently well to comment as he does. I had assumed that the same wasn’t quite as true for you. It seems I might have been wrong.

    At least you never fail to entertain.

  25. Joshua says:

    ==> “One feature of labour markets around the world is that people who are better at something, get paid more. …

    As I said: Smart, knowledgeable people can make really bad arguments when they are emotionally and ideologically identified with the subject they’re discussing – particularly if they don’t make an explicit effort to control for their biases and to engage in good faith.

  26. Eli Rabett says:

    Richard ties this and the last post together when he says:

    The context is whether or not there is a statistically significant trend in the temperature record. That is a time series question if there ever was one.

    True, but is a physically bounded time series question, something that the Keenan’s of the world never take into account which is why they keep getting the right answer to the wrong question.

  27. I have to say that Richard’s comment on this and on the last thread have made me feel a bit more sympathetic towards the Global Warming Policy Foundation. If this is the kind of Academic Advice that they’re getting, then it’s no wonder that they keep promoting nonsense.

  28. Sam taylor says:

    Eli,

    Yes. The pure statistics view completely misses that the earth system has shown bounded complex system like bahviour over the last few millennia. Ice core data is instructive in this regard. It’s like me swapping a 12 sided dice for a 6 sided dice halfway through a board game and arguing that it’s not statistically significant when I roll a 12.

  29. Sam,
    It’s interesting that you mention dice. Doug Keenan’s illustration of the supposed issue with the Met Office analysis involves dice. Essentially he argues that if you have a time series of dice throws, you determine the significance of this time series by comparing it to what you might expect for a fair die. What he seems to fail to recognise is that what he’s comparing to is a model of a fair die. Since a fair die is random, you can model that as a random process and then use that to determine if your given time series is consistent with being from a fair die or not. Since our climate is not purely random, the analogy breaks down.

  30. Willard says:

    > One feature of labour markets around the world is that people who are better at something, get paid more.

    Are you suggesting that women are less good at doing almost everything, Richard?

  31. Andrew Dodds says:

    Actually, many geoscientists of my acquaintance – including myself – went into software development for the money. Finance may be better paid, but you have to live in London; this was in any case before the quants rose to prominence.

    In any case – finance seems to like expertise in context-light pure math; by definition geoscience needs a lot of context. Time series analysis in the geosciences without background knowledge is a great way to go very wrong.

  32. Willard,
    A very apt observation 🙂

    Andrew,

    Time series analysis in the geosciences without background knowledge is a great way to go very wrong.

    I think you can replace geoscience with almost any science.

  33. Sam taylor says:

    Willard,

    And presumably that black American citizens are intrinsically worse workers than whites.

  34. TrueSceptic says:

    Let us not forget Nigel Lawson’s infamous “…this Slingo woman” comment. Can you imagine anyone saying something like “…this Jones man”?

  35. toby52 says:

    If financial statisticians “in the city” think that a temperature time series is just a sequence of numbers without any physical constraints or physical theory behind it, then their analysis would be useless.

    They would also need to understand how the data were gathered before any of their analysis were worth attending to. In other words, they would need to learn some physics, maybe a lot of physics.

    And I am sure there are good scientists who would prefer the opportunity of field trips to the Greenland Ice Cap, before any well-paid city job, even if they had hair like Richard Tol.

  36. Willard says:

    It might be interesting to know at what Donald Trump excels except justifying his salary:

    Donald overcombs bankcrupcies, that’s for sure.

  37. MIke Pollard says:

    Tol’s comments give very clear insight into his attitude toward science and research. Its about chasing the money, the actual science doesn’t even come into it. As I’ve said before, he can be dismissed as a scientist. He is the academic equivalent of an ambulance chaser.

  38. Willard says:

    Report from a study:

    During the boom from 1997 to 2006, London and the south-east was responsible for 37% of the UK’s growth in output. Since the crash of 2007, however, their share has rocketed to 48%. Every other nation and region – with the exception of Scotland – has suffered relative decline over the same period. The upshot is about a quarter of the population is responsible for half of the UK’s growth, leaving the remaining three-quarters of Britons to share the rest.

    The research also shows that the UK’s highest-earners have become relatively more prosperous after the crash, while many on middle incomes are being squeezed hard. In austerity Britain, the top 20% of earning households are enjoying 37.5% of all Britain’s income growth, even after accounting for taxes and benefits.

    http://www.theguardian.com/business/2013/oct/23/london-south-east-economic-boom

    Our top economic elite excels at crashing economies and is compensated accordingly.

  39. pbjamm says:

    If Richard Tol were truly talented or skillful he would obviously be employed in the private sector. Instead he is relegated to the minor leagues of academia with the rest of the second rate mathematicians.

  40. Harry Twinotter says:

    The place to argue it out is in the scientific literature of course. Let Doug Keenan publish a response or rebuttal peer-reviewed paper in a scientific journal.

  41. tol
    One feature of labour markets around the world is that people who are better at something, get paid more.
    ———-
    bankers are some of the best paid. But who was it that stated the last financial crash.
    Hmmm! come to think of it perhaps they came out of that financially enhanced anyway! Such skill!

  42. Magma says:

    I trust a clever person like Mr Keenan made a fortune by predicting the financial crash of 2007? johnrussell40

    In that spirit, may I quote from the parliamentary inquiry into the 2007 failure of Northern Rock, then chaired by global warming skeptic Matt Ridley?

    Mr Fallon: But it was your duty as Chairman and as a Board to ensure that your bank was liquid.
    Dr Ridley: We reviewed liquidity regularly and we reviewed our policy on liquidity and our policy on funding regularly.
    Mr Fallon: But you were wrong?
    Dr Ridley: We were hit by an unexpected and unpredictable concatenation of events.

  43. anoilman says:


    Richard Tol: claims “the best time series analysts are indeed in finance.”

    …. and he’s in academia. That would explain a lot of his woes;
    http://www.desmogblog.com/2015/08/03/richard-tol-s-gremlins-continue-undermine-his-work

    I wonder if Richard would kindly produce evidence that his personal private beliefs are true. I still can’t find his 300 missing papers, so I have my doubts about an awful lot of things he claims.

  44. Magma says:

    Keenan’s unpublished and reference-free note criticizes the IPCC choice of a simple linear model to calculate temperature trends:

    The third paragraph states that the IPCC has chosen a statistical model that comprises a straight line with first-order autocorrelated noise. If you are unfamiliar with such noise, that does not matter here. What is important here is that a model has been chosen, yet there is no scientific justification given for the choice. The failure to present any evidence or logic to support the assumptions of the model is a serious violation of basic scientific principles. (Keenan)

    Keenan has made his claim with reference to the following paragraph of Box 2.2 from AR5 WGI:

    The quantification and visualisation of temporal changes are assessed in this chapter using a linear trend model that allows for first order autocorrelation in the residuals (Santer et al., 2008; Supplementary Material 2.SM.3). Trend slopes in such a model are the same as ordinary least squares trends; uncertainties are computed using an approximate method. The 90% confidence interval quoted is solely that arising from sampling uncertainty in estimating the trend. (IPCC)

    Strangely enough, Keenan doesn’t bother to refer to the three pages of 2.SM.3 or the Santer et al. paper, both of which go into significant detail as to why this model was used. Keenan also ignores a later paragraph in Box 2.2 that states:

    Smoothing methods that do not assume the trend is linear can provide useful information on the structure of change that is not as well treated with linear fits. The linear trend fit is used in this chapter because it can be applied consistently to all the data sets, is relatively simple, transparent and easily comprehended, and is frequently used in the published research assessed here. (IPCC)

    But perhaps Dr. Tol can explain. Maybe it has something to do with money?

  45. Guys: This is getting ridiculous. Keenan’s benchmark is ARIMA(3,1,0). This is physically impossible in the very long run, but 150 years is short in climate terms.

  46. John Hartz says:

    ATTP: The downside of this OP and others like it is that you give climate science deniers like Doug Keenan free publicity. An upside is reading the absurd comments made by Richard Tol and the responses to them.

    Keep up the good work,

  47. BBD says:

    John

    ATTP is only helping to remind us that Doug Keenan doesn’t believe in radiative physics.

  48. Richard,

    This is getting ridiculous. Keenan’s benchmark is ARIMA(3,1,0). This is physically impossible in the very long run, but 150 years is short in climate terms.

    So what? Noone disputes that you need to use this kind of analysis to get trend and uncertainties for the instrumental temperature record. That’s bleeding obvious. The rest of what Keenan says simply illustrates his extreme ignorance about this topic. That you seem to not recognise this does you absolutely no favours whatsoever. Come on, this is getting ridiculous. Either you really haven’t bothered to check what Keenan is actually saying, or you’re as clueless as he is. At the moment, I’m somewhat divided as to which one it is.

    [Edit: Okay, I’ve just realised what you’re suggesting. You really seem to think that over a period of 150 years, it could simply be a random walk? If so, this is bizarrely and embarassingly wrong.]

  49. @BBD
    The question is whether the observed warming since 1880 is statistically significant or not. That is a univariate question.

  50. The question is whether the observed warming since 1880 is statistically significant or not. That is a univariate question.

    Jeepers, Richard, you really are doubling down here. Relative to a null of zero trend, it’s blatantly obvious that it’s statistically significant. Any other test of significance would require some kind of model that includes the physical processes that influence our climate. That’s called an attribution test.

    Are you really being serious? I’m starting to wonder again if you’re not some kind of random person on the internet impersonating a Professor of Economics who works on climate change, because the possibility that you really are a Professor of Economics who works on climate change is just a little too bizarre to be possible.

  51. You can read Richard Telford’s post about ARIMA(3,1,0) if you want more info on this (Richard you might learn something from this).

    Heres’s the key point that statisticians like Keenan and Tol seem to miss

    Non-stationary models are not physically plausible descriptions of global temperature. Temperature cannot simply drift up and down without violating the laws of thermodynamics. When it gets hot, heat loss by radiation increases, something has to provide that energy.

  52. anoilman says:

    Richard Tol (@RichardTol): I think we shall ask an expert but thanks for offering your academic opinion.

  53. Actually, Richard Telford’s posts (here’s another) are very good. A few choice quotes,

    Keenan is savaging a straw man. Nobody believes that a linear trend is a full description of climate change over the instrumental period. Climate forcings do not increase linearly with time, so it would be absurd to expect global temperature to. The linear trend model is simply a quick test of whether temperature is increasing. Replacing an oversimplified but informative model with a physically meaningless model is not progress.

    and

    But perhaps the climate is usually stationary, maybe from an ARIMA(3,0,0) process, and the non-stationarity that needs differencing only occurs during the last 150 years. What could possibly have caused this non-stationarity, this trend during the last 150 years? It couldn’t be greenhouse gases could it?

    Richard Tol, you should really read these. They’re very informative.

  54. @Wotts
    As I said before, arguments like Telford’s are valid on a much longer time scale than a century and a half.

  55. Richard,

    As I said before, arguments like Telford’s are valid on a much longer time scale than a century and a half.

    Rubbish. Okay, I think I’m now leaning one way.

  56. Marco says:

    “Guys: This is getting ridiculous. Keenan’s benchmark is ARIMA(3,1,0). This is physically impossible in the very long run, but 150 years is short in climate terms.”

    Indeed, it is getting ridiculous. Where is the evidence that ARIMA(3,1,0) is physically possible for periods less than 150 years?

  57. BBD says:

    Like I said, DK apparently does not believe in radiative physics.

    Richard, I would expect you to pay more attention to the implications of DK’s nonsense rather than endorse it.

  58. @Marco
    The impossibility is that in an ARIMA(N,1,P) model, the variance of the temperature tends to infinity as t goes to infinity. That’s physical nonsense (e.g., the temperature is bounded from below) and at odds with the geological record. But with 150 years, the variance is merely large.

  59. Richard,

    But with 150 years, the variance is merely large.

    Indeed, but that’s why you do attribution studies to determine what physical processes are most likely responsible for the observations.

    I’ll explain something to you, since you seem as confused about this as Keenan (surprising given how many years you’ve been associated with this). In the physical sciences, there are probably two main roles for statistical analysis. One is to simply determine the properties of a dataset (trend plus uncertainties, for example), another is to use statistical techniques to try and understand what physical processes might explain some observations (hypothesis testing, or attribution studies). Simply running some kind of statistical model and showing that you can produce a timeseries that matches some observation tells you nothing, unless your statistical model happens to be a reasonable representation of the physical processes (they’re actually random). Hence, what Keenan is trying to suggest is meaningless. As Richard Telford points out, the linear analysis is not intended to suggest that the time series is linear plus some variance, it is simply to answer the question as to what the linear trend actually is (positive, negative, statistically different from zero,…). That ARIMA(3,1,0) may produce a time series that might match the observations should lead to the response, “so what?”.

  60. Richard Tol wrote “Keenan’s understanding of the physics is irrelevant, as the context is a statistical one.”

    Applying statistics without any understanding or consideration of the physical processes is doomed to given meaningless answers, or as Eli said “the right answer to the wrong question”.

    E.g. see this analogy whereby one might claim that you can eat how much you want and never gain weight (based on statistics only, of course): https://ourchangingclimate.wordpress.com/2010/04/01/a-rooty-solution-to-my-weight-gain-problem/

  61. Bart,
    Thanks. Eli linked to your post yesterday. Good illustration.

  62. Has it not occurred to anyone that, while the second moment of an ARIMA model violates the laws of physics, the first moment of a linear trend model does?

    @bart
    HM Government made a statement about statistical significance and Keenan questioned the statistical merits of that statement.

  63. Richard,

    Has it not occurred to anyone that, while the second moment of an ARIMA model violates the laws of physics, the first moment of a linear trend model does?

    Yes, it’s not meant to represent a physical process. It’s simply a result of a form of statistical analysis of the data (didn’t you read my most recent comment). This isn’t complicated. Here’s a question for you. Let’s say you run an ARIMA(3,1,0) analysis until you find a timeseries that matches the observations. What would that tell you?

    HM Government made a statement about statistical significance and Keenan questioned the statistical merits of that statement.

    And, as a result, illustrated his ignorance. I really wasn’t expecting you to do the same.

  64. @Wotts
    That’s not the point. HM Government turned this in a statistics issues, agnostic of physics. Keenan just played within the government’s parameters.

  65. Richard,

    That’s not the point. HM Government turned this in a statistics issues, agnostic of physics. Keenan just played within the government’s parameters.

    No, it’s not a statistic issue, agnostic of physics. If you had a modicum of understanding of this, you’d be embarassed by what you’re saying (well, assuming that you’re capable of being embarassed by what you say). You still haven’t answered the question I posed in my previous comment.

  66. @Wotts
    I agree that this is not a statistics issue, and that statistics agnostic of physics should be avoided. But you can’t pin this on Keenan. HM Government issued a statistical statement based on a physically impossible model, and Keenan responded with superior statistics (and superior physics, by the way, as Keenan violates in second moment rather than first).

  67. Richard,

    I agree that this is not a statistics issue, and that statistics agnostic of physics should be avoided. But you can’t pin this on Keenan.

    Rubbish.

    HM Government issued a statistical statement based on a physically impossible model, and Keenan responded with superior statistics (and superior physics, by the way, as Keenan violates in second moment rather than first).

    No he did not. HM government did not issue a statement based on a physically impossible model. They presented results from basic statistical analysis of a dataset. This is standard practice in the physical sciences. Do you not realise this? Are you trying to sound like an ignorant fool, or do you not realise that you sound like an ignorant fool? This is a serious question. You could also answer my earlier question. What does an ARIMA(3,1,0) analysis tell us? I’ll even give you a clue. The first 4 letters of the answer are “n”,”o”,”t”,”h”.

  68. @wottsywotts
    Statement 1:
    T(t) = a + bt + u(t)

    Statement 2:
    T(t) = a + u(t)/(1-p1L-p2L^2-p3L^3)(1-L)

  69. Magma says:

    I’m unused to seeing anyone even peripherally associated with research in the physical sciences play the sort of empty semantic games that Dr. Tol seems so fond of.

  70. Richard,
    Hmm, I don’t think you understand my question. I’m asking you to explain what an ARIMA(3,1,0) analysis would tell us about a time series, not simply present the underlying equation. You do understand the difference, don’t you?

  71. I’m unused to seeing anyone even peripherally associated with research in the physical sciences play the sort of empty semantic games that Dr. Tol seems so fond of.

    Really? You haven’t seen him do this before?

  72. MarkB says:

    Time-series analysts in finance generally get paid 5–25 times as much as those in academia; so analysts in finance do naturally tend to be more skillful than those in academia

    A necessary (and arguably, sufficient) condition to see the world as a great many skeptics see it is that one unconditionally accept the postulate that the market provides the optimum solution regardless of the particular question at hand.

  73. jsam says:

    “Climate sceptics” (as opposed to real sceptics) insist they want polite discourse. They do not know what that entails from themselves.

    I may have missed “A response on statistical models and global temperature” this in the upposts. It’s a polite response to a rude (and wrong) man.
    http://blog.metoffice.gov.uk/tag/doug-keenan/

  74. Sorry. a=0 in statement 2.

  75. Richard,
    That still doesn’t answer my question. Was it too complicated, or is there no actual answer?

  76. @wotts
    You do need to re-read Box and Jenkins. An ARIMA(3,1,0) does not tell you anything. The fact that it is an ARIMA(3,1,0) does tell you something, namely that there is a unit root but no moving average and no linear trend. The AR part can only be interpreted in conjunction with its roots. ARIMA(3,1,0) is a system with fast moving parts and slow moving parts, but nothing much in between. (Sounds a bit like an atmosphere and ocean, innit.)

  77. Richard,
    Based on your answer, it appears that it can’t actually tell you anything specific about the data, which is one of the main points of doing data analysis.

    The fact that it is an ARIMA(3,1,0) does tell you something, namely that there is a unit root but no moving average and no linear trend.

    As Richard Telford was pointing out, we don’t think that the actual trend is linear. The reason we determine linear trends is because it can tell us some simple things about the dataset. That it isn’t necessarily actually linear isn’t particularly informative. Understanding the actual evolution of the system takes physical models, not statistical models that appear not to actually quantify anything and seem to simply tell us what we could probably establish by simply looking at the data itself.

  78. Magma says:

    ATTP: Really? You haven’t seen him do this before?

    I’ve generally skipped over such exchanges; this is the first I’ve followed in any detail.

    But thinking about it a little more, Curry does much the same, doesn’t she?

  79. Magma,
    Maybe a little, but I think Richard is the real master.

  80. entropicman says:

    I just spent six weeks debating Does Climate Science Exist at BH.

    They concluded that climate sccience is wrong because

    1) EM is a drunk
    2) EM is deluded.
    3) EM can’t spell.

    I had hoped for “climate science is wrong because {insert evidence here}”. Perhaps that was too much to expect.. 😕

  81. @wotts
    Very few people can estimate an ARIMA model by eyeball, and even they always follow up with a proper analysis.

    A statistical analysis tells you about the statistical properties of your data. In this case, the analysis tells you that the data are much more likely to contain a unit root than a linear trend. Any statement based on a linear trend is therefore spurious, a statistical nonsense.

  82. EM,
    I read some of that. I almost commented a few times, but decided against it. Was impressed by how you and Raff did. The rest – as you point out – was complete nonsense. That’s actually why I didn’t comment. The only comment that seemed appropriate was “are you complete idiots?”.

  83. BBD says:

    It’s like ploughing the sea, Entropicman. Except without the pleasure of frolicking in the surf.

  84. entropicman says:

    ATTP

    Against cognitive dissonance
    The gods themselves
    Contend in vain.

    I had hopes that one of the group was becoming interested in the actual science, which was why I persevered for as long as I did. Unfortunately he reverted to type.

  85. Richard,

    Any statement based on a linear trend is therefore spurious, a statistical nonsense.

    Well, this is nonsense. If you have a dataset that appears to have some kind of trend, approximating that by a linear trend is perfectly normal. That it may not actually be linear is not really the point. It simply gives you a number (positive, negative, zero, uncertainty,…). Of course, if it is very non-linear, you may decide that this analysis isn’t appropriate, but that it isn’t exactly linear, is not a reason to avoid determining the linear trend.

    If you want to understand the actual evolution, you use models of the physical system, not complicated statistical models that don’t tell you anything of the underlying physical processes, but tell you what is probably already obvious from the data itself.

    Oh well, this has been fun, but it is getting late. Maybe we can stop going in circles. Next time I criticise someone associated with the GWPF, you’re welcome to come back and defend the indefensible.

  86. Richard,
    Oh, since you’re commenting here, are you going to respond to Gavin’s comment?

  87. @wotts
    Nope. See Granger and Newbold (1974).

  88. Richard,
    What does a paper titled “Spurious regressions in econometrics” have to do with this?

  89. Sam taylor says:

    Nobel prize for peace really don’t mean what it used to.

  90. entropicman says:

    TE might be interested to know that HadCRUT4 monthly figures are currently showing an anomaly of 0.68C for 2015.

    That would put 2015 well within the range projected by the models.

  91. BBD says:

    Alarmist cherry-pick.
    🙂

  92. Phil says:

    ==> “One feature of labour markets around the world is that people who are better at something, get paid more. …
    When I left academia, I signed up to a recruitment agency; one of the questions they asked me was “did I want my CV sent to any defence contractors”. Apparently it was a common request; many people in my business didn’t want to work for defence contractors. Presumably that diminished the labour market, and forced wages up in those companies simply because candidates were rejecting them on (a) principle.
    It would seem likely that Tol’s comment on labour markets becomes less true the higher the income. Once you have fed, sheltered and clothed yourself you have the “luxury” of some principles.

  93. Eli Rabett says:

    Richard Tol

    Sorry. a=0 in statement 2.

    Good, no danger of your getting the sign wrong then.

  94. > Good, no danger of your getting the sign wrong then.

    Ouch!! 😉

  95. Did you see the Bank of England’s Mark Carney and other financial bods discussing the economy and when they’d need to put up interest rates on the news tonight? Bottom line: they haven’t got a fucking clue. And these overpaid creeps have the cheek to criticise climate models?

  96. TrueSceptic says:

    entropicman,

    I followed your link to BH. I admire your patience and good humour but did you really think at any time that you going to make any progress against such a terminally deluded bunch, as good an example of DK as you will find anywhere? I’ve seen others try this at other delusionist sites over the years, always with the same result.

    The denizens of such places are beyond help, minds shut out forever from the ability to consider that they just might have got it wrong. Isn’t there something a bit scary about people like that being so ignorant and yet so arrogant?

  97. Tom Dayton says:

    This week my business partner and I lunched with a former coworker of ours who is working at NASA’s Ames Research Center, where he is a contractor. NASA Ames is smack in the middle of Silicon Valley; Google literally is on the other side of the fence, where it is building a major campus on NASA Ames’s land. Private companies’ recruiters bark outside the fence, trying to drag away workers. My biz partner and I contracted there full time for years until a few months ago (not enough full-time contracting money to rely on). Now that my business partner and I are consulting to startup companies and venture capitalists outside of NASA, we offered to connect our most excellently qualified software developer friend to job opportunities in the private sector. He was not enthusiastic, despite the constant and very real threat of being unemployed at NASA with only two weeks notice and no severance pay, from any number of causes including whims of Congress. He could easily double his income by working for a private company, and not just in the space domain. But he’s not interested, because “It wouldn’t be NASA.”

    Likewise, my biz partner and I continue to try to get more work with NASA, including spending an entire freakin’ month submitting two proposals, when we could have been marketing ourselves to a bunch of private companies instead, for a lot more money. Why? Because NASA is cool. We want to support the common good, such as democratization of space by creating low-cost nanosatellite technology, low-cost rocket Guidance, Navigation, and Control (GNC) systems, robust general purpose object-oriented (e.g., internet of things) software frameworks (MCT–Mission Control Technologies from Ames), uncrewed aerial vehicle systems, . . . .

    For many people, money is not everything.

  98. Kevin O'Neill says:

    Richard Tol writes: “One feature of labour markets around the world is that people who are better at something, get paid more. …”

    Really? What world is this modeled from? I’ve worked in corporations large and small. Here in the US and overseas. I’d make the opposite argument; it’s rare that merit is rewarded with increased pay. In most manual labor and service industry jobs seniority determines pay, not merit. Likewise for most unionized workplaces. In salaried positions it’s as often who you know, who you’re related to, and/or whose ass you’re willing to kiss Even when merit is considered, the emphasis is usually placed on the absolute wrong metrics.

  99. anoilman says:

    Which financial guy accurately modeled the 2008 financial collapse I wonder?

  100. @wotts
    Granger and Newbold (1974) show that your remark of August 6, 2015 at 8:34 pm is simply false: Your statistical test for a linear trend is based on assumptions that are unsupported. Even if there were a linear trend, this is not how you would show there is one.

    By the way, this is what we mean when we say that natural scientists are “decades behind” when it comes to statistical theory. This is not an obscure paper — and yet, 41 years later, you were not aware of it(s contents — most now learn this from a textbook).

  101. Richard,

    Your statistical test for a linear trend is based on assumptions that are unsupported. Even if there were a linear trend, this is not how you would show there is one.

    You need to improve your reading comprehension (again). Read my 8:34pm again. I didn’t say “test for a linear trend”. I said that you can determine the linear trend, even if the underlying trend isn’t actually linear. Linear regression is not a statistical test to determine if the trend is linear or not. It is simply an analysis tool for determining the linear trend. It gives you a number (oC/decade). It tells you nothing of the underlying physical processes. It tells you nothing of whether or not such a trend will continue. It tells you nothing about what the trend was prior the time interval you're analysing.

    By the way, this is what we mean when we say that natural scientists are “decades behind” when it comes to statistical theory.

    Or, statisticians are so clueless when it comes to the physicals sciences that we’re occupying different temporal states. Here’s maybe the bit that you keep ignoring (or, not understanding). Physical scientists use statistics to quantify some properties of a dataset (for example, what is the best-fit linear trend?). If physical scientists want to actually understand the system being measured, you typically use physics (with statistics as a tool), not just statistics.

    Anyway, this is now gone from surreal to ridiculous, so maybe we can call it quits.

  102. Given Richard’s apparent suggestion that the Met Office are wrong because their test for a linear trend is wrong, I’ll simply repeat Richard Telford’s point.

    Keenan is savaging a straw man. Nobody believes that a linear trend is a full description of climate change over the instrumental period. Climate forcings do not increase linearly with time, so it would be absurd to expect global temperature to. The linear trend model is simply a quick test of whether temperature is increasing. Replacing an oversimplified but informative model with a physically meaningless model is not progress.

    I thought this was a fairly simple and straightforward point.

  103. semyorka says:

    “One feature of labour markets around the world is that people who are better at something, get paid more. Lionel Messi, for instance, earns more than Leroy Fer”
    So Raheem Sterling is the sixth best footballer ever. And David Luiz the 10th best.
    This is why there are no economists managing premiership football teams.

  104. semyorka says:

    Cricket analogy, look at the IPL valuation of English vs Australian cricket players then predict the fourth test at Trent Bridge….. 😉

  105. Sam taylor says:

    It’s quite amusing to see richard getting his knees dirty because he’s found someone who’s better at time series than he is.

  106. @wotts
    Estimation is a statistical issue. You can use physics to determine whether a trend could be expected to be linear, or whether a linear trend is a good approximation, but estimation is estimation.

    If you estimate a linear trend like that, then (a) your estimate is biased and (b) your estimate of the standard error is strongly biased downwards. Statistical significance thus becomes spurious (in Pearson’s sense of the word).

    The fact that is common practice in physics is an argument from authority.

  107. Richard,
    Keep savaging that strawman, defending the indefensible, and illustrating – once again – why insisting that pure statisticians should be checking the work of physical scientists is a really bad idea. Yesterday I was trying to implement something new in my code which needed to be recompiled over and over again, and so I had time to engage in this rather pointless discussion. Today, I have better things to do.

    If you estimate a linear trend like that

    Are you suggesting that there could be more than one linear trend?

  108. Richard Tol writes: “One feature of labour markets around the world is that people who are better at something, get paid more. …”

    When my youngest son came out of university with his MEng in computer science a few years back he was immediately invited to three interviews and they all wanted him. One of the jobs was with a bank. It offered him 50% more starting salary than the others. He turned it down as he said the work looked boring. The financial industries pay more, not because their people are better, but because money is no object. They just skim off, parasitically, whatever they need from the rest of society who create the value, and whose money passes through their hands.

  109. Quiet Waters says:

    “Lionel Messi, for instance, earns more than Leroy Fer”

    Leroy Fer earns a lot more than Stuart Broad but I don’t see him bowling the Aussies out any time soon…

  110. @wotts
    No I’m not suggesting there is more than one linear trend. The point is one of bias. In a finite sample, the estimated slope of the trend is not equal, in expectation, to the actual slope. More seriously, the standard error of the slope of the estimate is too small so that you find significance where you should not.

    All this is because stationary and non-stationary statistics are quite different, and procedures that are designed for stationary data come apart when applied to non-stationary data. It’s a bit like using Newtonian mechanics to model quantum effects.

    You argue that a linear trend is just a simple way to approximate any process. Well, so is an AR(1) model: It can approximate anything. The issues are, of course, the quality and validity of the approximation.

  111. Richard,

    The point is one of bias. In a finite sample, the estimated slope of the trend is not equal, in expectation, to the actual slope.

    There is no actual slope, at least not one that is perfectly linear. Linear regression is not being used to determine some “actual slope”, it is simply being used to determine a nice, simple property of a dataset; that property being “what is the best-fit linear trend?”. You really are savaging a strawman.

    Well, so is an AR(1) model: It can approximate anything.

    A linear trend isn’t really trying to approximate anything, it’s trying to give you a number that you can present to people who would like to know something of the system that is being observed. For example, what is that average warming trend over the instrumental temperature record? If you (or Doug Keenan) would like to determine a different property of that dataset, go ahead. That doesn’t mean that the properties that others have determined are wrong or have no information.

    The issues are, of course, the quality and validity of the approximation.

    It’s not really an approximation! If you want to actually determine more details of the system being observed you need to consider models that can actually represent this system. As I’ve pointed out numerous times, linear regression is NOT being used to claim that the trend is linear. It is simply being used to determine a best fit linear trend so that you can see if we’ve warmed (or not) and – on average – how fast.

    If you want to fit some kind of different function, go ahead. However, I’d be surprised if anyone gained anything from being told the coefficients of a polynomial that best fits this data, or the random number seed that produces the closest fit.

  112. @wotts
    So, you would fit a linear trend and then say “the linear trend is 0.2K/decade, but the trend is not linear, so 0.2K/decade is a meaningless number”?

  113. Richard,
    As I said, I don’t have time to deal with your strawman arguments today.

    So, you would fit a linear trend and then say “the linear trend is 0.2K/decade, but the trend is not linear, so 0.2K/decade is a meaningless number”?

    No, try reading harder. Actually, don’t bother, this isn’t going to go anywhere sensible.

  114. @wotts
    So what would you do? Above, you seem to argue that you would estimate a linear trend even if you know the trend is not linear, and even if you know that the estimator is biased.

  115. Richard,

    So what would you do? Above, you seem to argue that you would estimate a linear trend even if you know the trend is not linear, and even if you know that the estimator is biased.

    If I wanted to know the best-fit linear trend to a dataset, I would determine the best-fit linear trend (plus its uncertainties). I would, however, not claim that this best-fit linear trend represented anything other than the best-fit linear trend to that dataset. I’ll even repeat this. I’m not trying to estimate the trend, I’m trying to determine the best-fit linear trend. This is not complicated. Also, I’m not sure what you think is being estimated, other than the best-fit linear trend. Given that the best-fit linear trend is well-defined, it’s hard to see how it can be biased.

  116. @wotts
    The best-fit linear trend is consistent only; it is not unbiased.

  117. Tony Lurker says:

    @tol
    One feature of labour markets around the world is that people who are better at something, get paid more.

    I’m going to agree with this if that “something” is being able to bullsh*%t people. In my experience in both the public and private sector, the quality of your work is only weakly coupled to your salary. The strong coupling is how good you are at convincing people to give you money, regardless of your actual ability.

  118. Richard,
    The best-fit linear trend to a dataset is what it is.

  119. @wotts
    I agree that it is uniquely defined. Then again, if that is your criterion for using a method … AR(1) is uniquely defined, too, by the way.

  120. Richard,
    I didn’t specify a criterion, and I’m not the one harassing the Met Office. As I understand it, the standard analysis is a linear trend with first order autoregressive noise anyway.

  121. Willard says:

    > The fact that is common practice in physics is an argument from authority.

    Actually, it’s an appeal to tradition. Here would be an argument from authority:

    How do I know? My initial specialization was in time series (taught by none other than Bierens). I’ve taught the stuff, published a number of papers, and supervised master’s and PhD theses. Colleagues have contributed to the theory. I’ve been the editor of a journal that publishes a lot of time series analysis. I’ve read and reviewed time series analyses in finance, economics, political science, biology, epidemiology and geophysics. The most advanced time series analysts are in finance. Geophysics is several decades behind the curve.

    Stick to Gremlins, RichardT.

  122. @wotts
    Another argument from authority.

    @willard
    As you can see from the preceding comments, I was asked for my credentials in time series analysis.

    I did appeal to Granger’s authority, but his work can be readily found (as demonstrated by our host). There are plenty of accessible accounts of his work, including his Nobel lecture.

  123. Richard,

    Another argument from authority.

    It wasn’t even an argument.

    I was asked for my credentials in time series analysis.

    No you weren’t.

    Look this is getting tedious. I can understand that Andrew Montford (who I’ve explained this to before) simply doesn’t understand the difference between descriptive statistics, and inferential statistics. I can understand that Doug Keenan was simply paid way too much money to ever be capable of realising that there are some things he simply doesn’t understand. You – on the other hand – should be able to get this. You, also, should be able to realise that Doug Keenan’s apparent attacks on scientists and researchers is fundamentally despicable. That you would implicitly defend him is – unfortunately – no great surprise, but, in my opion, is indefensible.

  124. Sam taylor says:

    For all the talk in the last threat about conservation models, [Mod : Richard] appears to have forgotten that credibility can be both created and destroyed.

  125. Joshua says:

    ==> “Look this is getting tedious. ”

    Closing barn doors, bolted horses.

  126. @wotts
    I hardly know Keenan, and I have no opinion about his behaviour. He is correct in this case, however: HM Government said things in Parliament that they could not support.

  127. Richard,

    He is correct in this case, however: HM Government said things in Parliament that they could not support.

    No he is not. What was presented was an analysis of a dataset (descriptive statistics). It was very obviously an analysis of a dataset. That Keenan (and yourself, it would appear) was confused about what was presented, does not make him right. Keep savaging that strawman, though, and aligning yourself with people who think attacking others who present research with which they disagree is the way to go.

  128. Willard says:

    > Another argument from authority.

    The first one you identified wasn’t an argument from authority, RichardT. In a way, a tradition (not a school, but “how everyone somewhere do things”) is the opposite of an authority. Granger is an authority; econometricians’ common knowledge is not really an authority.

    Not every argument that appeal to something are appeals to authority, you know.

    ***

    > I was asked for my credentials in time series analysis.

    Gremlins top credentials, RichardT, and AT did not ask about your credentials, but how do you know that “the best time series analysts are indeed in finance.” I’m not sure how your credentials answer AT’s first question.

    Also, AT’s second question was: who cares? You seem to have forgotten about that one. That questions the relevance of our ClimateBall episode.

  129. FWIW, here’s the question asked by Lord Lord Donoughue

    To ask Her Majesty’s Government, further to the Written Answer by Baroness Verma on 30 October (WA 114–5) stating that global temperatures have risen less than 1 degree celsius since 1880, on what basis they assert that there has been a long-term upward trend in average global temperatures. [HL3048]

    To ask Her Majesty’s Government, further to the Written Answer by Baroness Verma on 30 October (WA 114–5) stating that there has been no significant global warming since around 1998, and deeming that period as a shorter timescale, how many years of non-warming they consider would constitute a long-term trend.[HL3049]

    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]

    and here is the the response made in by Baroness Verma.

    The assessment that there has been a long-term upward trend in global average near-surface temperatures since the late 19th century is based upon three global temperature records, compiled from observations, by groups in the US and UK. The rate of global temperature rise on different timescales is summarised in table 1 below. The underlying trend over the period from 1880 to 2011 is 0.062 celsius per decade, giving a total change of 0.81 celsius. Such a rate of change has been judged by major scientific assessments to be large and rapid when compared with temperature changes on millennial timescales.

    Over this period some parts of the world have warmed at a much faster rate. The land surface average temperature has risen by about 1.1°C and Arctic temperatures have increased by almost twice the global average rate. The consequences of this warming are already seen across the globe. For example, northern hemisphere sea-ice and snow cover have decreased markedly, most glaciers have retreated and the risks of certain extreme weather events occurring have increased.

    8 Nov 2012 : Column WA225

    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.
    Time period Linear trend (°C/decade) Absolute change in temp… described by linear trend (°C)

    1880-2011 0.062±0.009 0.81±0.13

    1900-2011 0.074±0.011 0.82±0.13

    1950-2011 0.106±0.025 0.66±0.16

    1970-2011 0.166 ± 0.038 0.70 ± 0.16

    Table 1. Trends fitted to monthly global temperature anomalies for HadCRUT4, with uncertainties describing 95% confidence interval bounds for the combination of measurement, sampling and bias uncertainty and uncertainty in the linear trend fitted to the data. The statistical model used allows for persistence in departures using an autoregressive process (ie that an individual value is not independent of the previous one).

    Statistical analyses and modelling of the global temperature record have shown that, because of natural variability in the climate system, a steady warming should not be expected to follow the relatively smooth rise in greenhouse gas concentrations. Over periods of a decade or more, large variations from the average trend are evident in the temperature record and so there is no hard and fast rule as to what minimum period would be appropriate for determining a long-term trend.

    Apart from pedantic nitpicks (which is all some people have) there’s not much wrong with this response.

  130. anoilman says:

    Finance analysts are always selling yesterday’s lottery ticket.

  131. @wotts
    Exactly. Baroness Verma presented a statistical analysis in Parliament. She wrote “0.062±0.009”. Both numbers are wrong: It is a biased estimate of the linear trend, and it is strongly biased estimate of the standard error.

    I think we should hold government representatives to account when they speak in Parliament.

  132. Marco says:

    “He is correct in this case, however: HM Government said things in Parliament that they could not support.”

    And Richard Tol would know how bad that is, since Richard Tol calculated there should be about 300 additional abstracts that support the contrarian position, but could not support that result when asked. He didn’t even try…

  133. Richard,

    she wrote “0.062±0.009″. Both numbers are wrong: It is a biased estimate of the linear trend, and it is strongly biased estimate of the standard error.

    What absolute and utter rubbish. It is a result of linear trend analysis with autocorrellated errors, just as stated. There is no “linear trend” other than that estimated using this method. This is not complicated. What you’re doing here is utterly bizarre. As I said, keep associating yourself with organisations that spread misinformation (GWPF) and people who attack other researchers (Keenan).

  134. Eli Rabett says:

    Perhaps Eli might lend a paw here. There is a set of numbers, a few tens, a few hundred, a few thousand. Whatever. Statistical analysis in isolation can reduce that large(r) set to a few parameters (mean, standard deviation, intercept, slope), by imposing a statistical model (ARIMA, Gaussian, etc). The more complex the model, the more parameters. In the worst case (see Tol 2009, Cook 2011) you have more parameters than data points, but so it goes. On the other hand, there are many cases where the simplest parameterization (the average, the linear model, etc) provides a useful summary of a large data set.

    That is why the IPCC reporters chose to report the fitting of the data to a linear model.

    Now the bunnies come to the next question, what statistical model should be used? The optimal choice for science is the model suggested by theory developed from first principles. This is what attribution studies use. When there is no physical guidance the most penurious choice is recommended consistent with the task at hand.

  135. Steven Mosher says:

    “the best time-series analysts tend to be in finance.”

    i suppose an alternative case could be made for other disciplines like signal analysis, so the best tend to be in electrical engineering or guidance and control… haha

    Looking at this from a publication stand point I think we could argue that the art of time series analysis has been pushed forward by a couple disciplines: econometrics and signal theory( think kalmen filter ).

    i think its fair to say that climate science tends to be a user of time series methodologies created for other disciplines rather than a creator of new approaches. Climate science tends to use a cook book created by folks who work in other disciplines. That’s probably because climate science doesnt present any unique or novel problems. With one exception perhaps and that is working with time series ( like paleo) where the sampling intervals are very different.

    I suppose it comes down to how you measure or determine the “best”

    1. creates to the most innovations/break throughs?
    2. makes the fewest errors?

    if we are talking about #1, then yes I think you can make a good case that the best come from
    say finance or control theory. That’s the rationale underlying Tol’s position.
    But Keenan, it seems, wants to rely on #2.

  136. TE might be interested to know that HadCRUT4 monthly figures are currently showing an anomaly of 0.68C for 2015.
    That would put 2015 well within the range projected by the models.

    Check the base reference periods.

  137. TE,
    I saw your comment on Judith’s blog. I’m not convinced you calibrated the models and observations properly.

  138. @Steve
    Apologies. Signal theorists are pretty good at this stuff too, and there is lots of cross-fertilization between signal theory, statistical physics and finance.

  139. @Wotts
    Same point again. The method used to estimate linear trend and AR(1) is incorrect (and has been known to be incorrect since 1974).

  140. Richard,

    The method used to estimate linear trend and AR(1) is incorrect (and has been known to be incorrect since 1974).

    Your 1974 references clearly does not show that this is true in this case.

    Anyway, I’m no longer interested. If you want to trash your reputation to promote misinformation and climate science denial, be my guest. Maybe you could avoid doing it too often here as I’m not that comfortable observing it happening, or being associated with it.

  141. pbjamm says:

    @Richard Tol :
    “She wrote “0.062±0.009″. Both numbers are wrong: It is a biased estimate of the linear trend, and it is strongly biased estimate of the standard error.”

    So what would the correct numbers be? Are they significantly different from the numbers presented? I am not a statistician so I am genuinely curious.

  142. @pbjamm
    You can find all that in Keenan’s work (or in Kaufmann-Stock, Estrada-Perron, you name them). The problem is not just that the method is incorrect, but the data are not trend-stationary in the first place.

  143. Richard,

    The problem is not just that the method is incorrect, but the data are not trend-stationary in the first place.

    As I’ve said before, so what? That’s rhetorical. The conversation has been unbelievable, which is no great surprise.

  144. pbjamm says:

    You must have missed the part of my comment where I said I am not a statistician. I am sure I would not be able to find the answer in Keenan/Kaufmann-Stock/Estrada-Perron nor recognize it if I did happen upon it.

  145. BBD says:

    The word, ATTP is ‘bullshit’.

  146. Willard says:

    > i suppose an alternative case could be made for other disciplines like signal analysis, so the best tend to be in electrical engineering or guidance and control…

    Add machine learning, psychophysics, demography, complexity theory, and many other disciplines requiring an analysis of time series that go beyond 14 data points, a baby time series on which RichardT fumbled not so long ago.

    RichardT’s claim goes beyond his own credentials.

    I blame Gremlins.

  147. pbjamm,

    I am sure I would not be able to find the answer in Keenan/Kaufmann-Stock/Estrada-Perron nor recognize it if I did happen upon it.

    That’s both intentional and because there isn’t an answer. As I understand it, Richard’s point is based on the idea that if you had a system in which the evolution was a linear trend with some correlated noise, then a linear trend analysis using autocorrelated noise would not identify the true underlying trend, nor it’s correct uncertainty. However, these observations are not of a system in which the underlying evolution is a linear trend with autocorrelated noise. It is a complex system that responds to changes in forcings, that are not linear, and to internal dynamics. The linear trend analysis with autocorrelated noise is simply an analysis tool for determining the best-fit linear trend and the uncertainty in the trend, which is assumed to be influenced by the correlations. It is not intended to identify some kind of real underlying linear trend, but simply return an answer to questions like; what is the best-fit linear trend?, have we warmed?, what is the likely range of warming?. It is an analyis tool. It’s is not trying to infer something about the underlying processes that drive the evolution of the system. This is not even all that complicated.

    BBD nails it.

  148. Willard says:

    Oh, and here’s RichardT’s authority:

    http://wolfweb.unr.edu/~zal/STAT758/Granger_Newbold_1974.pdf

    Since noone wants to specify climate as a linear model, one has to wonder where’s the misspecification in the first place.

  149. Since noone wants to specify climate as a linear model

    Exactly.

    I find myself having similar views to those of Andrew Gelman.

  150. @pbjamm
    In lay terms, what Verma presented is just nonsense. The results are meaningless. The numbers do not have an interpretation. The world did not warm on average 0.06K per decade, because the average is not defined.

    @willard
    No one may want to do this, but Verma did so nonetheless.

  151. Richard,

    In lay terms, what Verma presented is just nonsense. The results are meaningless. The numbers do not have an interpretation. The world did not warm on average 0.06K per decade, because the average is not defined.

    What BBD said. This is not what Verma said. I’m genuinely no longer interested in discussing this with you. If you really want to trash your reputation on blogs, feel free to do so. However, I’d really rather you didn’t do it here. Climate science denial blogs like WUWT and Bishop-Hill would welcome your presence, and applaud what you say.

    No one may want to do this, but Verma did so nonetheless.

    No, she did not.

    I seriously have no idea why you behave as you do, nor why you promote the garbage that you do. I, however, have no great interest in finding out. I really do hold the same view as Andrew Gelman and have no great interest in wasting any more of my time interacting with someone whom I hold in such low regard. There really are better things to do and there are plenty of other sites where you can promote your mis-information. I’d appreciate it if you stopped doing it here.

  152. John Hartz says:

    ATTP:

    There really are better things to do and there are plenty of other sites where you (Richard Tol) can promote your mis-information. I’d appreciate it if you stopped doing it here.

    Bravo!

    PS – I ran out of popcorn a couple of days ago. 🙂

  153. Willard says:

    Are you seriously claiming that Verma specified a model, RichardT?

    In any case, pray tell more about baselines:

    The serious question is what is the baseline for each study and which sets the zero of the temperature scale? Did Poor Richard align the baseline for all the calculations. For which is the baseline pre-industrial (in which case the world is now past Tol’s outlying positive point at 1.0 C), 1860 or so when instrumental records start in which case the world is pretty close to it, or some more recent time, in which case, another couple of degrees would, at least according to Tol, have little effect.

    http://rabett.blogspot.com/2015/07/an-honest-well-for-eli-question.html

  154. I can’t remember of I posted to this one of Richard Telford’s post. It ends with

    That the global temperature has not had a linear trend over the instrumental period is not in the least unexpected as the climate forcing has not had a linear trend. It would seem a mistake to reject a linear trend model that nobody thinks is perfect in favour of an ARIMA(3,1,0) process that violates the first law of thermodynamics.

  155. Here’s another one. Essentially you could add a linear trend of 50oC over 150 years (essentially a straight line) and a statistical test would still suggest that a driftless ARIMA(3,1,0) was superior to a linear model. This is clearly nonsense, given that we specified an extremely large drift in the data.

  156. pbjamm says:

    So if I understand this correctly Tol and Keenan are saying that the linear trend Is the wrong tool to use because the data does not have a linear trend. No one is disputing that as everyone knows the trend is not linear but finding the linear trend is helpful in determining if the trend is up or down and no more, over a certain period of time. I imagine it is slightly more nuanced than that but this is my Big Picture understanding of the disagreement.

  157. pbjamm,
    That’s pretty much it, I think. I would describe what Tol has done here as a form of climate science denial (we don’t know if we’ve warmed, how much we’ve warmed, if it’s significant,…..)

  158. @pbjamm
    I agree that it is hard to see what people are so excited about. Baroness Verma said something dumb in Parliament. For reasons beyond me, people feel the need to defend the Baroness and attack the guy who pointed out just how dumb it was. InNo one disputes that temperatures now are higher than they were then.

  159. Richard,

    I agree that it is hard to see what people are so excited about. Baroness Verma said something dumb in Parliament.

    Baroness Verma isn’t the one saying dumb things.

  160. pbjamm says:

    If ATTP will allow it, can you please explain what was so dumb? I have read all you have had to say as well as everyone else and still do not understand what was wrong with the original statement. Is my understanding as stated above incorrect? If so how? Perhaps the issue is too far into the weeds for a layman but since proper science communication is so important to you I hope you will make an attempt either here or somewhere else.

  161. “So if I understand this correctly Tol and Keenan are saying that the linear trend Is the wrong tool to use because the data does not have a linear trend. No one is disputing that as everyone knows the trend is not linear but finding the linear trend is helpful in determining if the trend is up or down and no more, over a certain period of time. I imagine it is slightly more nuanced than that but this is my Big Picture understanding of the disagreement.”

    What we could say is the following.

    when you fit a statistical model to data you are making an assumption about the underlying
    data generating process. DATA doesnt have trends! the models applied to data are the “things” that have trends or more explicitly trend terms. data is just data. never forget that. The minute you apply a model you are saying “Suppose this is the underlying process that generated this data” how well do the data and model fit with each other. ?

    If you fit a linear model your are assuming that the underlying process is linear. With temperature we know that this model is non physical. That is, we know that at some time either in the past or the future a linear model will give answers that are non physical ( if there is a non zero trend).
    But for pragmatic reasons we may make an assumption and say over periods of X years
    a linear model is useful. So if I wanted to compare the trend at two sites I might say
    that using a linear model over a short time period to compare two sites will give me an answer that
    i can rely on. for example, site A has a higher trend than site B under an assumption of a linear data generating process.

    I hope this makes things more confusing. sorry about that.

  162. This is disagree with

    If you fit a linear model your are assuming that the underlying process is linear.

    This I agree with

    The minute you apply a model you are saying “Suppose this is the underlying process that generated this data” how well do the data and model fit with each other. ?

    They’re not quite the same. I think this distinction is important.

  163. @pbjamm
    “the temperature rise since about 1880 is statistically significant […] linear trend (°C/decade) […]
    0.062±0.009”

    I agree that arcane statistical arguments are not easy to follow. Our dear host demonstrates that a professor of astronomy cannot grasp the issue. That said, this is undergraduate material in economics and finance. Keenan’s objections are immediately recognizable to people in banks and pension funds — and environmentalists’ botched attempts to defend the undefensible only undermines their credibility among those who control the money.

  164. “Yesterday I was trying to implement something new in my code which needed to be recompiled over and over again, and so I had time to engage in this rather pointless discussion. Today, I have better things to do.”

    I thought I was the only one.

    I have found that when my code isnt working I get really nasty on the internet.

  165. This is disagree with

    If you fit a linear model your are assuming that the underlying process is linear.

    This I agree with

    The minute you apply a model you are saying “Suppose this is the underlying process that generated this data” how well do the data and model fit with each other. ?

    They’re not quite the same. I think this distinction is important.

    ##################

    I think I agree. my first statement is a bit over reaching.

  166. I note you still haven’t answered the question.

    Our dear host demonstrates that a professor of astronomy cannot grasp the issue.

    I would be more offended if it wasn’t for the fact that you’ve just spent the last 2 days illustrating your ignorance and sprouting all sorts of climate science denial. Bizarre.

    As if it isn’t obvious, the statistically significance refers to relative to there being no trend. As if that wasn’t obvious.

    environmentalists’ botched attempts to defend the undefensible only undermines their credibility among those who control the money.

    Ironic, coming from someone who is implicitly defending Keenan. Of course, you probably see nothing wrong with throwing around accusations of fraud. Another reason not to take you seriously.

  167. Eli Rabett says:

    Steve, in Eli’s experience in such cases you have to speak to your loved ones if they are still speaking to you.

  168. I have found that when my code isnt working I get really nasty on the internet.

    I actually quite like it when it’s not working, as it means I actually have to do things. When it’s working it just runs and take weeks. This means I’ve often forgotten why I started it, when it finishes 🙂

  169. @Richard

    @Steve
    Apologies. Signal theorists are pretty good at this stuff too, and there is lots of cross-fertilization between signal theory, statistical physics and finance.

    #####################

    no worries

  170. pbjamm says:

    @Richard Tol
    You have so far explained nothing in this thread and have only insulted the host and anyone who challenges your position. This in not the least bit helpful for someone trying to grasp the issue at hand. I have seen you in the past criticize climate scientists for their poor communication skills. You have demonstrated here that you are an expert at poor communication and I find your criticisms hypocritical at best.

    @Steven Mosher
    Thank you. I do not think that you made things more confusing. I guess I am confused about the point ATTP disagreed with. How does applying a linear model to data that is known to be non-linear mean you are assuming it to be linear? Can you not just use that simple test to gain some understanding of the data without assuming that it is complete? It seems that what you get out of the linear model/trend is useful if no where near the total story. I have seen it said frequently that all models are wrong but some are useful. That seems to me to apply here.

  171. Willard says:

    Here’s Jonathan Giligan’s comment at Andy’s:

    Over at Eli Rabett’s blog, the lagomorph asks a very important question on top of all these others: In aggregating the data, did Tol correctly ensure that the “warming” numbers all referenced the same baseline? “The serious question is what is the baseline for each study and which sets the zero of the temperature scale? … For which is the baseline pre-industrial (in which case the world is now past Tol’s outlying positive point at 1.0 C), 1860 or so when instrumental records start in which case the world is pretty close to it, or some more recent time, in which case, another couple of degrees would, at least according to Tol, have little effect.”

    Brandon Shollenberger answers that “Tol didn’t actually use any real baseline. All the papers used different temperature baselines, and Tol didn’t put them on a common one!” (more detail here at Shollenberger’s blog)

    http://andrewgelman.com/2015/07/23/instead-he-simply-pretended-the-other-two-estimates-did-not-exist-that-is-inexcusable/#comment-229684

    Is it true you did not actually use any real baseline, RichardT?

    ***

    Jonathan used to comment at Keith’s. He was my main reason to read Keith’s.

  172. pbjamm,
    I think Steven and I are in agreement about the first statement.

    insulted the host

    To be fair, I’ve insulted Richard too.

  173. @Steve
    Your second statement, the one you agree with, is of course the correct phrasing.

  174. John Hartz says:

    ATTP:

    Dang it! Now you have forced me to go to the store for more popcorn. 🙂

  175. Your second statement, the one you agree with, is of course the correct phrasing.

    Yes, it is. It’s roughly what I’ve been saying for two days now. Of course, you’ll now find some reason to dispute that. That’s because spreading misinformation, doubt, climate science denial, and promoting any old trash, is your modus operandi.

    Now, I’m asking nicely, can you please go and spread your misinformation elsewhere? It’s really tedious and irritating and I really don’t like my site being used to spread this kind of garbage. As I said above, there are plenty of climate science denial sites that would welcome you with open arms and where you could feel like more of a kindred spirit.

  176. pbjamm says:

    Yes ATTP, we crossed there. I still have no idea what Tol is criticizing here and he seems not to have an interest in explaining it in terms a layman might understand. I think you had it correct earlier when you referred to him as savaging a strawman, not for the first time here on your blog.

  177. pbjamm,
    I think the term is FUD.

  178. pbjamm says:

    Indeed. This is all very reminiscent of Cook/97% argument with Tol a few months back. A subject that NO ONE WANTS TO REVISIT. I will let it drop now since I feel your point is made and your antagonist has failed to to explain his position in any meaningful way. Sorry for revving this up again.

  179. Hyperactive Hydrologist says:

    “All this is because stationary and non-stationary statistics are quite different, and procedures that are designed for stationary data come apart when applied to non-stationary data.”

    Climate impact scientists are well aware of this especially in analysis of rainfall. See this paper as an example: http://dx.doi.org/10.1002/joc.3669

    I also don’t think most academic are in academia for the money. I am sure some of them could quite easily get high paid jobs in finance but instead they choose to do some thing useful with their lives.

  180. @wotts
    Perhaps should not have stopped at “The minute you apply a model you are saying “Suppose this is the underlying process that generated this data” how well do the data and model fit with each other?” but added “if the fit is not good, perhaps you should not take the model results too seriously (and definitely not present them in Parliament)”.

  181. Richard,
    I’m asking nicely. Go and spread your mis-information somewhere else, please.

  182. @HH
    Sure. Hurst gave us long memory.

  183. John Hartz says:

    Threads like this play into the deniers’ hands because they divert attention from what’s going within the Earth’s total climate system. A single metric of annual changes in the temperature of the lower tropsphere does not even reflect all of the spatial and temporal anomolies that occur in the lower troposphere in a given year — to say nothing of reflecting what’s happening in the other components of rhe climate system.

  184. Eli Rabett says:

    Eli would still like to know what Richard did with the baseline shuttle in Tol 2009 and elsewhere.

  185. Serious question: If we want to know how much the world has warmed between 1880 and 2015, shouldn’t we just subtract the temperatures of 2015 and 1880? The trajectory of the in-between years isn’t really relevant, is it? One could of course argue that there’s some measurement uncertainty in individual years, but to overcome that you could subtract decadal averages instead.

    One could also argue that there is autocorrelation in the timeseries, so a decade might be biased high or low. But with that argument you’re again hip-deep in assuming a model that generated the data.

    Is there anything wrong with my line of thought?

  186. Bouke,

    Serious question: If we want to know how much the world has warmed between 1880 and 2015, shouldn’t we just subtract the temperatures of 2015 and 1880? The trajectory of the in-between years isn’t really relevant, is it?

    The problem is that this doesn’t take into account the natural variability in the data. However, if the time interval were long enough (decades) what you suggest would give you a reasonable estimate, but without any uncertainty estimate. If you wanted to do something a bit better than this, you could do linear regression with uncorrellated errors. That would give you a slightly better estimate. Given that the data isn’t uncorrelated, you could do what is commonly done, which is linear regression with auto-correlated errors. That would give you a trend with a reasonable estimate of the uncertainties in the trend.

    Beyond this, you have to start doing proper comparisons with physically motivated models. This, however, is more related to trying to understand what is driving the warming, than trying to estimate how much warming we’ve experienced.

  187. MarkB says:

    Keenan’s claim is that Baroness Verma’s response to the question “on what basis they assert that there has been a long-term upward trend in average global temperatures” is based on a flawed model and he proposes a different (better?) model. Having spent looking at the details he’s presented in various places, I’ve not seen something similar to the “trend +/-uncertainty” estimate that comes out of linear regression. To be clear I understand that he’s applied some criteria to the respective models and declared a winner and I understand that the “better” model may be unphysical,but has he actually provided an answer to “What is the central estimate and uncertainty of warming over the instrument period?”

  188. but has he actually provided an answer to “What is the central estimate and uncertainty of warming over the instrument period?”

    No, I don’t think he has. As I understand it there are a number basic goals. To suggest that the Met Office’s model is wrong. To suggest that we can’t say anything significant about the temperature trends. To, consequently, suggest that we know nothing about the rise in temperature since the late 1800s. What Keenan doesn’t realise is that just because he doesn’t understand this and can’t answer these questions, does not mean that noone else can do so. For reasons I don’t fully understand, Tol is aiding in this mis-information campaign.

  189. BBD says:

    ‘The Greens’ made him do it.

  190. Roger Jones says:

    This thread certainly warrants a bowl of popcorn or two but I came in on it late.

    ATTP, you are certainly correct when you say the linear trend is a method, but I also agree (conditionally) with Steven Mosher’s first statement that he retracted:

    “when you fit a statistical model to data you are making an assumption about the underlying
    data generating process”

    This is incorrect, but nevertheless, many people do assume this and will have an idea of the physical process informing it or will form one by mentally processing a statistical description of the change. There are many, many papers in climatology where the method is defended as if it were the process – and these are by some of the most notable scientists working today. There is also an industry in contrarianism that gainsays the theory by arguing that climate change does not trend in the manner scientists say it does, so the theory is wrong.

    The assumption that most people – and working climatologists have – is that the warming process is gradual but irregular and mediated by climate variability which may be non-linear and is chaotic. I disagree with important aspects of this model but have no problem with linear trends being used over long timescales (of about a century) as applied by the Met Office as a descriptive method.

    At the same time, ATTP’s point about descriptive and inferential statistics is well made and people tend to forget this all the time. I do think that the IPCC WGI should state this a little more often, because the relationship between socially-constructed models of climate change and theoretically-informed models needs to be emphasised clearly and often.

    The unfortunate thing about all of this is that future climate is projected on decision-making timescales of decades as if it were linear, and this is incorrect. All economic models based on this are doubly wrong because they treat both the economy and climate as quasi-linear systems.

    But to treat this as a problem that statistics can solve, without being informed by the underlying theory, is just plain dumb.

  191. Roger,
    Thanks. I actually only objected to this statement of Steven’s (which, I think, he withdrew)

    If you fit a linear model your are assuming that the underlying process is linear.

    whereas this statement

    when you fit a statistical model to data you are making an assumption about the underlying data generating process

    is somewhat weaker. I agree with what I think you’re saying; even if you aren’t assuming that the underlying process is linear when using a linear model, you are assuming that a linear assumption will give you a reasonable descriptor of the dataset.

    But to treat this as a problem that statistics can solve, without being informed by the underlying theory, is just plain dumb.

    Exactly.

  192. Roger Jones says:

    ATTP, ta. I am actually saying something stronger, which is because the linear models is used so much and relied upon so strongly, that people, including scientists, fall into the trap of seeing the underlying system as linear.
    Certainly, climate-related decision-making often assumes that – if you pull someone up and say “You know the system is inherently non-linear,” they will say “yes, I know”, and if you say “why don’t you make your decisions based on that?” they might respond “Well, we don’t have the tools to do that, so we use the tools we have”. Those tools are based on climate trending into the future.
    This is a real problem, and with the work we do on decision making for adaptation, we frequently use scenarios of rapid change that shock a system, and explore what the ramifications of that are.

  193. Roger,

    I am actually saying something stronger, which is because the linear models is used so much and relied upon so strongly, that people, including scientists, fall into the trap of seeing the underlying system as linear.

    Okay, yes, I see what you’re getting at and I agree.

  194. The problem is that this doesn’t take into account the natural variability in the data.

    Does data have natural variability? Doesn’t that presuppose a model?

  195. Boeke,

    Does data have natural variability? Doesn’t that presuppose a model?

    Oh, I just meant that the data clearly isn’t simply a straight line. Hence, you don’t know if the first and last points are really the best representations of a best-fit trend.

  196. anoilman says:

    Bouke van der Spoel: For instance in 1998 we had a record 2C hotter El Nino. If that was your end point you’d reach alarming conclusions. If you did it in 2000 you’d reach different conclusions.

    There has been a lot of work in this field.

    http://www.skepticalscience.com/graphics.php?g=52

  197. Tapani L. says:

    Reading this thread kinda makes me have second thoughts about my current undergraduate studies majoring in statistics… especially being a physics dropout from long before.

    In a nutshell, when Richard says

    “Baroness Verma presented a statistical analysis in Parliament. She wrote “0.062±0.009″. Both numbers are wrong: It is a biased estimate of the linear trend, and it is strongly biased estimate of the standard error.”

    it’s clearly correct in statistical sense. No argument. If data don’t really fit your statistical model’s assumptions, then the statistics you get out of it are suspect (biased). But then a physicist might say, “so what, the numbers describe the situation well enough to have a sense of the change over the instrumental period”, and go back to modelling from first principles – the nature follows physical laws and energy conservation bounds possible temperatures.

    So I think ATTP and Richard were talking past each other and Richard’s cryptic comments that mostly had something to do with obscure (to others) time series analysis were just more fuel into the fire. Only a kettle and a bag of popcorn was needed… and I was late to the party – but earlier I did get a chuckle over lunch from Richard’s comment about the ARIMA(N,1,P) model being wrong over long timeframes since Earth’s surface temperature is physically bounded by absolute zero!

  198. John Hartz says:

    Here’s the reality that the human race is facing:

    The climate system, in all its infinite complexity, is impossible to predict entirely.

    There are some things happening that scientists don’t completely understand yet, such as why ice floating on the sea around Antarctica is currently growing slightly. Scientists think, perhaps counter-intuitively, that it’s down to climate change, too, as the winds encircling the continent push freezing water outward from the coastline, extending the icy platform offshore.

    And the climate system could still hold some surprises. As the Arctic warms, the once-frozen ground is thawing and releasing the powerful greenhouse gas methane. Scientists are unsure yet just how much 4C of global warming could speed up this process.

    Two, three and four degrees are all points along a global warming continuum. None represents a climate precipice, but it’s clear that as the temperature rises, so do the risks.

    What’s left to decide is, how much of a chance are we willing to take? The science is solid enough that whatever we choose, we can’t tell future generations that we didn’t know the risks.

    The reality of global warming: We’re all frogs in a pot of slowly boiling water by Roz Pidcock, The great Debate, Reuters, Aug 7, 2015

  199. T-rev says:

    >ATTP: Doug Keenan is largely clueless about this topic, his views are simply wrong,

    Professor Dunning has term for this “Confident Idiot”
    http://www.psmag.com/navigation/health-and-behavior/confident-idiots-92793/

  200. T-rev says:

    >JH: The downside of this OP and others like it is that you give climate science deniers like Doug Keenan free publicity. An upside is reading the absurd comments made by Richard Tol and the responses to them.

    This 🙂 As to people who are worth more being paid more, that seems compelty at odds with real life observations. I am going with Dave Graber’s take on it.

    In terms of employment, a persons “worth” is in how upset other people are if they don’t turn up. Tube workers don’t turn up, Chaos. Financial time series analysts don’t turn up ? Would anyone even notice.

    If you’re basing it on money, surely Johnny Depp etal are the real deal 🙂 Leonardo diCaprio is the person we should be getting advice from.

  201. @Tapani
    If you model temperature as a random walk, ARIMA(0,1,0), then there is nothing to stop the model from going below 0K, and if you wait long enough it will.

    @Others
    If you model temperature as a linear trend, you have a similar problem.

  202. izen says:

    Here is a paper specifically investigating, because of its complete failure to predict the economic collapse of 2007, the development of econometrics in the decades since its adoption of the mathematical formalisms of time series analysis.
    Perhaps this might explain why a linear trend is rejected as wrong and the basis for a personal attack on another scientists.

    http://www.econ.qmul.ac.uk/papers/doc/wp669.pdf
    “Wide conviction of the superiority of the science method (scientisation) has converted the econometric community largely to a group of fundamentalist guards of mathematical rigour and internal consistency. It is often the case that mathematical rigour is held as the dominant goal and the criterion for research topic choice as well as research evaluation, so much so that relevance of the research to business cycles is reduced to empirical illustration. To that extent, probabilistic formalisation has entrapped the econometric business cycle research in pursuit of means at the expense of ends. It is thus not unforeseeable that those studies have failed to generate any significant breakthrough in predicting and explaining business cycles in the real world.”

  203. Richard,

    If you model temperature as a random walk, ARIMA(0,1,0), then there is nothing to stop the model from going below 0K, and if you wait long enough it will.

    If you did this, what would it tell you that the original data couldn’t tell you without you having done such an analysis in the first place? I already gave you a clue as to the answer.

    If you model temperature as a linear trend, you have a similar problem.

    Except – as has been explained numerous times already – linear trend analysis isn’t really trying to model temperatures. It is answering the question “what is the best-fit linear trend” or (as Steven put it) “if we assume that the underlying model is linear, what do we get?”.

    Now, I’m asking really nicely that you go and spread your misinformation elsewhere. There are plenty of places where you can do so.

  204. John Hartz says:

    Richard Tol:

    How many Brownie Points do you get in Deniersville for being banned from commenting on this website?

  205. Eli Rabett says:

    Richard Tol:

    If you model temperature as a random walk, ARIMA(0,1,0), then there is nothing to stop the model from going below 0K, and if you wait long enough it will.

    As Boltzmann was rumored to have remarked, you should live so long.

    Of course, since all models have limits of application, idiots who try and use them outside those boundaries have bad hair days.

  206. MarkB says:

    If you model temperature as a linear trend, you have a similar problem

    But the context being addressed is “What is happening with the instrument era data we have in hand.” It is not claimed to be a projection into the future,

  207. John Hartz says:

    ATTP:

    Perhaps it’s time for you to unfurl the Do Not Feed the Tol! banner.

    Tol’s “I’m right! You’re wrong!” messaging has become rather boring.

  208. Kevin O'Neill says:

    Eli writes: “Of course, since all models have limits of application, idiots who try and use them outside those boundaries have bad hair days.”

    LOL. No more coffee for you this morning, Eli. I think you’ve had enough caffeine 🙂

  209. Willard says:

    Last year, I tried to make coffee with my random physics expresso machine.

    I’m still waiting.

  210. John Hartz says:

    Willard:

    You should invite Richard Tol over to stir the pot. 🙂

  211. anoilman says:

    In a very circuitous way, Tol could be right, it’s just that its difficult to unravel.
    http://www.desmogblog.com/2015/08/03/richard-tol-s-gremlins-continue-undermine-his-work

  212. Howard says:

    Shorter Tol: All models are wrong and I’ll be damned to find any useful.

  213. redbbs says:

    Richard Tol’s greatest strength is his inability to be embarrassed.
    The University of Sussex’s greatest weakness is its inability to be embarrassed by Tol’s conduct.

  214. The University of Sussex’s greatest weakness is its inability to be embarrassed by Tol’s conduct.

    I’m not sure that that is fair. I don’t think there is any evidence to suggest that the University of Sussex is not embarassed by Richard’s conduct.

  215. redbbs says:

    Sussex’s has had many years to solve its problem.

  216. Sussex’s has had many years to solve its problem.

    Not really true, plus academic freedom. Let’s not start making suggestions about people’s employment. This is just a blog. I would defend Sussex’s right to employ anyone they want, and that person’s right to express their views freely. That I might disagree strongly with that someone says, does not mean that they should not say what they wish (within the law, that is).

  217. redbbs says:

    Stuff like this is within the law but does nothing to help Sussex with its very low ranking.

    https://andthentheresphysics.wordpress.com/2013/06/02/watt-about-richard-tol/#comment-453

  218. Steven Mosher says:

    Wow.

    Check out the title of this post.
    Read the comments.

  219. Steven,
    I’m guessing you’re complaining about supposed attacks on Richard Tol? If so, you may have a point. However, I think you would be hard pressed to convince me that it isn’t deserved.

  220. izen says:

    @-Steve Mosher
    It is surprising to see the attacks on Met office scientists defended by a world leading econometrician ( according to Tol) on the grounds that a linear trend is a false description of the data.

    Tol-“But you can’t pin this on Keenan. HM Government issued a statistical statement based on a physically impossible model, and Keenan responded with superior statistics (and superior physics, by the way, as Keenan violates in second moment rather than first).”

  221. Joshua says:

    ==> “However, I think you would be hard pressed to convince me that it isn’t deserved.”

    The title of the post refers to personal attacks. Harsh criticism of arguments made…pretty easy to justify. Personal attacks, IMO, not so much.

  222. izen,

    and superior physics, by the way, as Keenan violates in second moment rather than first

    I’d missed this comment by Richard. Bizarre. There’s no physics in Keenan’s work, superior, or otherwise.

  223. izen says:

    Apologies if this seems of-topic, but I am trying to find an analogous example where we use the linear description of the relationship between two variable to derive useful information even though that relationship is not a simple linear process.

    It is possible to record how many miles you drive and how much fuel you use in any car you drive. The result is a figure for the miles per gallon for that vehicle for that distance.

    But modern vehicles can give a continuous readout of the mpg, this reveals that the specific conditions affect the mpg, it is not a linear relationship between fuel and distance, many other factors influence the value and these change continuously with distance traveled. There is a tenuous metaphorical link between climate sensitivity and mpg…

    As a result the useful display that modern cars often have that gives you the distance or range left with the fuel available is based on a physically impossible model of fuel consumption and distanced traveled as it is assumes a simple linear relationship based on the past average mpg. It does a simple extrapolation of a past trend.

    Despite the contravention of the mathematical formalisms of time series analysis that this gross and physically unrealistic linear averaging of mpg provides, it does not seem to be the target of ideologically driven criticism accusing the motor manufacturers of fraud and lying to the public.

    And despite the formal shortcomings of the underlying model of fuel consumption implicit in the range left display in modern cars, most people seem to find it a useful feature, not mathematically invalid and intentionally misleading information that justifies accusations of incompetence and fraud on the part of the car manufacturers and their scientists.

  224. John Hartz says:

    ATTP:

    Richard Tol’s modus operandi is to bait commenters into responding with personal attacks. As long as he is not banned from commenting, he usually succeeds. Politely asking him to post comments elsewhere seldom succeeds.

  225. Andrew Dodds says:

    izen –

    Worse still, those ‘Miles left’ readouts are deliberately misleading – they are designed to considerably under-read, so as to guarantee that if it says 30 miles left and the next petrol station is 30 miles away, you’ll make it. They are basing the readouts on the precautionary principle, because the potential damage from over reading (saying 40 miles left when you only have 30) is much worse than the damage from saying 30 miles left when you have 40.

  226. Don Gisselbeck says:

    So we can continue to pump CO2 into the atmosphere with impunity. This is true because how much statisticians make tells us how skilled they are. Good to know.

  227. Eli Rabett says:

    Does anybunny realize that what Keenan is claiming as state of the art are the damn chartists, an empty barrel if there ever was one. Here is a lovely taste

    http://www.ft.com/cms/s/0/2f325d8a-4d90-11e1-bb6c-00144feabdc0.html#axzz3iYvQKqdF

  228. I have to thank you for letting Richard Tol reveal himself in all his (in)glory. The fact that he seems to regard income (money) as the highest good and greatest signifier of quality explains a lot. It explains a lot more than he would have liked to reveal. Fact is, he has rather a nice job, and he appears to have bungled quite a bit lately. So if his income is as he asserts the exclusive signifier of his value, he need not worry about that.

    But money isn’t everything, particularly when we have a rather intense question in hand about planetary survival. So he should stop worrying about his own standing and worry about the standing of the human community on their home planet, and himself as a member not standing above or apart from that community, but needing to make common cause with the rest of us in making sure that we don’t destroy that planet’s habitability.

    It might be too late, but there is rather a strong moral element here, and no amount of income will compensate for being a contributor to making it not be so.

  229. Brian Dodge says:

    Professor Tol said “… and it is strongly biased estimate of the standard error.”
    I’m guessing, not being a mathematician, that he is implying a rather greater uncertainty in temperature trends than the Met office thinks. Since damage done by flooding from extreme rainfall or storm surge, and crop losses from higher than optimum temperatures are high order polynomial functions, plus there is the acceleration of ice loss from Greenland and Antarctica, this misunderestimation of the range of future temperatures has serious policy implications. My lay understanding of risk analysis is that the risk is based on the probability function,times the damage function, and if the probability upper bound of warming is being underestimated, then the risks are much higher, and the need for drastic action much more urgent than currently though.

    If skeptics like Professor Tol have reached these conclusions, perhaps joining the Catholic Church in prayer on September 1st is in order, even for agnostics like me.

  230. lerpo says:

    New York University economist Paul Romer recently complained about how economists use math as a tool of rhetoric instead of a tool to understand the world. – http://paulromer.net/mathiness/

  231. Pingback: Richard Tol’s climate sensitivity estimate | …and Then There's Physics

  232. Sam taylor writes:“Any discipline which uses the hodrick-prescott filter in anger (hello economics) is, to my mind, less than clueless when it comes to time series analysis. “

    Not all economists, Sam. James Hamilton has written Why you should never use the Hodrick-Prescott filter

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s