NAS Panel deja vu

I wrote earlier about McIntyre’s attack on the NAS Panel on temperature reconstructions. McIntyre objected to two panelists because they were co-authors of co-authors of Mann, but not to the panelist who was a co-author of a co-author of McKitrick. In another post he also objects to another panelist, Kurt Cuffey, because Cuffey wrote:

Mounting evidence has forced an end to any serious scientific debate on whether humans are causing global warming. This is an event of historical significance, but one obscured from public view by the arcane technical literature and the noise generated by perpetual partisans.

McIntyre claims that this demonstrates prejudice, but Cuffey does not include the Hockey Stick as one of the pieces of mounting evidence. The only thing that Cuffey writes about temperature reconstructions is this:

Reconstructions of past climates, on timescales of millennia to millions of years, demonstrate that small changes in climate influences (like greenhouse gases) cause significant climate changes. And the magnitude of the changes is in the mid- to high range of predictions from the best climate models (published October 2004).

Obviously he’s not referring to the Hockey Stick, which shows little change in temperature before the 20th century.

However, it could be argued that one of the panellists has made up his mind about the Hockey Stick. That panelist is John Christy who reckons that the Hockey Stick was broken by Soon and Baliunas:

The conclusion in IPCC 2001 that human induced global warming was clearly evident was partly based on a depiction of the Northern Hemisphere temperature since 1000 A.D. This depiction showed little change until about 1850, then contains a sharp upward rise, suggesting that recent warming was dramatic and linked to human effects.[3] Since IPCC 2001, two important papers have shown something else.[4] Using a wider range of information from new sources these studies now indicate large temperature swings have been common in the past 1000 years and that temperatures warmer than today’s were common in 50-year periods about 1000 years ago. These studies suggest that the climate we see today is not unusual at all.

McIntyre, of course, makes no objection to Christy’s presence on the panel.

McIntyre is also unhappy with another panelist

[Bloomfield] is cited in two pers. comms. in Briffa et al [Holocene 2002] where Briffa describes how they went about estimating confidence intervals for their MXD reconstruction – you know, the one where they chop off the period after 1960. Out of all the statisticians in the world, why would they pick one who consulted on confidence intervals for one of the Hockey Team studies?

Ummm, because he knows something about the statistics of reconstructions?

I’ve seen all this before. Back in 2001, John Lott made similarly specious attacks on the NAS panel on firearms research, claiming that it was biased. You see, there is nothing to lose in making such criticisms. If the panel comes down against you, you can dismiss their findings as biased (this is what Lott ended up doing). And if they come down on your side, you can present their findings with: “Even though they were biased against me …”

Update: John Fleck finds a coauthorship chain connecting McIntyre with Mann: McIntyre – McKitrick – Michaels – Christy – Pielke Senior – Mann. OK, Lambert – Stanton – Stinson – Erdos.

Comments

  1. #1 James
    March 4, 2006

    Chris:

    “McIntyre also has the idea that there is no formal rule for deciding how many PCs are needed. Yes, I’m sure he’s looked for one.”

    Which rule do you prefer, Chris? Scree test? Eigenvalue rule? Preisendorfer?
    There is no formal rule for retention of PCs (or otherwise stated, there are several).

  2. #2 Spence_UK
    March 4, 2006

    Chris,

    Thanks for another fascinating series of mischaracterisations on McIntyre’s arguments. Nice of you to throw in a logical fallacy as well – I’ll get to that later. The key that I’ve been trying to describe here, is that the multi-proxy methods are not robust, either to data or variations of method. A statistical method that is not robust should not be relied upon.

    Anyway, on to your points.

    If you argue both centred and uncentred PCA are valid, then the reconstruction must be robust to this decision. But, as shown by both McIntyre, Burger and Cubasch, it isn’t. The problem isn’t so much uncentred PCA is fundamentally wrong, the problem is that it has consequences for the robustness of the results.

    You then go on to claim that McIntyre doesn’t think there is a formal method for selecting PCs. This is simply false. McIntyre has talked about a number of formal methods for selecting PCs, but points out none of them tie in with the PC roster used by Mann. MBH claims to use Preisendorfer’s “Rule N” – discussed extensively on ClimateAudit (so how you can claim McIntyre doesn’t know about formal rules just shows how unaware of his arguments you are) but this step cannot be reproduced either. (The code W&A made available simply read off the PC roster archived at Mann’s SI, so they haven’t even tried to replicate this step).

    On his blog, McIntyre provides references to allow people to find out about PC selection criteria such as the Kaiser-Guttman criterion; bootstrapped Kaiser-Guttman; Scree Plot; Broken-stick; Proportion of total variance; Sphericity test; Bartlett’s test of the equality of ?1; Lawley’s test of ?2, bootstrap eigenvalue-eigenvector. Hardly indicative of someone who knows nothing about PC selection. I thought “Broken-stick” was a particularly aptly named method.

    “Fallen apart meaning 0.5 deg C from the original” – that’s enough to show lack of robustness, but isn’t the most important aspect. By changing these parameters (which shouldn’t matter), you get a wide range of different values for statistical skill; by cherry picking one with high significance, you make the results appear more meaningful than they actually are. Unfortunately this cherry picking can be applied to the cross validation statistics as well as the calibration period, turning the cross-validation into another calibration, further undermining the results.

    “One of the advantages (if you can call it that) of the 20th century for calibrating proxies is that the 20th century covers such a wide temperature range” – This is an example of inserting your conclusions into your assumptions. This is the logical fallacy of affirming the consequent. Find out more here:

    http://en.wikipedia.org/wiki/Affirming_the_consequent

    By assuming the 20th century covers such a wide range of temperatures, of course you’ll find that the 20th century had the warmest period of it. But that doesn’t teach you anything about the truth.

    “As far as I know the paleo-proxies are always below the highest temperature values occuring in the calibration period” – no they’re not. Read Burger and Cubasch. Just the abstract explains this much. Remember proxies indicate local, not global temperature. Nobody else (that I know of) is claiming temperatures have never been this high in some localities.

    “Nice to be reminded of the bleeding obvious.” – it’s a shame you had to be, by suggesting “When will the contrarians move on to arguing the thing that really matters, i.e. that we have the right result now?” – clearly showing you want the science to stop. I don’t think we have the right result now, and from their statements, I believe von Storch, Burger, Cubasch, McIntyre don’t either. Some obviously do think we have the right result. The debate – and the science – will continue.

  3. #3 Ian Gould
    March 4, 2006

    Tim Curtin: nuclear energy… the proven most cost-effective way to reduce CO2 emissions whilst maintaining living standards,

    Keep asseting thatm Tim. Eventually you may convince someone, if only yourself.

  4. #4 Tim Lambert
    March 4, 2006

    Tim Curtin, in the passage that you wrote and I quoted above, you calculated percentage changes in Celsius. You weren’t doing to object to something you now allege the IPCC is doing — you did it as part of an argument that a small percentage change in atmospheric compostion could not cause a large percentage change in Celsius temperatures. You’re not helping your credibility by now claiming that you’ve known all along that this was invalid.

  5. #5 Spence_UK
    March 4, 2006

    Just an extra note.

    I didn’t bother reading Chris’ link to “a tutorial” on PCA, because I’m quite familiar with how PCA works. After posting, and out of morbid curiosity (I’m a sucker for punishment), I thought I’d take a peek. (NB. You have to knock out the space in the link for it to work properly).

    Scanning quickly through it, it is a simplistic – but not bad – introduction to PCA; a bit low-level for someone with reasonably strong maths skills, but well presented for someone who doesn’t. What made me laugh was that in the document it states (pg 12, 3.1 step 2):

    “For PCA to work properly, you have to subtract the mean from each of the data dimensions.”

    I’m not saying I necessarily agree or disagree with this point, the real answer is, as ever, rather more complex than that, but you are using a reference to demonstrate one point that actually disagrees with an earlier point you make.

    Is that ever a wise move in a debate?

  6. #6 Eli Rabett
    March 4, 2006

    The temperature of the Earth without an atmosphere is 256 K and the temperature with the atmosphere. The current global average temperature is ~288 K. That is a difference of about 32 K. Meaningful percentage changes should be stated relative to that figure, i.e. a change of 3K is about 10%. Ice ages were globally about 7K lower, i.e. ~20% lower

  7. #7 Tim Curtin
    March 4, 2006

    Ian Gould: I am sure you know the cost of electricity generated by nuclear energy varies by plant depending on age and type and discount rate for valuing the capital cost component. The largest all-in (including capital) costs of 14 nuclear power plants cited by the OECD’s NEA is one in Japan, at US$70 per kWh with a discount rate of 10%; One in Canada comes in at less than US$40, and there’s one in Korea at US$30. Fuel costs alone were much less of course, as little as US$3 per MWh in the Korean case (but would be appreciably higher at today’s costs of uranium). The OECD’s NEA paper by Bertel and Morrison (“Nuclear Energy Economics in a sustainable development perspective”) shows that at a 5% discount rate the all-in generating cost falls to as little as US$25 per MWh for the Canadian plant for which the fuel cost is US$9.

    Data from US Utility Data Institute on nuclear energy costs in USA from 1981 to 1999 are in a range between US$20 and US$30 per MWh and by 1999 were just below the comparable cost of coal fired power at US$21, and well below oil and gas, in constant 1999 US$.

    Australia’s NEMMCO publishes average electricity prices which can fluctuate widely in the course of a week, eg first week in February, from as little as A$15 per MWh at weekends (US$11) to as much as A$57.48 (US$42) on weekdays (NSW, 1st Feb 06). Other things equal, it would appear that nuclear power could well be competitive even with coal in Australia.

    This conclusion was independently confirmed by Martin Sevior (School of Physics, Univ of Melbourne) in the Australian Financial Review as recently as 1st March 2006: “Nuclear power now an affordable option”.

  8. #8 James
    March 7, 2006

    Those that recall my dust up with Dano on the r2 statistic in MBH98 might like to have a look at this:

    http://www.climateaudit.org/?p=564#comments

  9. #9 Chris O'Neill
    March 7, 2006

    I wrote:

    “Nice to be reminded of the bleeding obvious.”

    to which Spence_UK replied:

    “- it’s a shame you had to be, by suggesting “When will the contrarians move on to arguing the thing that really matters, i.e. that we have the right result now?” – clearly showing you want the science to stop.”

    When I was talking about moving on I meant moving on from the method of MBH98. The thing that really matters is getting the right result regardless of the method used. I can’t see how this means that I want the science to stop, even though apparently it’s clear to some people that this is my meaning.

    Regarding Burger and Cubasch, Mann says they didn’t study their most up to date methods as in Proxy-Based Northern Hemisphere Surface Temperature Reconstructions: Sensitivity
    to Method, Predictor Network, Target Season, and Target Domain
    .

  10. #10 Spence_UK
    March 7, 2006

    “The thing that really matters is getting the right result” – exactly, and B&C have shown that the current results are not reliable enough to be called the “right result”. But even when we find a result which actually holds water, there is always the potential for that to be overturned in the future. That is what science is all about. Keep questioning what we think we know.

    I haven’t looked at the REGEM method in great detail yet (I’ve downloaded the MATLAB, fixed the errors that prevented the archived code from running, but not yet analysed it in much detail).

    The REGEM method certainly seems to address one problem of PCA, in that PCA demands that there is no missing data, and some process must be put in place to handle this.

    But this isn’t the problem identified by Burger and Cubasch. Their main concern is that outside of the calibration interval “temperature” extent, there is no guarantee that the proxy data remains linear and even if it does, the extrapolated error grows with the distance outside of the calibration interval. I don’t see anything in the paper you cite which addresses specifically that concern. If you can point me to a particular paragraph that you feel addresses this point, that would be useful.

  11. #11 Dano
    March 7, 2006

    Yeeees, james, I said it was .02.

    What’s your point? that a blog wants to keep hammering an old paper? How is that news?

    D

  12. #12 Chris O'Neill
    March 8, 2006

    “B&C have shown that the current results”

    According to Mann, B&C don’t test all the current results.

    “But even when we find a result which actually holds water, there is always the potential for that to be overturned in the future.”

    Yes, that’s the perfect climate skeptic argument. Just say the results showing we have a problem could always be overturned in the future so just ignore them.

    “Burger and Cubasch’s main concern is that outside of the calibration interval “temperature” extent, there is no guarantee that the proxy data remains linear and even if it does, the extrapolated error grows with the distance outside of the calibration interval.”

    You’d be best to ask Mann or his colleages if they think this is still a problem but it appears that the main problem with extrapolation error, both linear and non-linear, is for temperatures colder than the calibration range, as shown by figure 7 in Burger, Fast and Cubasch’s “Climate reconstruction by regression – 32 variations on a theme”. In this figure there are very few reconstruction proxy values above the warm end of the calibration proxy values. So the reconstruction proxy values near the warm end would generally be quite well covered by the calibration proxy values. So the reconstruction would not be particularly sensitive to extrapolation error near the warm end. As long as uniformitarianism applies, the warmest reconstructed temperatures should be as accurate as the reconstruction proxies allow. Of course, uniformitarianism might not apply and that’s another issue.

  13. #13 Spence_UK
    March 9, 2006

    “that the main problem with extrapolation error, both linear and non-linear, is for temperatures colder than the calibration range” – that statement is not correct in the non-linear case. If the response is a quadratic (“inverted U”) for example – certainly an appropriate model for tree rings – warmer temperatures could appear below today’s levels. There are plenty of other proxies that exceed 20th century values by a large margin, but those proxies seem to be less likely to be selected in paleoclimate studies. Unfortunately justification of proxy selection criteria is rarely laid out and appears to be a little “ad hoc”, which should be anathema to statisticians.

    “According to Mann, B&C don’t test all the current results.” – B&C question an assumption at the heart of most of the multiproxy studies, whilst it is true some of them may not be affected, at present it is not clear which are and which are not. Perhaps if we could see some of the data that we could work it out more easily.

    “You’d be best to ask Mann or his colleages if they think this is still a problem” – I have tried asking questions on RealClimate in the past and, mysteriously, they never appear. I gave up bothering some time ago, as it is not worth the time taken to compose the questions. I’ll find time to do a bit more digging into the RegEM procedure. I can see the merit in the RegEM procedure (procedural mechanism and error model for missing values) but at a top level I just can’t see how it addresses the issues raised by B&C.

    I’m not sure what you mean by “Just say the results showing we have a problem could always be overturned in the future” – I’m not sure what “problem” the paleoclimate reconstructions show we have. I’m interested in the detail of the science, not the broad brush conclusions people try to draw from it (which are more often than not politically motivated, on both sides of the debate).

  14. #14 Chris O'Neill
    March 16, 2006

    “If the response is a quadratic (“inverted U”) for example”

    You have to assume worse than that. You have to assume non-uniformitarianism because if such a response actually occurs it should also occur in the warm times of the twentieth century. In nearly all of B&C’s alternative reconstructions the middle of the twentieth century has average proxies reaching higher values than any other time back to 1400. If there really was an inverted U response and if it really was warmer in the past then the proxies in the past should at least reach the middle twentieth century average before turning down.

    “- certainly an appropriate model for tree rings”

    Care to give a reference?

    “There are plenty of other proxies that exceed 20th century values by a large margin”

    If that’s true then they’re not showing an inverted U response. Is there any evidence of significant inverted U response? One thing about the climate is that because of its variability, even if it was warmer on average around 1400, there would still be some years in the 20th century that were substantially warmer than most of the years around 1400. Any undetected non-linearity could only affect the small fraction of the years in the reconstruction that could have been warmer than the warmest years in the twentieth century.

    “I have tried asking questions on RealClimate in the past and, mysteriously, they never appear.”

    I know McIntyre has had an article asking for questions asked at realclimate that were ignored. If he hasn’t formalised this it might be worth asking him to set it up. It would be easier to pick out the questions with substance if all such questions were in one place. I know that some people try to ask repeated questions based on incorrect information originating from McIntyre, such as the idea that the data in the ftp://holocene.evsc.virginia.edu/pub/MBH98/TREE/ITRDB/NOAMER/BACKTO_1400-CENSORED directory only has bristlecone pine proxies removed when it actually has the entire North American tree ring data set and Queen Anne data set removed. This data made up 70% of all of the proxy data used by MBH98 prior to AD 1600 and was removed from the MBH98 dataset by McIntyre and McKitrick in their original 2003 E&E paper. I’m not surprised that the contributors to realclimate are not interested in answering questions based on this misinformation.

  15. #15 Spence_UK
    March 16, 2006

    Chris,

    You seem to still be pushing a point that the proxies never show higher values than the twentieth century. This simply isn’t true. Sargasso, Conroy, Indigirka for example, all of which have been used on the various multi-proxy studies at some point.

    On the topic of inverted U – no, proxies being warmer some time ago does not falsify the inverted U hypothesis, as the modern period might not be at the optimum growth condition (peak of the U) – it could be either side, so higher values earlier are not incompatible with this theory.

    I’m surprised you’re questioning the inverted U. To me, it seems obvious that a tree will have an optimum temperature for growth, too hot or too cold and it will grow less well. Also, the same tree species will have a different optimum temperature at different sites, and may be on different parts of the curve at the same time.

    Anyway, you asked for a reference. Here’s an example:

    http://www.fs.fed.us/rm/pubs/rmrs_p032/rmrs_p032_124_135.pdf

    Page 128, figure 4 shows how photosynthesis varies with temperature for some of the very trees used in MBH98.

    Again, I’m not quite sure what your point is regarding RealClimate. I wanted to ask the people at RealClimate a question, not Steve McIntyre, I don’t see how posting on McIntyre’s blog would help get an answer. If they were willing to answer the questions, they would have posted up and responded. I can’t see them going over to McIntyre’s blog to answer there instead.

    Looking at one of M&M’s articles, their comment was:

    “MBH98 did not report the results adverse to their conclusions from calculations excluding bristlecone pines (contained in the BACKTO_1400-CENSORED directory).”

    It does not say the bristlecones were the sole exclusion (as you imply), it just says that a run was performed without the bristlecones, that would have caused adverse results, but the adverse results were not reported (indeed the paper claimed that the results were robust to the removal of all dendroclimatic indicators). At best you could argue McIntyre could have worded it a bit better, but his conclusion that adverse results were not reported stands, and claims over the specifics of that run seem to me to be a weak straw man.

  16. #16 Dano
    March 16, 2006

    O’Neill said:

    a buncha stuff

    Surely decision-makers across the planet look at just one paper and don’t ask their scientists to look in other areas to help them decide about stuff. Makes it easier to think, dont’cha know.

    And I’m not surprised that some want to focus, atomistically, on one tiny fraction of all information.

    Certainly this is better than looking at the rates of ecosystem change we are currently experiencing measured in plant movements uphill and northward, ice-free days on new England and upper Midwest lakes, population changes, biodiversity impacts, invasive spp., changes in population species richness and diversity, fishery collapse, British Columbia pine devastation…

    [takes deep breath]

    …population crashes in Amazonia, Madagascar, changes in glacier mass balance across the globe, coral bleaching, changes in precip frequency and intensity, changes in runoff, melting permafrost, arctic biota changes, increasing temperature, O3 hole in Ant….

    [takes deep breath]

    …human appropriation of NPP changes, LU/LC indices changes such as NVDI, BIBI changes in urban streams due to changes in hydro regime, drought in US SW, SW Eur, Murray-Darling impacts, salmon pop. crashes in US Pacific NW…

    Ah, you get the picture.

    So, it is far more useful to atomistically quibble about one paper – from eight years ago – than it is to step back – today – and behold the planet-wide changes and challenges wrought by human enterprise.

    Thank you, O’Neill, for helping spread the good news that the scientific data about climate change is based on misinformation.

    Your misdirection will make it easier for me to sleep tonite! Things are getting better, Tinkerbell!

    Thanks, O’Neill!

    Best,

    D

  17. #17 John Cross
    March 16, 2006

    Spence, I am one of the people who seems to feel that the Sargasso is not a good proxy to use. Or rather I am unsure what it is a proxy of. Is it a proxy of sea temperature (unlikely given the thermal inertia involved) or is it a proxy of current location. If the latter, then it may actually be a proxy of the NAO but a more definite linkage would be necessary.

    Also, I don’t have the paper in front of me but I seem to recall that there were issues with salinity as well.

    I am not familiar with the others proxies you cite, but I can assert that the sargasso has been overused by some who don’t understand (or want to understand) the paper – Drs S & B for example.

    But I am with Dano – lets flag it and move on. There are a huge number of other proxies that do seem to indicate we are in a time of warmth higher than in the last thousand years. I look forward to future studies that incorporate them as well as the past ones.

  18. #18 Spence_UK
    March 16, 2006

    John,

    You might be leaning on one of my pet peeves there, watch out :-)

    Selection criteria for inputs into statistical studies have to absolutely strict and quantitative, never subjective because it is too easy to add bias – even subconciously (this is a well known and observed effect).

    Every single proxy out there has other dependencies. Tree rings are dependent on sunlight, availability of water, competition and fertilisation effects. Borehole temperature measurements are dependent on fractures and water contamination. dO18 measurements are dependent on precipitation.

    If you make an argument “I don’t like proxy X because it is contaminated by other influences” then you are introducing immediate bias into the process. If you reject one proxy this way, you must reject them all. This is one of my greatest concerns about the various multi-proxy studies: not so much the data that are put in, but why data are left out.

    This is a real weakness from a statistical point of view. Statistical studies in other fields have been thrown out because an apparently rigid selection criteria turned out to be ambiguous or tunable, yet in multi-proxy studies more often than not we have no clear reason for rejecting proxies at all.

    “There are a huge number of other proxies that do seem to indicate we are in a time of warmth higher than in the last thousand years” – and there are a huge number that suggest we aren’t. The ones I picked are deliberately chosen as ones used in the current multi proxy studies (Sargasso sea is in Moberg ’05). There are many out there which, for some reason, don’t get chosen.

    I would agree this kind of debate in the blogosphere won’t settle this, but the current crop of multi-proxy studies won’t settle it either.

  19. #19 John Cross
    March 17, 2006

    (Sargasso sea is in Moberg ’05).

    In fact if you look hard enough on the net you can find where I questioned Moberg’s use of the Sargasso sea data.

    I would agree this kind of debate in the blogosphere won’t settle this, but the current crop of multi-proxy studies won’t settle it either.

    Actually, we do seem to be somewhat in agreement. I think that paloclimatology is an interesting topic (and fun to argue about since I suspect that most of us don’t really have a clue about the science) but it is of limited use in today’s look at global warming. If there are statistical questions on the proxies then there are even more on the forcing functions.

    That is why I think the way forward is by looking at current conditions not past conditions. Papers like Philipona or Minschwaner or even Annan are examples of this.

    So as I said – flag it and more on – nothing to see here folks.

  20. #20 Spence_UK
    March 17, 2006

    “In fact if you look hard enough on the net you can find where I questioned Moberg’s use of the Sargasso sea data.”

    I had a quick google for your name and Sargasso and I note on RealClimate you asked the question as above – but this does not counter my argument above. If you reject Sargasso on the subjective basis that it has confounding influences then you must reject all proxies. You are allowing your own bias to select proxies that conform to the argument you are making.

    The points you (and Dano) subsequently raise are to do with the wider issue of global warming, which is really an entirely different debate to that of historical temperature reconstruction (IMHO). If you try to tie a result from historical temperature reconstruction to a view on AGW then you are simply likely to introduce more bias into the debate (because it is such a polarised issue). I think the science would progress more effectively if people didn’t try to link the two so closely.

  21. #21 John Cross
    March 17, 2006

    Spence, I don’t think you understood my point. I don’t think that the Sargasso sea temperature is a bad proxy, I don’t think it is a proxy at all. This would seem to be a fairly well defined criteria for rejection. Of course this is an opinion on my part – obviously Moberg disagrees with me and I suspect he knows a fair bit more than I do about paleoclimatology.

    If you try to tie a result from historical temperature reconstruction to a view on AGW then you are simply likely to introduce more bias into the debate (because it is such a polarized issue)

    Well, yes and no. I agree that the reconstructions do seem to polarize the debate, but if you take them away then there is a fair amount of agreement (at least in the published work). The only other area where I could see a similar polarizing issue was that of atmospheric temperatures (NAS convened a panel for that as well). That now seems well on its way to being resolved so I don’t think that it is such a polarized issue these days.

  22. #22 Spence_UK
    March 17, 2006

    John,

    My point is an important statistical point. Sure, subjectively, you can question the merit of the Sargasso sea (you still don’t lay down a criteria other than your opinion). I can sit here and claim that the Bristlecone pines aren’t a temperature proxy. Gavin Schmidt can claim the Thames freezing doesn’t make a good temperature proxy. It may not be obvious to you – but it is obvious to a statistician – that we would all be guilty selecting and rejecting proxies on subjective criteria and introducing bias. On such a loose reason as you provided, applied fairly to all proxies, would result in all proxies being rejected. You may have convinced yourself this isn’t the case but your reasoning would not bear up to statistical analysis.

    My key point here is that very few of the multi-proxy studies have clearly defined criteria for selecting proxies, and the selection should not be on the whim of an individual (I’m talking about the initial selection process, not the secondary cut that is often performed and described in some papers – although even this secondary cut is rarely adequately described).

    I certainly wouldn’t trust a statistical study without this, which is one of the reasons I don’t think the current crop of studies have much credibility.

  23. #23 Dano
    March 17, 2006

    If you try to tie a result from historical temperature reconstruction to a view on AGW then you are simply likely to introduce more bias into the debate (because it is such a polarised issue). I think the science would progress more effectively if people didn’t try to link the two so closely.

    This presumes the folk doing the data collectin’ do such a thing.

    Oh, wait: you must mean the folk who chatter over policy implications after the data are collected.

    Never mind. I’m sure you don’t confuse the two – completely separate – issues.

    Best,

    D

  24. #24 Dano
    March 17, 2006

    correction to one of my comments above: I mistakenly replied to O’Neill rather than Spence.

    Apologies.

    Best,

    D

  25. #25 John Cross
    March 17, 2006

    Spence: We appear to be at an impasse. I am not saying that the Sargasso sea is a broken proxy, but rather that it is not one at all – however perhaps this horse (or thread) is dead.

    I notice that Tim has a page up where he is offering to match pledges to The Global Fund to Fight AIDS, Tuberculosis and Malaria. In light of these enjoyable discussions I will show my appreciation by offering to match any pledge that you make (up to $25 – Tim is richer than I am). This means that anything you pledge gets increased by a factor of 3.

    Goodnight
    John

  26. #26 Chris O'Neill
    March 19, 2006

    “You seem to still be pushing a point that the proxies never show higher values than the twentieth century.”

    I’m not saying that at all. I’m saying that some of the proxy values in the twentieth century are higher (at the warm end) than most of the past proxy values over periods of say, 50 years. There were periods when a small fraction of the proxy values were higher than the highest values that occurred in the twentieth century but this could only at worst have affected a small fraction of years in any given period of 50 years. Considering also that there is noise in the proxies, this issue is only going to come into effect in those years where unusually high temperatures coincided with noise peaks so the fraction of affected years would be even less.

    ” Sargasso”

    Yes well, this indicates that you didn’t realise what I was talking about which is annually resolved proxies such as the ones used by MBH. The Sargasso foraminifera proxies have a resolution of 50 years at best so they won’t pick out the individual warm years. Since MBH didn’t use this type of low resolution proxy you can’t say their results are affected by a possible shortcoming of low resolution proxies. Also BTW, Moberg’s Sargasso proxy only goes up to 1925 but that doesn’t make much difference anyway because of the low resolution.

    In any case B&C’s issue was about linear extrapolation error which is clearly a problem at the cold end. They didn’t appear to be very interested in non-linear error so I don’t know if it’s that much of an issue among experts. If it isn’t then there’s not much point in non-experts talking about it.

    ” I’m surprised you’re questioning the inverted U. To me, it seems obvious that a tree will have an optimum temperature for growth, too hot or too cold and it will grow less well.”

    We’re not talking about trees in general here, we’re talking about trees in locations that are specifically chosen to cause high sensitivity of growth rate to temperature, i.e. near the cold limit at which the trees can grow. In such locations the temperature is a very important factor limiting how fast it can grow and it’s usually a lot colder than the optimum growing temperature. This is why the locations are usually near the high-altitude or high-latitude tree-line. So in these circumstances the response is not likely to be near an inverted U.

    ” Again, I’m not quite sure what your point is regarding RealClimate. I wanted to ask the people at RealClimate a question , not Steve McIntyre, I don’t see how posting on McIntyre’s blog would help get an answer.”

    Well if you had logged your questions somewhere then someone might have been able to tell you why they doesn’t get responded to at realclimate rather than just pointlessly whining about it to people who don’t even know what your questions were.

    “Looking at one of M&M’s articles, their comment was:

    “MBH98 did not report the results adverse to their conclusions from calculations excluding bristlecone pines (contained in the BACKTO_1400-CENSORED directory).”

    It does not say the bristlecones were the sole exclusion (as you imply),”

    BTW, McIntyre says elsewhere “If the bristlecones are excluded from the network and principal components are calculated without the bristlecones, there is no hockey stick shape.”

    Yeah sure, none of this is misleading and I don’t know why people try to ask questions at realclimate based on being mislead in this way.

    McIntyre’s “conclusion that adverse results”

    caused by removing 70% of all the proxy data used by MBH98 prior to AD 1600

    “were not reported stands.”

    Removing 70% of all the proxy data has an adverse effect. Gee that’s a surprise.

  27. #27 Spence_UK
    March 21, 2006

    Dano, I was trying to be polite by not drawing attention from your little Freudian slip, but since you insist on drawing attention to it yourself: we’re talking about historical temperature reconstruction, not hysterical climate contemplation. (Hysterical seemed appropriate from all the hyperventilating going on. You should see a doctor about that. Seriously, I worry about it.)

    “This presumes the folk doing the data collectin’ do such a thing. Oh, wait: you must mean the folk who chatter over policy implications after the data are collected.” – you present a false dilemma; I’m referring to neither of these. The people who go out and drill holes in things can’t subconciously bias results, because they can’t see the results before they do the drilling. But there are thousands upon thousands upon thousands of holes drilled, yet only a small percentage of these make it into the multi-proxy studies. The decision here is made with the data visible a priori, which should require a quantified and unbiased approach. This seems lacking in the studies I have seen. I’m not talking about the second cut here either (selection of temperature sensitive sites) but the first cut (selection of which sites to test).

    John, we aren’t going to get agreement on this topic. “Broken proxy”/”bad proxy”/”not a proxy at all” are just different ways of applying a subjective threshold, which unfortunately will have skew or bias depending on the contents of the results. As I said, a statistician would follow this argument, but a plain old scientist may not. Subconcious bias is prevalent even in such mundane fields as particle physics. In such a polarised field as climate, you can imagine the problems are likely to be much greater.

    To both Dano and John: I recommend you read “MYTH 9: SCIENTISTS ARE PARTICULARLY OBJECTIVE” from http://www.usc.edu/dept/education/science-edu/Myths of Science.pdf

    John: regarding donating to charity, I generally give to smaller charities; large charities these days create a strong brand image and are getting a larger and larger slice of the pie, leaving some small organisations marginalised. Since I don’t really engage with Tim I don’t feel any great compulsion to give to his preferred charity, however if you are willing to accept then I will donate a nominal amount to a small UK charity of my choosing, you can then give to a charity of your choosing. I hope you consider this “in the spirit” of your request.

  28. #28 Spence_UK
    March 21, 2006

    “I’m not saying that at all. I’m saying that some of the proxy values in the twentieth century are higher (at the warm end) than most of the past proxy values over periods of say, 50 years.”
    And some of them aren’t. Local and global temperature variations cannot be compared in this way. My original point was that if you go much outside the calibrated region of the twentieth century, you cannot guarantee linearity. Nothing you have said has addressed this claim. This point has been made in the scientific literature, and is presently being discussed. In fact, there has been a decline in tree ring growth of some supposedly “temperature sensitive” sites in the late 20th century which has been the cause of much discussion: most of it remains conjecture, but non-linearity has to be right up there as a candidate.

    The straw man about Sargasso that you have posted up is unbelievable. I lost count of the number of logical flaws in your statement here. I gave three example proxies, one of which is a tree ring chronology providing annual data for over a thousand years, and you reject them all because you were talking about MBH (I was talking about proxies and multi-proxy studies in general) and one of the examples I gave “isn’t appropriate” (even though the others could be). This is wrong on so many levels I don’t know where to start. Aside from the fact that I gave a range of proxy types, including low frequency and high frequency: if the high frequency data differed from the low frequency data, wouldn’t this indicate a problem that needed to be resolved anyway? Also, you suggest MBH seeks high bandwidth data to pick out individual warm years; if it was capable of doing this, it would have a halfway decent cross-validation r^2. It doesn’t. So much for that idea.

    Even worse, since there are plenty of proxies showing elevated values compared to today’s temperature, your observation that these appear not to be present in the B&C paper begs the question: why not? Was the sample selection in MBH biased?

    “In any case B&C’s issue was about linear extrapolation error which is clearly a problem at the cold end”
    B&C raised more than one issue. They mention both linear extrapolation problems (clearly a problem at both ends when you select an unbiased set of proxies) as well as the possibility of non-linearity.

    “We’re not talking about trees in general here, we’re talking about trees in locations that are specifically chosen to cause high sensitivity of growth rate to temperature, i.e. near the cold limit at which the trees can grow”
    Yes but as discussed already this can only be tested for the calibration period, and may not hold outside of that interval.

    “So in these circumstances the response is not likely to be near an inverted U.”
    Only if you assume previous temperatures are close to the temperatures in the calibration interval, as I stated earlier, this is the logical fallacy of affirming the consequent (does it feel like we are going around in circles here?). Given that some of these sites are not reflecting the recent 20th century warmth, why would they have reflected earlier warmth?

    “BTW, McIntyre says elsewhere “If the bristlecones are excluded from the network and principal components are calculated without the bristlecones, there is no hockey stick shape.” “
    Yes, that conclusion is based on his own runs using his own emulation of MBH98, so it stands on its own.

    “Removing 70% of all the proxy data has an adverse effect. Gee that’s a surprise.”
    MBH claimed in their paper that the NH reconstruction was robust to the removal of all dendroclimatic indicators. If they hadn’t made this claim, it might not have been a surprise to find out it isn’t. Given what they claimed, yes it is a surprise.

  29. #29 Dano
    March 21, 2006

    But there are thousands upon thousands upon thousands of holes drilled, yet only a small percentage of these make it into the multi-proxy studies. The decision here is made with the data visible a priori, which should require a quantified and unbiased approach. This seems lacking in the studies I have seen. I’m not talking about the second cut here either (selection of temperature sensitive sites) but the first cut (selection of which sites to test).

    This presumes many of those holes are adequate for study.

    I’m quite sure you won’t want the holes cored in, say, red alder in a river bottom. There are few suitable sites for a reason.

    And, yes, everyone appreciates the fact that your e-friends want retroactive data storage rules. Good on ya. Until then, if your boy want’s folk to share data with him, your boy shouldn’t act like such a…a…well, hmmm…if he wasn’t, he wouldn’t appeal to that group, would he?

    Best,

    D

  30. #30 Spence_UK
    March 22, 2006

    “This presumes many of those holes are adequate for study.”

    Nice straw man. I’m not claiming we shouldn’t be selecting sites. I didn’t even mention archiving (which is an issue, but for different reasons). I’m saying that site sampling should be transparent and quantified explicitly in any study that leans heavily on the use of statistics. Lack of transparency and subjective selection criteria results in subconscious bias in statistical studies. Well known and well documented effect.

    Statisticians understand why this is necessary, although not all scientists do (and I wonder if this is part of the reason you are struggling a bit with the concept). It is related to the link I posted above – the myth that scientists are objective.

    Introducing a proper, clear, verifiable selection criteria that eliminates the risk of bias is straightforward, and failure to do so is just lazy. It is also important to show robustness to sampling and data issues, which cannot effectively be tested properly with the current crop of studies.

  31. #31 Tim Curtin
    March 22, 2006

    Spence_UK said: It is related to the link I posted above – the myth that scientists are objective.

    Spot on, and you have shown conclusively that Kyoto is a naked emperor, devoid of any empirical substantiation for its hypothesis.

  32. #32 Dano
    March 22, 2006

    Timmy pecked out on his keyboard:

    [there is a] myth that scientists are objective…[s]pot on, and you have shown conclusively that Kyoto is a naked emperor, devoid of any empirical substantiation for its hypothesis.

    Aside from the fact that the science behind Kyoto has empirical substantiation, policy action is not rooted in empiricism [see, it’s the knowledge behind it that is], that policy action has no hypothesis, and you are mixing up science and policy,

    It matters not whether scientists are objective. Of course they are not.

    Your hand-waving and misstatements only work for dumb people.

    What matters is whether scientists follow the scientific method in their material gathering and analysis. The scientific method is a construct to eliminate bias.

    So please. Take your widdle FUD phrases where they’ll work, sir – where the rubes hang out. That ain’t here.

    Best,

    D

  33. #33 John Cross
    March 22, 2006

    Spence, Tim:

    That is a great site you have found Spence. Tim I would encourage you to read all the Myths involved. I found 1, 2, 3, 6, 10 and 13 to be very interesting and if only more people had read these there would be far less debate.

    Spence: The thing about Tim’s charity was that a contribution by you would have resulted in a matching by me and both of our contributions would be matched by Tim so it would have increased 3 times. If you can find me another cahrity where someone will match my contribution 3 times I will probably contribute (if I feel that it is a valid charity – no contributiuons to the Spence retirement fund).

    John

  34. #34 stewart longman
    March 23, 2006

    Coming late to this party. The ‘hockey stick’ paper seems a useful straw man, because the ins and out and controversies of statistical analysis can keep many people unfruitfully occupied for a long time, while distracting from the ongoing data collection (I note that glacier melt is speeding up, arctic temperatures are rising, the distribution of cyclones is globally broadening in area and season, and areas of drought are increasing in the US, while Europe is losing some of the Gulf Stream warmth), all consistent with the current models for global warming. Instead of discussing facts such as these, let’s focus on a fraction of the data, and never update it. PCA is both robust (data normalization is not as crucial in many contexts as fetishists would have you think) and weak (with number of principal components leading to different findings. Automated rules can be applied, but their use will vary depending on the context (especially the reliability of the individual observations), and reliance on automated rules without comparison of over and under-extraction is dangerous.
    It’s nice to see that so many of the general public have now decided to become experts in this area, though.
    As for the progress of science, normal science proceeds by reducing uncertainty (making predictions more accurate), new science proceeds by answering a niggling question, while pathological science proceeds by sticking to one point, insisting on anomalies in the standard model and ignoring any change in the data available. I don’t see how the denialists have done anything other than pathological science (i.e., all normal science is wrong, and we will ignore any improvement or extension of their data).
    Stewart

  35. #35 Dano
    March 23, 2006

    Stewart may be fashionably late to the party, yes, but I notice he places his hat at a jaunty angle, confidently strides into the room, sees the denialists huddled in a dark corner and then declares them unsocialized and maladapted wankers.

    Welcome, Stewart. The bar is over there and there’s a rack of my homebrew behind the bar – 9.0% and hoppy in front with a chocolate finish.

    Best,

    D

  36. #36 stewart longman
    March 23, 2006

    Thanks, Dano:
    I brought a couple bottles of homebrewed pale ale as well, and a few bottles of our local brew 9Wild Rose Industrial Park Ale – nice beer mats, too).
    I think that methodological fetishism is probably a useful wanker identification tool.
    Stewart

  37. #37 Dano
    March 23, 2006

    …and I’m sure you noticed, Stewart, that there aren’t any women huddling in that dark corner?

    There’s someone I want you to meet, BTW, she was asking about your hat…

    D

  38. #38 Chris O'Neill
    March 25, 2006

    “”I’m not saying that at all. I’m saying that some of the proxy values in the twentieth century are higher (at the warm end) than most of the past proxy values over periods of say, 50 years.”
    And some of them aren’t.”

    No kidding.

    “you cannot guarantee linearity”

    No kidding again. The point I keep trying to make is that non-linearity can only affect a small fraction of proxy values as long as it doesn’t affect observed proxy values in the twentieth century.

    “In fact, there has been a decline in tree ring growth of some supposedly “temperature sensitive” sites in the late 20th century which has been the cause of much discussion”

    at McIntyre’s blog? Dendroclimatologists have been dealing with these issues for a long time and one of the requirements (among many) for a multiproxy reconstruction is that each individual proxy has to sufficiently correlate with a local temperature record and have sufficient linearity. Is there anywhere in the scientific literature that discusses the effect of non-linearity on particular multi-proxy reconstructions (this doesn’t include McIntyre’s blog BTW).

    ” The straw man about Sargasso that you have posted up is unbelievable.”

    because you still didn’t get the point.

    “one of the examples I gave “isn’t appropriate””

    One of the examples showed that you missed the point.

    ” even though the others could be”

    The others were relevant to the issue of non-linearity affecting annually-resolved proxies which is the issue I was dealing with. I wasn’t trying to make any point about non-linearity affecting low resolution proxies. There are plenty of reconstructions that only use annually-resolved proxies to make the low resolution ones of lesser importance. BTW, speaking of low resolution proxies, how is McIntyre going with his demolition of Moberg’s reconstruction?

    ” if the high frequency data differed from the low frequency data”

    Generally speaking, no-one is suggesting that the data is free of noise.

    “Also, you suggest MBH seeks high bandwidth data to pick out individual warm years; if it was capable of doing this, it would have a halfway decent cross-validation r^2.”

    What’s the definition of halfway decent cross-validation r^2?

    ” your observation that these (proxies) appear not to be present in the B&C paper begs the question: why not? Was the sample selection in MBH biased?”

    What observation was that?

    “”In any case B&C’s issue was about linear extrapolation error which is clearly a problem at the cold end”
    B&C raised more than one issue. They mention both linear extrapolation problems (clearly a problem at both ends”

    thank you for your opinion,

    “when you select an unbiased set of proxies) as well as the possibility of non-linearity.”

    They mention non-linearity (again at the cold end) but their detailed discussion was on linear extrapolation error. The issue of extrapolation error would affect local reconstructions as well as multiproxy reconstructions so I would expect this issue should have been considered in the past in connection with local reconstructions and thus creators of multi-proxy reconstructions should be aware of it. BTW, B,F&C in their Tellus paper say “the calibrating 20th century climate is much warmer than the rest of the millennium.”

    ” “So in these circumstances the response is not likely to be near an inverted U.”
    Only if you assume previous temperatures are close to the temperatures in the calibration interval,”

    No this doesn’t assume they are close, not near also means not close.

    ” Given that some of these sites are not reflecting the recent 20th century warmth”

    How many of these sites are used in multi-proxy reconstructions and how much influence do they have on reconstructions? Also, if you have proxies that indicate warmer temperatures in the past then these have a good chance of not turning down at temperatures higher than the present so could be useful in reconstruction.

    ” “BTW, McIntyre says elsewhere “If the bristlecones are excluded from the network and principal components are calculated without the bristlecones, there is no hockey stick shape.” ”
    Yes, that conclusion is based on his own runs using his own emulation of MBH98, so it stands on its own.

    Oh so now it is just bristlecones that are excluded and you’re giving up on the idea that McIntyre actually meant it was more than just bristlecones that were excluded. The above quote came from one of McIntyre’s articles were he was talking about the contents of MBH’s BACKTO_1400-CENSORED directory. He was not talking about his own emulation. So my original point stands that McIntyre originated incorrect information such as the idea that the data in the ftp://holocene.evsc.virginia.edu/pub/MBH98/TREE/ITRDB/NOAMER/BACKTO_1400-CENSORED directory only has bristlecone pine proxies removed when it actually has the entire North American tree ring data set and Queen Anne data set removed.

    ” “Removing 70% of all the proxy data has an adverse effect. Gee that’s a surprise.”
    MBH claimed in their paper that the NH reconstruction was robust to the removal of all dendroclimatic indicators.”

    Before we find out if you’re playing with grammatical errors, exactly where did they say this?

    I’ll say in conclusion that all this wondering about details of dendroclimatology that McIntyre and other amateurs who read his blog do, should motivate them to read a few papers on the subject other than the ones McIntyre picks out for them. If they did they might get some idea of the amount of detail that has already been considered by professionals in the field.

  39. #39 Dano
    March 26, 2006

    Chris, chris, chris.

    You are presuming that the personality type with a self-identity that finds the words on CA appealing will read primary sources.

    Best,

    D

  40. #40 John Cross
    March 26, 2006

    Chris:

    Thank you for your last post. You corrected a misconception I had about MBH98. I have heard the skeptics say it so many times that I thought the paper said that the results were robust to the exclusion of the dendroclimatic indicators.

    So I went to the paper and had another read. From the paper:

    In short, the inclusion of the proxy data in the ?multiproxy?
    network is essential for the most skilful reconstructions. But
    certain sub-components of the proxy dataset (for example, the
    dendroclimatic indicators) appear to be especially important in
    resolving the large-scale temperature patterns, with notable
    decreases in the scores reported for the proxy data set if all
    dendroclimatic indicators are withheld from the multiproxy network.
    On the other hand, the long-term trend in NH is relatively
    robust to the inclusion of dendroclimatic indicators in the network,
    suggesting that potential tree growth trend biases are not influential
    in the multiproxy climate reconstructions.

    This only goes to show that Mr. Dano is correct with his thought that amateurs should be careful when reading technical papers.

  41. #41 Eli Rabett
    March 26, 2006

    WRT John Cross’ latest, MBH 99 has more on this issue.

  42. #42 Dano
    March 27, 2006

    Your last para., John, just goes to show why the women at Stewart’s party don’t go for the corner-huddling denialist-type with the Cheeto-stained boxers.

    See, women can tell when people make sh*t up.

    Best,

    D

  43. #43 Spence_UK
    March 29, 2006

    Hmm. Seems people are confusing the term “not as good” with “robust”. These are two separate aspects of a statistical system that mean different things. Let’s temporarily move away from MBH98 to explain the difference, in simple to understand terms.

    Let’s consider a simple system. We have 100 noisy measurements of a data point, each measurement being an independent normally distributed measurement with standard deviation of 1. If I average those data, I get a refined version of the measurement – the standard deviation drops with the square root of the number of samples, so my new measurement based on the average has an standard deviation of 0.1. Let us say (for instance), the result comes out to be -0.56.

    Now let’s say we are concerned half of those measurements come from a particular source that may contain bias. We can eliminate those measurements and see what happens. In an unbiased system, by only averaging 50 samples, the standard deviation increases from 0.1 to 0.14. So, yes, our new measurement is worse, but only by a small amount. Furthermore, our samples are correlated, so we would expect the new measurement to be close to the original one; but “not as good”.

    If by removing those samples, our result shifts from -0.56 to -0.48, it shows the result is robust. The statistics of the system are well behaved, and the -0.56 value should be more accurate.

    If by removing those samples, our result suddenly shifts from -0.56 to +0.63, we know we have a serious problem. That is a far larger swing than a well-behaved statistical system should allow. This shows a lack of robustness. On this basis, we cannot trust the -0.56 as it is far more sensitive to small changes in the data than it should be.

    The problem either lies with the data or the method. In the example above, we can quite quickly eliminate the method – a simple average is easy to understand – and question the underlying data, or the assumptions implicit in the data.

    So, having clarified the difference between “not as good” and “not robust”, let’s go back to MBH98. From the paper, and John Cross’ quote,

    “But certain sub-components of the proxy dataset (for example, the dendroclimatic indicators) appear to be especially important in resolving the large-scale temperature patterns, with notable decreases in the scores reported for the proxy data set if all dendroclimatic indicators are withheld from the multiproxy network.”

    OK, what he is saying here, is if you leave out the dendro samples the verification of the networks (in the late 19th century) becomes worse, i.e. the results are “not as good”, which is what we would expect. But then the next sentence addresses a different issue:

    “On the other hand, the long-term trend in NH is relatively robust to the inclusion of dendroclimatic indicators in the network, suggesting that potential tree growth trend biases are not influential in the multiproxy climate reconstructions.”

    Now, we can’t verify what happened in the 15th century because we don’t have instrumental records going back that far, so we can’t truly determine if the 15th century is “not as good”. But we can check that the statistics are well-behaved, and this is what this sentence refers to. Mann claims that the long-term trend in NH temperature (which dominates the global mean temperature, and contains the bristlecone pines) is robust, i.e. doesn’t suddenly swing when removing some parts of the data.

    Yet, we know this simply isn’t true. The 15th century data (i.e. the long term trend) swings by the amount of the entire reconstruction when eliminating the dendroclimatic indicators – or even just the bristlecone pines.

    So yes, Mann does claim that the reconstruction is robust to the inclusion or exclusion of the dendroclimatic indicators. But this point has been clearly falsified by McIntyre. To understand this point, you have to understand the differences between “robust” and “not as good” in the context of a statistical system. Mann knows the difference, which is why he realised the importance of making the claim in his study. It appears very few people here understand the important statistical differences between being not as good, and lacking robustness. I suppose this was pretty clear from some of the naive commentary on Burger and Cubasch’s results above, attempting to interpret B&C’s graphs as “temperature”.

    In my simple example above, we could eliminate the method as being a simple average. We cannot do this for MBH98 because of the complex methodology used by Mann. All we know – demonstrated by McIntyre, Burger and Cubasch – is that either the data, or the method, or both, is flawed in MBH98, as the results lack robustness.

    While I’m here, it would be worth pointing out that I wasn’t aware of the full details of what was left out in the CENSORED directory, and simply took Chris’ claim at face value. It seems Chris’ claim was a distortion of the truth. That’s a little disappointing, but not entirely surprising.

    “This only goes to show that Mr. Dano is correct with his thought that amateurs should be careful when reading technical papers.”

    I agree John; as an amateur, you should take care when reading such papers. You’ve misunderstood an important statistical point.

  44. #44 John Cross
    March 29, 2006

    Spence:

    I agree John; as an amateur, you should take care when reading such papers. You’ve misunderstood an important statistical point.

    You wound me deeply! Well, actually not – since I am not a statistican. My intrepertation was based on what I read. You equate “long-term trend” with “reconstruction” and I do not do so.

    I would intrepert the comment as saying if you remove the dendrochronology (leaving you the corals and the ice cores) these would then show a general trend up as we get towards the present.

    Regards,
    John

  45. #45 Dano
    March 29, 2006

    Wow, S_UK really has a lot invested in quibbling about an old, first paper, eh?

    Has someone chosen an identity that is dependent upon something being not true?

    Best,

    D

  46. #46 Spence_UK
    March 29, 2006

    John,

    Of course we have to judge what Mann meant from his words, and who knows he might have meant what you are saying, but it doesn’t particularly read like that. The late 19th century and 20th century are covered by the cross-validation and calibration statistics respectively (which effectively covers the “blade” of the hockey stick). The first sentence is referring to this period (because it refers to the scores). The “long term trend” is referred to in contrast to the previous sentence (“on the other hand”), and is a cruder test of robustness (since it is not possible to verify the earlier temperatures any other way), so I read it as referring to the “shaft” of the hockey stick, i.e. 15th century through to the 18th century.

    Dano,

    I originally tried to push the discussion down the route of the NAS panel and the wider issues of multi-proxy studies (such as those raised by Burger and Cubasch 2005, which raises a common problem across most papers even though it uses MBH98 as an example), but John and Chris have steered the discussion very much to McIntyre and MBH98. But I note you blame me for this. The irony is not lost on me.

    There is an old joke that is appropriate in this situation. Well, I use the word joke in it’s loosest possible terms, but it is apt.

    A fly-on-the-wall documentary team interview the mother of a small family, asking her how decisions are made in her household. “Well”, she says, “I make all the small, detail decisions for the household; my husband makes all the big decisions”. When asked to give an example of the decisions she makes, she says “I decide things like where we live, what schools the children go to, things like that.”

    The interviewer, taken aback by this description of “small” decisions, asks what sort of decisions the husband makes. “Oh”, she says. “He makes decisions like, who should be the next President, or whether we should go to war with another country”

    So there we have it. Dano, you go ahead and answer the big questions of the day. I’ll stick to the small ones.

  47. #47 Dano
    March 29, 2006

    …John and Chris have steered the discussion very much to McIntyre and MBH98. But I note you blame me for this. The irony is not lost on me.

    My sincere apologies, Spence, for not rereading this 100-some comment thread to keep what has transpired at the fore of my awareness. I do hope you have some forgiveness in your heart and I’ll try to do less throughout the day so I can follow along more better-like.

    But, I must say, re-scanning this thread to avoid further mistakes, though, suuure is interesting.

    I see that the argumentation upthread uses stuff that is pre-chewed for folk from only one source. The paleo/dendro community, apparently – even though they do this for a living – doesn’t move as fast as someone who does not do dendro for a living (or ever).

    Odd how that works, no?

    In the meantime, the decision-making community doesn’t read that one website that complains it can’t get traction (maybe it’s the amen chorus, or the cheer squad, or the periodic attacks or something, but hey).

    The decision-making community, see, doesn’t have time for small questions, Spence, so they get good counsel from folk who look all over the place (not just in one place, see) and make decisions.

    Too bad for the contrascientists, folks with oddly-chosen identities and anti-science ideologues that no one looks to one website for counsel, huh?

    Yup. Toooo bad.

    Best,

    D

  48. #48 Spence_UK
    March 30, 2006

    Dano,

    Good of you to admit you didn’t bother reading the thread. So your primary contribution to this discussion, is to drift on, not really follow what is being talked about, declare anyone who disagrees with you as being a wanker, and criticise people’s oddly-chosen identities. Am I right in assuming that is what you call “science”, and my discussion is “anti-science”? I also note your willingness to cheerlead Realclimate yet you get upset about it when it happens on other blogs. I wonder how you square that circle?

    My arguments specific to MBH come from 3 sources, yes McIntyre is one of them, also von Storch and Burger and Cubasch. My arguments about good scientific and statistical practice date back long before multi-proxy temperature reconstructions were a glint in their author’s eyes.

    Your link to the “decision-making community” – Seattle no less – is laughable. I can find a politician who will swear blind black is white if it helps his cause. That’s why I avoid politics and stick to the science.

    As for people “from the outside” knowing more than those “on the inside”, from my work with statistics, it doesn’t surprise me. Often people stuck in one community are just too close to the coal face to see what they are doing wrong, and it takes an outsider to wake them up to this fact. Which is one of the reasons I like to occasionally look at debates in fields outside my own. The Torah codes was a very interesting statistical debate, and I could draw parallels between that and some of the errors made in the multi-proxy studies. Interestingly, that concept was finally falsified when both sides of the argument got together, agreed a proper sampling protocol and failed to replicate the statistical significance claimed. The lack of a proper study protocol in the multi-proxy studies prevents this kind of a test. That type of statistical malpractice makes a statistician suspicious.

    BTW, your extensive use of the prejudicial language logical fallacy continues to amuse. Keep up the good work.

  49. #49 John Cross
    March 30, 2006

    Spence: I put up with your snide comment that I misunderstood an important statistical point..

    But I can not fathom why you made this comment: but John and Chris have steered the discussion very much to McIntyre and MBH98.

    If you please re-read the thread you will find that I do bring up McIntyre in a discussion with Dennis Williams in regards to a claim about Mr. McIntyre being funded by industry (as a side note I don’t think he is and I don’t really care if he is, but even you have to admit the irony of making a statement that you are not willing to back up).

    However let me outline my discussion with you:

    –>We started discussing the composition of the NAS panel. I discussed Cuffey’s work and Christy’s contribution.
    –>Out next conversation started with a statement by yours about misunderstanding McIntyre’s point to which I did respond.
    –>My next comment was again a response to yours where you brought up M&M, vS and B&C.
    –>My next comments were about the Sargasso Sea (not a proxy of MBH98 and nothing to do with McIntyre) and contained a statement that we should move on which you did not apparently agree with since you said “The points you (and Dano) subsequently raise are to do with the wider issue of global warming, which is really an entirely different debate to that of historical temperature reconstruction (IMHO).”
    –> I did respond to a post by Chris about how Mann used robust in the paper since the phrase I had always heard was “robust to the presence or absence of …” and I had accepted it without question. I am willing to admit I was wrong in this respect.
    –>You then attacked me again and brought the discussion back to McIntyre.

    Now, I request that you back up your statement and point out where I have “steered the discussion very much to McIntyre and MBH98″. Absent that I expect to receive a retraction!!!

  50. #50 Dano
    March 30, 2006

    Good of you to admit you didn’t bother reading the thread. So your primary contribution to this discussion, is to drift on, not really follow what is being talked about, declare anyone who disagrees with you as being a wanker, and criticise people’s oddly-chosen identities.

    As you can see, S_UK, my comments are throughout, so I have read the thread. What I did not do was find it important enough to remember.

    Am I right in assuming that is what you call “science”, and my discussion is “anti-science”?

    That’s quite a stretch.

    I can find a politician who will swear blind black is white if it helps his cause. That’s why I avoid politics and stick to the science.

    The decision-makers are acting on the science. Good on you to stick to the science, which is good enough to act on!

    Best,

    D

  51. #51 Jeremy
    March 30, 2006

    The earth is warming. Ecosystems are changing at rates not seen before according to our current understanding. This is not in dispute…

    The earth and life are infinitely complex. So much so that noone can understand how they were created. Therefore there must be a creator being called God. This is not in dispute.

    I can pretty much say whatever I want after that.

    Saying, “These are facts, accept them, they are not in dispute,” is no different than an evangelist preaching from the pulpit. I should know, I dealt with such people while growing up.

    Science is different, it questions everything, including itself.

  52. #52 Spence_UK
    March 30, 2006

    John: let’s be clear here. My remark about you being an amateur misunderstanding a statistical point was a parody of the fact that you made the same sort of statement. If you feel put out at being labelled with that statement, perhaps you shouldn’t have used it in the first place.

    Looking back down the thread, our first discussion was regarding the NAS panel (on topic! shock!!!). My second comment was to do with a misrepresentation raised by Chris about the M&M paper, in which I tried to broaden the discussion outside of just M&M (by including B&C, etc). Looking down the thread I would agree that we continued to discuss wider issues, while Chris continued to raise new points specific to the M&M vs. MBH98 debate.

    So I retract my earlier statement, and amend it to stress to Dano that whilst John and I were generally discussing the broader issue of multi-proxy studies, Chris was very much narrowing the debate to MBH98 vs. M&M.

    While I’m at it, I’d like to stress that I have no issue with narrowing the debate to M&M vs. MBH98, or broadening it to the wider issue of multi-proxy studies: there is nothing wrong with either of those approaches. I’m just bemused as to why Dano is trying to dress the former up as a bad thing and label me with it, when it clearly takes two to tango. I guess he only has petty arguments left to respond with, although I’m almost descending to his level by offering a reply.

  53. #53 Spence_UK
    March 30, 2006

    As you can see, S_UK, my comments are throughout, so I have read the thread. What I did not do was find it important enough to remember.
    Posting comments to a thread does not require reading the thread. If it isn’t important enough to remember why do you feel the need to comment? It would seem to me your apparent disinterested status is greatly exaggerated.

    That’s quite a stretch.
    Mental note to self: Dano is incapable of detecting irony. Are you American? That would explain a lot.

  54. #54 Dano
    March 30, 2006

    Jeremy, if you have some facts and empirical findings that show something other than what I said, bring them.

    Otherwise, according to our understanding what I said is not in dispute. See, just because somebody doesn’t like a finding doesn’t mean its in dispute.

    Your sloppy argumentation does nothing to change this fact. It’s great that you can say science questions stuff, but typing all those wurdz does nothing to refute what I said.

    ===========

    S_UK:

    If you flap your arms hard enough, you can fly. That’s what I hear, anyways.

    They also tell me that too much hand-waving leads to Carpal Tunnel Syndrome. With plentiful evidence already pointing towards wankery, one might suggest that you be careful of your right wrist, if you get my drift.

    Best,

    D

  55. #55 Spence_UK
    March 30, 2006

    If you flap your arms hard enough, you can fly. With plentiful evidence already pointing towards wankery [snip]
    Darn you Mr Dano, with your cunning scientific logic and reasoning. I got no answer to that one.

  56. #56 Chris O'Neill
    April 2, 2006

    “the data in the ftp://holocene.evsc.virginia.edu/pub/MBH98/TREE/ITRDB/NOAMER/BACKTO_1400-CENSORED directory only has bristlecone pine proxies removed when it actually has the entire North American tree ring data set and Queen Anne data set removed”

    I made this statement in the mistaken belief that this was the source of McIntyre’s misinformation that just leaving out the bristlecone pine proxies makes a substantial difference to MBH98’s reconstruction. I have found out that McIntyre used data from the above directory, apparently without any statement about what Mann’s actual use of that data was, for his own clearly flawed reconstruction process to produce a different result from MBH98 and thus try to claim that this proves that MBH98 is not robust to leaving out bristlecone pine proxies. As Mann points out in this interview, MBH “performed a set of sensitivity tests to determine if a skillful reconstruction was available without correcting certain high-elevation tree-ring chronologies”, i.e. the bristlecone pine proxies, “for sensitivity to possible non-climatic (e.g. co2-fertilization) effects. These calculations were performed as part of these analyses, after MBH98. This is all discussed quite clearly in our follow-up paper to MBH98 published in the journal GRL in 1999″ (MBH99).

    So Mann was quite aware of the possible non-climatic effects on the bristlecone pine proxies and presumably the absence of these proxies in the BACKTO_1400-CENSORED directory means he was doing some sort of robustness test involving these proxies. So Mann certainly gave these proxies some robustness testing.

    Mann also points out “It is sad that McIntyre and McKitrick have been reduced to scouring our website for things like this, to take out of context, and make false and misleading assertions.”

  57. #57 Spence_UK
    April 2, 2006

    Lets be clear here Chris: you made this mistake in an unprincipled attempt to defend MBH98.

    The Mann method of reconstruction – whether you include the dubious CO2 fertilisation correction Mann applies in MBH99 or not (which Mann’s best attempt to justify to date involves telling people to look at google scholar – classy) – is not robust to the presence of the bristlecone pines, representing less than 10% of the MBH98 data set, as shown by M&M.

    It is not robust to a posterior selections in the regression step, as shown in B&C.

    These can be trivially proven, by yourself, by downloading R code either from the ClimateAudit site or Ammann and Wahl (the two pieces of software agree to machine accuracy, so it really doesn’t matter which) and applying the test (note: you have to set this test up with the W&A code, they are rather selective as to which tests they provide input data to run).

    The “reconstruction” fails significance against r^2 cross-validation (ref. W&A) and fails significance against properly computed RE statistics (ref. M&M reply to Huybers, the last word on this subject in the scientific literature)

    Bottom line: neither MBH98 or MBH99 are robust to a large number of changes (important note: “not robust” and “not as good” do not mean the same thing – see my note above).

    Whether or not Mann knew about the problems is just a sideshow IMHO. The “reconstruction” is junk for more than one reason, I personally have no interest in whether Mann knew about this at the time or not (others may feel differently), let’s just accept more work is needed in the area and move on.

  58. #58 Chris O'Neill
    April 6, 2006

    “Lets be clear here Chris: you made this mistake in an unprincipled attempt to defend MBH98.”

    unprincipled. Thank you for your opinion. I’ll admit that McIntyre can be pretty confusing to a non-expert. Especially when he gives an article a particular title and says he is going to show something relevant to that title and then without skipping a beat changes the issue. I can see that relying on McIntyre to stick to the issue (as a non-expert needs) is not a wise choice. Let me point out something else Mann said:

    “This claim by MM is just another in a series of disingenuous (off the record: plainly dishonest) allegations by them about our work.”

    It is staggeringly hypocritical for someone supporting McIntyre’s arguments to accuse others of lacking principle.

    “The Mann method of reconstruction is not robust to the presence of the bristlecone pines, representing less than 10% of the MBH98 data set, as shown by M&M.”

    Sure, just like they showed it was not robust to changing the principle components analysis centring choice.

    “It is not robust to a posterior selections in the regression step, as shown in B&C.”

    We’ll have to wait to see what the experts say about B&C’s work.

    “These can be trivially proven, by yourself, by downloading R code either from the ClimateAudit site or Ammann and Wahl (the two pieces of software agree to machine accuracy, so it really doesn’t matter which) and applying the test (note: you have to set this test up with the W&A code, they are rather selective as to which tests they provide input data to run).”

    The W&A emulation, that I’m aware of, of MM05’s E&E method using the MBH 1450 proxy network produces a reconstruction without the bristlecone proxies that passes validation and is robust to the absence of the bristlecone proxies. So there’s no problem going back to 1450. That hockeystick is there from 1450 with or without bristlecone proxies. Their emulation of MM05’s E&E method using the MBH 1400 proxy network produces a reconstruction without the bristlecone proxies that fails validation while if the bristlecone proxies are included it passes validation. So I guess robustness in this case means the temperature reconstruction passes validation against the instrument record or more recent proxies in spite of part of the proxies having growth trends that are not related to the instrument record or other proxies.

    “The “reconstruction” fails significance against r^2 cross-validation (ref. W&A)”

    We know that’s vitally important to McIntyre.

    “and fails significance against properly computed RE statistics (ref. M&M reply to Huybers, the last word on this subject in the scientific literature)”

    M&M’s reply is the last word. Sure.

  59. #59 Spence_UK
    April 6, 2006

    Wow.

    You use a quote from Mann that describes M&M as being “dishonest” (off-the-record, in a letter, whatever the hell that means) without any supporting evidence, and you suggest McIntyre is the unprincipled one? Now that is funny.

    As for “non-expert”, I have considerable experience in statistics and statistical issues. Whilst I recognise I have little expertise in paleoclimatology, the issues that I have raised are primarily statistical issues, on which I am quite capable of understanding and responding to. I have seen little evidence that your statistical skills are up to understanding the substantive issues (e.g. your failure to understand the differences between “not as good” and “not robust”).

    Many of my arguments come from the scientific literature, apart from one or two things from W&A, most of yours seem to come from stuff Mann and co have posted on blogs.

    “Sure, just like they showed it was not robust to changing the principle components analysis centring choice.”

    I see you don’t have a real answer to that, but ironic mockery of the point probably helps make it seem less real in your own mind, at any rate.

    “We’ll have to wait to see what the experts say about B&C’s work.”

    B&C are experts, and the work is in the scientific literature, which means it has passed peer review by “experts” of some description. That isn’t to say it won’t be found to be flawed, or superseded in the future. That’s the way science works. But at the moment it is “current”.

    “The W&A emulation, that I’m aware of, of MM05’s E&E method using the MBH 1450 proxy network produces a reconstruction without the bristlecone proxies that passes validation and is robust to the absence of the bristlecone proxies.”

    Hmmm, you still haven’t figured out this lack of robustness in a statistical study have you? If person A shows a method or data lacks robustness through demonstration, picking cases that appear to be robust cannot falsify the observed lack of robustness. It doesn’t make the problem cases go away. W&A selectively running cases that give similar results adds nothing. They judiciously avoided the M&M cases that shows no robustness. It is obvious just looking at their website how selective they were in what they ran. That is why I said above you have to set up your own input files with W&A. The W&A emulation is mathematically and algorithmically identical to the M&M emulation, which is why it gives the same answer. (You can check this easily yourself by downloading the code and running it). None of the emulations pass a true test of statistical skill, including MBH or W&A (unless you can provide me the reference below).

    “”and fails significance against properly computed RE statistics (ref. M&M reply to Huybers, the last word on this subject in the scientific literature)””

    “M&M’s reply is the last word. Sure.”

    If you think someone else has set an RE benchmark for Mann’s work after McIntyre’s reply to Huybers in the scientific literature, please point me to the reference, and I’ll back down on this. W&A set no benchmark in CC, they rely on their GRL article which (as we know) did not pass peer review.

  60. #60 Chris O'Neill
    April 14, 2006

    “You use a quote from Mann that describes M&M as being “dishonest” without any supporting evidence”

    Refer to what Mann said,

    “As for “non-expert””

    I was referring to myself as a non-expert who was reliant to a degree on McIntyre making a straight argument which he doesn’t always do.

    “B&C are experts”

    I said the experts, which doesn’t mean just two of them.

    “But at the moment it is “current”.

    Whatever “current” means. Could that be “awaiting confirmation”?

    “”The W&A emulation, that I’m aware of, of MM05’s E&E method using the MBH 1450 proxy network produces a reconstruction without the bristlecone proxies that passes validation and is robust to the absence of the bristlecone proxies.”
    Hmmm, you still haven’t figured out this lack of robustness in a statistical study have you?”

    You don’t seem to be getting the point here. The issue is robustness to whether bristlecone proxies are included or excluded. W&A showed that it is robust to including or excluding bristlecone proxies from the 1450 proxy network. McIntyre’s issue is that the reconstruction doesn’t work without bristlecone proxies (which is true before 1450) and that somehow these bristlecone proxies produce an incorrect result.

    ” picking cases that appear to be robust cannot falsify the observed lack of robustness”

    You seem to have the idea that these proxies are all just independent random variables with the same mean etc. Well they’re not. They’re correlated so that when the number of proxies available drops quite low (such as before 1450), leaving out one set can have quite a large effect on the result. There is a very non-linear relationship between proxy numbers and accuracy of results. I haven’t looked up what the number of proxy sets for MBH 1400 to 1450 is but at 1000 in MBH99 it is seven in the northern hemisphere.

    “None of the emulations pass a true test of statistical skill, including MBH or W&A”

    Yes McIntyre thinks he has a true test of statistical skill, considering that in his rejected comment to Nature he incorrectly calculated all of his verification statistics.

    “”and fails significance against properly computed RE statistics (ref. M&M reply to Huybers, the last word on this subject in the scientific literature)””
    “M&M’s reply is the last word. Sure.”

    I was taking “the last word” in the sense that there will never be another word said about it. “the latest word” would have been more accurate. The status is that M&M disagree with W&A’s view on RE statistics. W&A say they have done work to show a different view but have not yet published it.

    “W&A set no benchmark in CC, they rely on their GRL article which (as we know) did not pass peer review”

    for reasons other than the material on RE statistics. Hopefully this material will be published very soon.

  61. #61 Spence_UK
    April 17, 2006

    “Whatever “current” means. Could that be “awaiting confirmation”?”

    In a Popperian sense, “valid until falsified” seems reasonable to me. Yes, it may be corrected or superceded, but until then it is the best available.

    “for reasons other than the material on RE statistics.”

    No – for many reasons including the material on RE statistics. Steve explains this better on the thread noted below.

    There seems little point in continuing this here since the same points are being discussed under this topic, including comments from Steve (who will put these things more accurately than I can). There seems little point in having the same debate twice in two places, so I’ll suggest that the debate continues there [where it won’t need me :-)].

  62. #62 Anonymous
    April 25, 2006

    “”Whatever “current” means. Could that be “awaiting confirmation”?”

    In a Popperian sense, “valid until falsified” seems reasonable to me. Yes, it may be corrected or superceded, but until then it is the best available.”

    So if someone publishes a paper like for example “Corrections to the Mann et al (1998) Proxy Data Base and Northern Hemisphere Average Temperature Series” by McIntyre and McKitrick then even though it is full of rubbish it becomes the best paper available until it is corrected or superceded simply because it’s published. Sure.

    “”for reasons other than the material on RE statistics.”

    No – for many reasons including the material on RE statistics. Steve explains this better on the thread noted below.”

    Not that I noticed.

  63. #63 Chris O'Neill
    April 25, 2006

    The previous message was mine. (In case anyone can’t guess.)

  64. #64 concerned of berkely
    April 17, 2007

    Given that you guys have demonstrated your dispassionate, forthright, and determined commitment to objective and sound science, I would be interested to know what you think of this article.

    http://libertyunbound.com/archive/2007_05/jason-warming.html

    How do I respond to those who argue this viewpoint?

    Also, while I am here, it is disappointing to see the dendroclimatologists allow a long and damaging discussion on tree/temperature relationships without careful and measured response.

The site is currently under maintenance and will be back shortly. New comments have been disabled during this time, please check back soon.