Trip Report – Holland

A late report on my visit to Holland. I don’t think that I’ve talked as much in a month as I did in 36 hours in Holland. I had two main presentations -one at KNMI in the morning; one at the Free University in the afternoon. I also had two long newspaper interviews and a long meeting on Friday morning with a Dutch mathematician. After the KNMI presentation, I had lunch with Rob van Dorland, Nanne Weber, Jos de Laat of KJNMI, all of whom were very cordial, and spent much of the afternoon talking with them.

Throughout I was very cordially entertained and guided by Marcel Crok of NWT (and his charming wife) Any success that I had was largely due to Marcel’s initiative.

At the Free University lecture, there were some CA readers: Ferdinand Engelbeen, Hans Erren, Hans Labohm; and perhaps a couple of others. (Larry and Lena Hulden were at the Stockholm conference and I had a chance to talk to them there.) Ferdinand kindly gave me some very interesting Belgian beers; Hans Erren a 2003 Burgundy – from the hot summer. All in all, I felt very welcome in Holland.

The Lion’s Den?

KNMI is the Dutch meteorological institute. It is located in De Bilt, which is a train ride outside Amsterdam. There were about 40-50 people present for my presentation.

KNMI has felt obliged to respond to criticisms of the Hockey Stick. They ran an endorsement of the Hockey Stick in their website in February 2005; in a newspaper article by Rob van Dorland in October 2005 and, most recently, in its most recent Annual Report in August 2006, where they said that the “points criticised by McIntyre and McKitrick have been mostly refuted in various studies”.

They are participating in a new study (MITRIE headed by Martin Juckes – see for example) of the Hockey Stick financed by the Dutch government, involving pretty much the European Hockey Team (Briffa, Moberg, Esper) plus Nanne Weber from KNMI. This study will be submitted to Climates of the Past in the next couple of weeks. (I wasn’t given a copy of the article to comment on.)

Last spring, KNMI had a workshop on the Hockey Stick at which Marcel Crok of NWT gave a presentation. He had also interviewed Nanne Weber in the spring, so I had a pretty good idea of their ideas and background. For example, Crok’s notes on his interview indicated that she believed the following:

ⵉWe had failed to show the impact of Mann’s PC methodology on the NH reconstruction even though this was shown in our EE article(which she had not read);
ⵉYou have to take more PC’s into account until you represent the underlying dataset well enough (without going into detail about Preisendorfer rule she said something like: you add new PC’s until the sum of the PC’s is not changing anymore).
ⵉIt’s not easy to see what’s the effect of the decenterd PCA on the end result, so you have to calculate it. Von Storch and Zorita did this with pseudo proxies, concluding it has little effect.
ⵉW&A showed according to her that there is no difference in end result between centered and decentered PCA
ⵉShe claimed that W&A showed that a simple average of all the MBH data gives a hockey stick as well;
ⵉWhen asked about WA confirming that the R2 failure, she told Crok that
she was not a specialist on this and that Martin Juckes was responsible for this. When asked by Crok whether she had investigated the R2 failure, she said that she hadn’t and had no intention to do.
ⵉWhen asked by Crok whether they were planning to calculate the error bars in MBH, she said that she would ask Juckes about this as well.
ⵉShe gave Crok a copy of the EOS paper of July 12 2005 in which I was slagged by Crowley. Crok told her that he had read a lot of the correspondence and that I was nowhere not friendly. She said that I was equally responsible that things got out of hand.
ⵉshe said that she understood why people don’t want to make their data available: they invested so much time in it and it’s so much work to archive everything.
ⵉCrok asked her about the crucial datasets: Polar Urals, Tornetrask, bristlecone pines and foxtails. She told Crok that Tornetrask is only in Esper and Moberg and Polar Urals only in Mann & Jones, Crowley and Lowery and Esper, but would look into it and come back to him on it.

So while most of the audience were unfamiliar with the territory, KNMI as an institution had taken an active interest in the matter and had been officially both critical and dismissive. My trip to KNMI was supposed to be a trip into the lion’s den.

I sometimes get the impression that some of my critics expect me to fold at the most minor confrontation. Any one expecting blood on the floor of the stadium would have been disappointed. The scientists were polite. The questions tended to be pretty general rather than about proxy reconstructions. None were hard or even detailed questions – somewhat surprising given the previous statements by the institution.

Now one can’t draw any very strong conclusions from this; I might simply have bored then to tears. My presentation at KNMI was definitely the weakest of the three presentations as I tried too hard to bring in details to argue against the Weber perceptions and completely missed my time markers. However, the people stayed for the over-length presentation and seemed politely interested. I think that I conveyed an impression that I knew my material and that I was fairly pleasant. I am increasingly coming to the view that that is all you can do with technical material in this type of presentation. For actual technical points, you are are really just trying to interest people in what you’re doing rather than presenting information that is going to convert anyone.

I was interested in trying to persuade Weber and von Dortland about the fatal weakness of Wahl and Ammann. I presented the diagram about von Storch and Zorita that had interested von Storch in Stockholm – noting von Storch’s comments in Stockholm that he now understood how a contaminant proxy could affect a network. My position on Wahl and Ammann is that their results yield a q.e.d.:
ⵉThey say that reconstructions without bristlecones/foxtails have no climatological meaning because of failed RE statistics. We agree with this;
ⵉThe NAS panel, following earlier articles and recommendations by Graybill and Idso 1993; IPCC 2AR; Biondi et al 1999; etc. stated that strip-bark should be avoided in temperature reconstructions;
ⵉStrip-bark avoidance includes the bristlecone/foxtail sites
ⵉtherefore MBH cannot assert 20th century uniqueness based on their data and methods as stated in McIntyre and McKitrick 2003. q.e.d.

This argument seems completely obvious to me. It seems obvious to anyone that is not a climate scientist. However, climate scientists seem to feel that there is some sleight-of-hand in the argument. Thus, despite my best efforts, this seemingly simple argument had not impressed van Dorland or Weber, who remained firmly in the Wahl-Ammann camp.

I classified three families of problem in the multiproxy literature:
1) the specific and somewhat exotic problems of MBH98 – bristlecones, biased PC methodology and failed verification statistics;
2) the “divergence” problem in which proxies didn’t work in the last half of the 20th century;
3) cherry picking and data snooping; in which the proxies were used over and over and the selection was biased, showing the particular impact of the Yamal series.

Afterwards, I talked to Weber, van Dorland and de Laat at some length, with Weber being the point person.

On the Divergence Problem – they were “working on it”. Weber denied that the Divergence Problem applied to ring widths – here she was simply wrong. The point is buried pretty deeply in the literature – there’s one graphic in passing in Briffa et al 1998, but the graphic exists and was one of the graphics that I presented. As to the mismatch between proxies and temperature, she thought that wasn’t a problem – we had a temperature record in the 20th century and could use it; for earlier periods, we could use the “proxies”.

On the selection of proxies – Weber said that they used the proxies that were “available”. I offered to bet 500 Euros that their CPD submission would use the Yamal series, rather than the equally “available” Polar Urals. She didn’t take the bet. This is a big deal in my opinion – results in these little 10-series subsets, such as their forthcoming CPD submission – are highly sensitive to picking. The issue had been raised last spring by Marcel Crok, but didn’t seem to have registered.

We talked about bristlecones/foxtails. She argued that they were “necessary” to represent an important region that would otherwise be unrepresented. They didn’t seem to care about the statement by the NAS panel that they should be “avoided” in temperature reconstructions.

We talked for a while about “data snooping” as the term is used in economics – if you know what the series look like as a result of a literature search, including searches by prior authors, then usual statistical tests have no meaning. She either disagreed with this point or didn’t understand it – she definitely didn’t agree with it.

We talked about failure of verification statistics. Weber said that Juckes was their statistics guy; he knew about r2 and RE, while she wasn’t much interested in statistics.

It’s pretty easy to predict what their CPD submission will look like. I’ll bet that it has almost exactly the same proxy network as Osborn and Briffa 2006. They will argue that they can get a HS from this network without using PC methods; ergo, everything is fine in Team-world. Anyway, we shall see.

Free University
Afterwards, Marcel and I travelled back to Amsterdam, where I was interviewed by a Belgian newspaper; then a very pleasant dinner with Marcel and some of his friends and family. Then off to the Free University. When I arrived at the university, there was a sign “Lezing McIntyre”‘?, together with a photographer from the university newspaper. It was like being a very minor celebrity.

I’ve posted up my powerpoint here for the Free University presentation, which is pretty much the same as the Stockholm presentation and is what I should have presented at KNMI, where I tried for too much detail.

This lecture was very well attended with about 120-140 people. A couple of people snickered loudly when they heard that Mann’s verification r2 was nearly zero. I got a lot of questions, but the questions were mostly about global warming and politics rather than proxies or statistics. I realize that this somewhat comes with the territory, but generally I do not attempt to provide opinions on broader issues.

Afterwards, a lot of people came to talk – it was particularly nice to talk to the friendly faces of the two Hanses and Ferdinand. Some young students wanted to have their picture taken with me – my daughter would have rolled her eyes at this.

Through the trip, I stayed in a very pleasant small hotel overlooking a park arranged by NWT. Amsterdam is a very nice city to visit; it was fun being a small celebrity; the trip to the lion’s den was stimulating and Marcel Crok was a terrific host.



  1. James Lane
    Posted Sep 24, 2006 at 10:02 PM | Permalink | Reply

    Steve, where’s the link to your Free University presentation?

  2. Steve McIntyre
    Posted Sep 24, 2006 at 10:20 PM | Permalink | Reply


  3. Howard Wiseman
    Posted Sep 24, 2006 at 10:49 PM | Permalink | Reply

    From R.Pielke Sr. Cliamte Science Today-not quite as collegial as your European experience.

    There was a news article today in the Denver Post entitled “Global Warming?”. My views are clearly articulated on Climate Science (see), but unfortunately, the article does not clearly distinguish that there are a diversity of views among climate scientists on the role of human- and natural- climate forcings. The article also has errors; for example, the statement attributed to me that

    “Man-made changes to the land, in addition to about 30 other greenhouse gases – some man-made, some natural – may play an even a bigger role, he said.”

    is in error. I stated that besides land use changes, the input into the climate system of natural- and human-aerosols have a range of effects on the climate system, which may exert larger effects on the global climate system than the radiative effect of the human input of CO2. We have documented this conclusion, for example, in our peer-reviewed paper “Measurement-based estimation of the spatial gradient of aerosol radiative forcing”. This does not mean we should not seek to reduce the human input of CO2 into the atmosphere, but it does mean that controlling CO2 emissions alone only addresses a fraction of the human alteration of the climate system.

    Of most concern to those who value courteous scientific debate, however, is the quote from Jim Hansen. It reads

    “Some of this noise won’t stop until some of these scientists are dead,” said James Hansen, head of the Goddard Institute for Space Studies in New York City, and among the first to sound the alarm over climate change.”

    Regardless as to your perspective on climate change, this callous statement from a federal administrator who is very visible in the climate science debate should be strongly repudiated by everyone who accepts that the debate should be about the science. I invite Dr. Hansen to expand on, clarify, or correct the comment that he made for this news article

  4. James Lane
    Posted Sep 24, 2006 at 11:09 PM | Permalink | Reply

    If anyone wants to comment on Pielke/Hansen, could I suggest they do it over at Climate Science?

  5. Posted Sep 25, 2006 at 2:59 AM | Permalink | Reply

    Steve, which was the Belgian newspaper that made the interview?

  6. Michael Hansen
    Posted Sep 25, 2006 at 4:20 AM | Permalink | Reply

    Hi Steve M;

    Given the fact that most of the points raised by Nanne Weber is also the (talking) points of Real Climate, Lee, and everyone else who wants Mann vindicated, I think you should addressed them in a collective fashion, with references as to where more in-depth information can be found. I cannot judge Nanne Webers points without going though a dozen posts & +100-pages-reports, and spending a lot of time. That may be the rule of the game if you want to follow the hocky stick debate — tough luck, so to speak — but it also would de facto cut out a lot of people, who has the basic training to follow the debate, but simply don’t have the spare time to find though this labyrinth of he said she said.

    More generally — and I do know this is not a small project — I’m dreaming of some sort of timeline, starting with the Mann e-mail exchanges and ending with…I don’t know…the Wegman report. Each point on this timeline should hold a nested structure with an abstract explaining the key points, the rebuttal from The Team, you answer to those rebuttals and so fourth. The idea is to collect all the good arguments — pro and con — spread all over this and other sites, in a more comprehensive and easy-to-access way. A nested DHTML menu — and those are freeware — somewhere on this page should do the trick, though the time consuming part of the job, of course, would be to find, concentrate, and organize data.

    Sorry for the somewhat off-topic post, but I just remembered that occasionally people complain about the “browsability’ of this site. And I agree. It IS difficult for none-experts to find information and wade through the Hocky Stick controversy.

  7. Posted Sep 25, 2006 at 4:21 AM | Permalink | Reply

    Funny that no one from the dutch readership of this site has mentioned that an interview with Steve appeared in the Dutch Volkskrant of last saturday. Could this be that author Martijn van Calmthout has so often misrepresented climate issues that this daily is not read so much anymore. Anyway, I do not read the Volkskrant, but a friend mailed me a scan of the piece. Below is my translation.


    Bodywarmer, beard, map of Amsterdam on the table, infamous climate-critic Steve McIntyre is in a trendy Grand Cafe and eats his Vietnamese chickensoup. Did he like to give a lecture to the KNMI (Dutch Metorological Org) end last week. It’s in the lion’s den isnt it?

    “I think I might have shown less then 100 slides. The questions afterwards didnt really surprise me, but then again that would be odd since I think I am the biggest expert in this field. Lion’s den, yes, but the lion was sleeping’.

    You have convinced them that the earth is not warming?

    “That is not the point. It is even quite plausible that it’s getting warmer. As a Canadian I know that many glaciers have shrunk this century’.

    Then what is the point?

    “That the researchers who make these themperature-reconstructions may be right but come up with evidence that is no evidence’.


    “They base themselves on the analyses of the widhts of tree rings of the past 1000 years. Warm years, thick rings, cold years thin rings. First several things are missing in their analyses. Worse is that the methods used produce the well-known hockeystick-curve by themselves: always a flat line that runs upward at the end. No matter what data you use. So it is an artefact.’

    Hasnt that been disproven already?

    “Not at all. The American NAS has recently established after hearings that the tree ring researchers have been too certain with their story about the warmest years in a millennium. It seems now especially difficult to acknowledge that errors have been made’.

    The NAS sees problems, but doesnt reject the hockeystick.

    “They point to other studies, but these are based on the same trees. Only meteorologists do not know that.’

    But they are also based on glacier studies like our Hans Oerlemens?

    “He never mentions 1000 years’.


    There are things that are not told unless you ask for it. That is remarkable. In the business community such a thing would be punsihed.’

    Your criticism has been embraced by climate-skeptics who think that the greenhouse effect is nonsense.

    “Correct, but that is a side issue’

    A side issue?

    “I know from the mining industry how difficult it can be to combine the measurement data of several locations. It is foremost a mathematical problem’.

    So the treerings could have been exploratory drillings as well?

    “Or something else. Is you story strong enough or not, that is what it is about’.

    The KNMI says not to be convinced after your visit

    “They are meteorologists. I did not have the impression that they really see the problem, nevertheless these are things that you can explain to your mother’.

    Is it strange that climatologists think you are rather unfriendly?

    “I hold on. If I do not get an answer I send another email. And another one. If I would have had to buy stamps I would never have started this hobby’.

    Martijn van Calmthout

  8. Steve McIntyre
    Posted Sep 25, 2006 at 6:25 AM | Permalink | Reply

    #6. I’m working on a detailed response to Wahl and Ammann- which actually has not appeared in print yet. I’ve got most of a draft done, which Ross and I will go over.

    #7. Perhaps it’s the translation and re-translation, but there are a number of points that didn’t come out the way I said them. For example, the statement about HS and red noise with the use of the term “artefact”. That’s not how I express the matter; it is how realclimate expresses the matter, when they put words into my mouth supposedly making the “primary” claim that the HS is “simply an artefact” of the PC method; they then argue that they can get a HS another way; ergo, our “primary” claim is disproven. My usual formulation is that Mannian PC methodology is biased towards selection of HS shaped series and that the flawed method intereacts with flawed data (bristlecones). There are other points in the article like this, but I’m not going to parse it.

  9. Ric Techow
    Posted Sep 25, 2006 at 6:56 AM | Permalink | Reply

    When we are talking about R2 figures that are approximately zero ah … ahem … just how low exactly are we talking about here???

  10. fFreddy
    Posted Sep 25, 2006 at 7:21 AM | Permalink | Reply

    Re #9, Ric Techow
    Have a look at this post.
    Also others in the category Wahl and Amman (see RHS sidebar).

  11. Chas
    Posted Sep 25, 2006 at 7:45 AM | Permalink | Reply

    Was the de Laat you met up with the author of “Industrial CO2 emissions as a proxy for anthropogenic influence on lower tropospheric temperature trends’?
    -It seems to be a really important analysis.
    But I keep wondering if the results are driven by northern/southern hemisphere differences in economic activity coincidental to the differences in temperature trends. Is there a good way to untangle it all?

  12. Steve McIntyre
    Posted Sep 25, 2006 at 7:49 AM | Permalink | Reply


  13. Chas
    Posted Sep 25, 2006 at 8:08 AM | Permalink | Reply

    The request part of #11 was silent :-)

  14. Ross McKitrick
    Posted Sep 25, 2006 at 9:18 AM | Permalink | Reply

    #6: At my web site (which needs updating) the 1st five bullet points provide links to summary/chronological articles that bring the M&M story up to spring 2005, and our NAS presentation in March 2006. Pretty much all the technical issues were on the table by early 2005, and are covered in our 2005 E&E article. I just re-read our backgrounder from Jan 27 2005 and it’s remarkable how it “responds” to the current thinking at KNMI as reported by Steve. The story since 2005 has involved some technical explorations, such as in response to Huybers’ GRL comment. Otherwise it’s been mostly process: the NAS panel, Wegman, the Barton letters/hearings, and other assorted attempts to articulate the findings in a variety of contexts.

    Down the page at my site I have links to the earlier pages where we chronicled our interactions with Mann, Nature etc. in 2003 and 2004.

  15. Steve McIntyre
    Posted Sep 25, 2006 at 9:43 AM | Permalink | Reply

    Ross, I looked back to our first Nature submissions, which included observations on the PC methodology and NH reconstruction in the same paper – which we later separated into two papers. Although people like Weber criticize this separation after the fact, von Storch in Stockholm mentioned that he agreed with our approach – he thought that the pragmatism of this approach was essential at the time. Maybe it would be different now, but probably not too much, as evidenced by his suggestion on presenting the impact of contamination in pseudoproxy networks without mentioning bristlecones.

  16. TCO
    Posted Sep 25, 2006 at 10:13 AM | Permalink | Reply

    1. I think some of the points (PC1 versus reconstruction, number of PCs) have validity and that you are to blame for overstating and misstating your case. Continuing to protect your honor with lame debate comments like citing the EE paper (telling the truth in EE doesn’t justify a lie/error in GRL, Steve), is a waste of time. Just concede valid points and move on. Your Huybers performance was disgraceful.

    2. Why was your presentation over-technical and over-time? For someone who touts his business background, this is completely unsat. Show some sympathy to your hosts and your audience. Learn how to give a talk. Did you really need to flop to learn that or couldn’t you read a book or do some practices with a critic, etc. The NASA pub (written in 1955) has good brief direction on the philosophy of science talks, Steve. You shouldn’t have to “rediscover North America” by learning on your own that talks are less technical then papers.

  17. fFreddy
    Posted Sep 25, 2006 at 10:37 AM | Permalink | Reply

    Re #16, TCO
    This is annoying.

    telling the truth in EE doesn’t justify a lie/error in GRL

    What exactly is a lie or an error in the GRL paper ?

  18. bender
    Posted Sep 25, 2006 at 10:56 AM | Permalink | Reply

    Very annoying.

  19. Steve McIntyre
    Posted Sep 25, 2006 at 11:14 AM | Permalink | Reply

    TCO – I’m getting tired of your baseless and inflammatory criticisms. I’ve deleted less inflammatory comments about Richard Courtney and I’m trying to decide why I shouldn’t delete your comments. I’m not going to spend time arguing with you.

    My presentations in Europe went fine, thank you very much. You weren’t there and have no idea. I think that I made a very decent impression on people and was far from being a “flop”.

  20. Barney Frank
    Posted Sep 25, 2006 at 11:44 AM | Permalink | Reply


    They go beyond annoying.
    A charge of dishonesty goes to a person’s fundamental character and trustworthiness and should be reserved for only blatant and obvious cases of duplicity, not flung about as though it is a trivial matter.

    If TCO can document actual, purposeful lies (not disagreements or differing interperatations or opinions) then he should do so. Otherwise he should stop defaming people’s characters. Not only is it dishonorable it’s not legal.

  21. Ken Fritsch
    Posted Sep 25, 2006 at 12:30 PM | Permalink | Reply

    re: #16

    TCO, I believe that you must have been convinced by now or at least should be well aware that the serious posters and readers (I will assume) at this blog have their own understandings and judgments of the issues to which you continue to make vague references and personal accusations of Steve M. Since you have been unwilling to be more specific when challeged to a discussion of these issues, the only excuse for you to keep bringing them up is your claimed “mentoring” interests in Steve M’s public image. If that is the case, please do all the rest of us a big favor and direct your comments to Steve M via email and he can then dispose of them as he sees fit.

  22. Dave Dardinger
    Posted Sep 25, 2006 at 12:36 PM | Permalink | Reply

    Re: TCO

    I agree that he has totally failed to get anyone here to understand what his complaints are. If just one person would “get it” and translate his complaints into something the rest of us could discuss that’d be one thing, but he hasn’t. And there’s tons of brain-power here. He needs to find someone to meet with in person and go through his problem with. Myself, I just skim over any of his messages recently where he gets into one of his mini-tirades.

  23. bender
    Posted Sep 25, 2006 at 12:47 PM | Permalink | Reply

    … yes, and mostly just for entertainment value.

  24. TCO
    Posted Sep 25, 2006 at 4:00 PM | Permalink | Reply

    Steve and community: an incorrect or misleading statement (and an inappropriate comparison is a misleading statement) in GRL is still an incorrect statement regardless of what was done in EE. Steve has A PATTERN of over-emphasizing the impact of “off-centering” on the hockey stick by referring to it’s impact on PC1. However “the hockey stick” is not the PC1! And guess what, the off-centering has a lot more impact on the PC1 then it does on the overall hockey stick. So Steve is running a game when he does this. Steve has shown this pattern:

    -in his GRL article (I’m not satisfied that EE article puts the most relevant comparison upfront either, but at least a toe drags over second base in that article…)
    -in multiple (misleading by failing to draw this distinction or better yet just answer directly with impact on the reconstruction) posts on this site
    -by failure to answer my clarifying question for months and until I raised it to a pain level and wiggling like a worm when pinned down.
    -In a comment to John Cross, where he called the PC1, “the reconstruction”. Then when corrected, instead of calmly thanking John and correcting himself, he kveteched at John for pushing the point.

    Note this type of behavior (on slightly different issues) is shown in Steve’s argumentative addressing of Preisendorfer’s n and in his (Huybers comment) failure to articulate that comparison of Mannian transform to a correlation matrix changes TWO variables (not just off-centering, but standard deviation dividing). Bottom line, ya gotta watch Steve a bit, he will overstate things, will debate defensively rather then being objective and phlegmatic.

    Steve, you want to start snipping my critiques of you, fine. It’s a censor of a relevant science/science policy issue though. My comments about your misleading style are meant entirely honestly from me, have been supported by example, are the result of a heck of a lot of reading on this site, and are in scope for discussion on this board (we talk about Mann’s proclivities). Even if you disagree, you should leave them up, or find yourself in the position of running a blog that does not allow disagreement.

    If you went over time and were over technical and covered new material with people who needed an explanation of the basics, then that is poor practice at a talk. Maybe flop is too harsh. But the basic point remains regardless if I say “flop” or “had a bad talk” or had a “not so good talk”. Don’t worry about the flop word. Worry about my basic point.

  25. Ken Fritsch
    Posted Sep 25, 2006 at 4:44 PM | Permalink | Reply

    Steve, you want to start snipping my critiques of you, fine. It’s a censor of a relevant science/science policy issue though. My comments about your misleading style are meant entirely honestly from me, have been supported by example, are the result of a heck of a lot of reading on this site, and are in scope for discussion on this board (we talk about Mann’s proclivities). Even if you disagree, you should leave them up, or find yourself in the position of running a blog that does not allow disagreement.

    TCO, you have done it again with your vague references that have not even come close to making a case for your accusations. You then turn around and make a threat that you should be allowed to make these vague accusations or Steve M will be publicly condemned for censorship. You are obviously attempting to game the process and in doing so you add little or nothing to discussion, not even any entertainment value.

    Your approach may work in some areas. I just wish it did not work here. You seem to have a need for attention and this your way of getting it. Please consider some alternatives. If your were on a sincere search for truth, I would not think you would be so vague.

  26. Hans Erren
    Posted Sep 25, 2006 at 4:47 PM | Permalink | Reply

    TCO shut up.

  27. Steve McIntyre
    Posted Sep 25, 2006 at 4:56 PM | Permalink | Reply

    TCO, I disagree with you. The interest in the PC1 arose out of the reconciliation between MM03 and MBH98. In Mann’s internet response, he identified the PC1 as the difference. This stuff has been discussed over and over. If any of your concerns had any merit, you can bet your boots the NAS panel would have raised them.

    Whatever my other sins may be, I’m pretty reasonable with critics and I doubt that many people would accuse me of “kvetching” at critics. I don’t recall the John Cross situation that you speak of. I just searched “John Cross” in past threads in Admin mode and couldn’t find anything remotely like it.

    BTW I can’t possibly deal with all comments on this site.

  28. TCO
    Posted Sep 25, 2006 at 4:57 PM | Permalink | Reply

    #20 To my knowledge, Steve has never made a deliberate lie. He has obfuscated issues and allowed misunderstanding to form or to remain which serve the effect of “improving his case” and has defended himself from criticisms or answered probing questions by sidesteppting the immediate issue and adressing other ones. I consider this dishonest.

  29. Steve McIntyre
    Posted Sep 25, 2006 at 5:03 PM | Permalink | Reply

    TCO, you take my breath away. I am 100 times more detailed and more patient than any other person in this field and you give this crap. Especially given the amount of disinformation and crap from Wahl and Ammann, Mann, realclimate etc.

    Our statements on the connection of the PC1 and the reconstruction have always been very clear. Here’s a typical example:

    Although considerable publicity has attached to our demonstration that the PC
    methods used in MBH98 nearly always produce hockey sticks, we are equally
    concerned about the validity of series so selected for over-weighting as temperature
    proxies. While our attention was drawn to bristlecone pines (and to Gaspé cedars) by
    methodological artifices in MBH98, ultimately, the more important issue is the
    validity of the proxies themselves. This applies particularly for the 1000–1399
    extension of MBH98 contained in Mann et al. [1999]. In this case, because of the
    reduction in the number of sites, the majority of sites in the AD1000 network end up
    being bristlecone pine sites, which dominate the PC1 in Mann et al. [1999] simply
    because of their longevity, not through a mathematical artifice (as in MBH98).

    Given the pivotal dependence of MBH98 results on bristlecone pines and Gaspé
    cedars, one would have thought that there would be copious literature proving the
    validity of these indicators as temperature proxies. Instead the specialist literature only
    raises questions about each indicator which need to be resolved prior to using them as

  30. Ken Fritsch
    Posted Sep 25, 2006 at 5:11 PM | Permalink | Reply

    BTW I can’t possibly deal with all comments on this site.

    And that is what TCO is evidently counting on. Throw out many vague accusations and evidently then accuse you of sidestepping the issues. In my eyes TCO is showing a very lazy streak of irresponsibility by not clearly stating his reasons for his accusations. I think there should be show down where he can either clearly state his case or he drops it or he, in a gentlemanly fashion, goes away.

  31. Steve McIntyre
    Posted Sep 25, 2006 at 5:18 PM | Permalink | Reply

    I’m also not interested in debating things over and over, although I do so more than I want. I’ve got nothing left to say on, for example, Huybers. I think that our Reply to Huybers was quite reasonable. BTW the NAS Panel effectively agreed with our position that there was no mathematical or statistical basis for preferring correlation PCs to covariance PCs and that any of these results needed to be established scientifically.

  32. TCO
    Posted Sep 25, 2006 at 5:26 PM | Permalink | Reply

    Steve, do you really want to measure yourself against them? My standard is the response I would expect from a technician working on a problem: a forthright one and not an argumentative one.

    You make a pattern of:
    1. “Publishing in a protected environment” versus one with better critical examination of your arguments.
    2. “Publishing” incomplete criticisms. (These blog posts.) Thus making it harder to examine and refute/confirm them.
    3. Overstating your case by not attaching caveats on points that are arguable either way (Preisendorfer’s n, correlation versus covariance matrix) and that might impact your result. This is doubly damning since, you don’t even assert for sure a preferred method: when pinned down–you are ambivalent on those choices–one more reason to give the caveats, Steve.
    4. Conflating issues (faults).
    5. Responding to in depth examination of a particular issue by trying to shift the discussion to another one.

  33. TCO
    Posted Sep 25, 2006 at 5:29 PM | Permalink | Reply

    I’m also not interested in debating things over and over, although I do so more than I want. I’ve got nothing left to say on, for example, Huybers. I think that our Reply to Huybers was quite reasonable. BTW the NAS Panel effectively agreed with our position that there was no mathematical or statistical basis for preferring correlation PCs to covariance PCs and that any of these results needed to be established scientifically.

    This is so twisted, I’m not sure if you are being stupid or dishonest. The issue is not preferring one to the other, the issue is that you showed the one that made your case look better rather then showing both and that the one you showed, changed TWO factors at the same time, rather then the one of interest: “off-centering”.

  34. TCO
    Posted Sep 25, 2006 at 5:33 PM | Permalink | Reply


    a. Understand your disagreement. I still disagree with your disagreement. My comments stand on the PC1. You have been asked about mining causing the hockey stick and responded with impact on the PC1. This is misleading. Here is the ref to Cross:

    b. The comment about NAS would have addressed the PC1 versus recon is a complete example of your being either misleading or blind. You’ve already indicted them for their lack of thoughfullness, for broadening their mission beyond the MM/Mann controversy, etc. Very weak and bogus comment, Steve.

    c. I agree that you can’t answer every question and that I tend to non-trivial essay questions as a mode of discussion. That said, my question “is PC1 the same as the hockey stick” can be answered with one word: “no”. You are much more responsive to answer questions that will show Mann as bad then that will limit the impact of your criticism. That’s my impression from interacting with you over the months. A true scientist would just want to clariy things. You are more of an advocate then a true scientist.

  35. Steve McIntyre
    Posted Sep 25, 2006 at 5:48 PM | Permalink | Reply

    TCO – just for the record one more time: we started off by simply doing a conventional PC analysis of MBH networks and getting different results. A “conventional” PC analysis in this type of network was a covariance analysis. When we started, we had no idea what Mann had done – only that we couldn’t replicate his results. Later we found out that he hadn’t done a conventional PC analysis, but had both off-centered and divided by the detrended standard deviation. We described all this. We didn’t “try” to change two things at once – Mann was the one that had changed two, actually three things. We described all this.

    You’ve quoted your own comment in #34, not anything that I said. What am I supposed to have said to John Cross?

  36. TCO
    Posted Sep 25, 2006 at 5:55 PM | Permalink | Reply

    I’m looking for it, Steve. It was a while ago and I can’t remember the thread. You said something to the effect that he had ruined a good point by how he pushed it. I think it was Cross, not Hunter. Is there another John?

  37. TCO
    Posted Sep 25, 2006 at 6:00 PM | Permalink | Reply

    35. That’s what I have a problem with. If you are looking at OFF-CENTERING, you should change only that. Changing 3 things at one time is crappy analysis, and is misleading if you just tout one of the changes (makes it look like all the effect comes from the uncontested fault rather then the non-fault, arguable choice*).

    If you don’t know what is going on, fine. When you do or when someone else does, then just say, ok, now we know better.

    * BTW, you can’t have it both ways (except in the most tendentious, lawyerly way), arguing equality “there was no mathematical or statistical basis for preferring correlation PCs to covariance PCs” and “covariance is conventional”.

  38. TCO
    Posted Sep 25, 2006 at 6:18 PM | Permalink | Reply

    I’m not sure if this is what I was thinking of and it was Hunter, not Cross. Thought I remembered your actually misusing the term, itself. Let me look some more.
    *Hate the sidebar*
    *Hate the sidebar*
    *Hate the sidebar*
    *Hate the sidebar*
    *Hate the sidebar*

  39. bender
    Posted Sep 25, 2006 at 7:02 PM | Permalink | Reply

    TCO, if you’re going to call someone dishonest, you had better have a full command of the facts, which you NEVER do.

  40. TCO
    Posted Sep 25, 2006 at 7:05 PM | Permalink | Reply

    I can’t find that one reference right now, so scratch it from the list (until I find it). The rest stands, though. And what I did find still shows the same practice. Care to adress that stuff?

  41. Ken Fritsch
    Posted Sep 25, 2006 at 7:08 PM | Permalink | Reply

    We talked for a while about “data snooping” as the term is used in economics — if you know what the series look like as a result of a literature search, including searches by prior authors, then usual statistical tests have no meaning. She either disagreed with this point or didn’t understand it — she definitely didn’t agree with it.

    We talked about failure of verification statistics. Weber said that Juckes was their statistics guy; he knew about r2 and RE, while she wasn’t much interested in statistics.

    Surprising how otherwise knowledgable people do not “get” the data snooping issue. It is almost as if the more you truly believe something and data snooping finds it for you that the harder it is to see the problems connected with data snooping.

    Not understanding or wanting to understand a near 0 for an R^2 would seem to me much less excusable.

  42. bender
    Posted Sep 25, 2006 at 7:16 PM | Permalink | Reply

    Data snooping is an accepted part of climate science culture. It’s as simple as that. You can’t experiment on planet Earth; all you can do is observe. So what do you need statistics for, right? It’s Bloom’s “count ‘em, five fingers, no error” argument.

  43. Ken Fritsch
    Posted Sep 25, 2006 at 7:39 PM | Permalink | Reply

    I can’t find that one reference right now, so scratch it from the list (until I find it). The rest stands, though. And what I did find still shows the same practice. Care to adress that stuff?

    You cannot find a specific reference and than you allude to the rest stands as if you proved your case there. I am sincerely thinking that if you have problems remembering these issues you should stop the accusations. The referenced link that you gave showed me that the question of the PC1 versus the total reconstruction was answered by Steve M and exhaustively discussued for anyone to reach a judgment on it.

    Your “care to address that stuff” statement demonstartes once again just how lazy and imprecise you are.

    Maybe you should withdraw from the discussions at this blog for a while so you could prepare and gather evidence for your case against Steve M. What you are doing takes no intelligence, no logic and no effort.

  44. TCO
    Posted Sep 25, 2006 at 7:40 PM | Permalink | Reply

    I gave several specifics, Ken.

  45. Ken Fritsch
    Posted Sep 25, 2006 at 8:05 PM | Permalink | Reply

    re: #44

    I gave several specifics, Ken.

    I am ending this discussion by saying that one of us must be delusional on this topic and that makes the whole excercise futile. If you think that your reply in comment #37 is specific, clear and concise and can get others here to agree, I will go back on my meds.

  46. bender
    Posted Sep 25, 2006 at 8:07 PM | Permalink | Reply

    “Stuff” is specific. Isn’t it?

  47. Dave Dardinger
    Posted Sep 25, 2006 at 8:10 PM | Permalink | Reply

    re: #44

    I gave several specifics, Ken.

    If you’re talking # 32 you gave several generalities.

    Picking point 4 there at random

    4. Conflating issues (faults).

    This doesn’t give a citation either here or in a paper. Nor does it even try to prove the point. When I’ve seen you make this complaint before, it’s because Steve brings up two or more problems with something, like the off center PCs and the lack of scientific backing for BCPs as temperature proxies in MBH9x. Now the NAS panel, and Wegman both agree that the off-center PCs are incorrect. And NAS and any sensible reading of the literature, (such as the Graybill and Idso 1993 paper that is presently under discussion) show that it’s quite unlikely that BCPs and particularly strip bark ones are suitable as temperature proxies. Therefore Steve was and is correct on BOTH points there. And that’s generally or always the case. He makes two valid complaints and you complain that he’s “conflating” the points. This is wrong for you to do and unless you can actually demonstrate your point you should stop doing it.

    Just because you can’t separate out various lines of argument doesn’t mean that others can’t.

  48. TCO
    Posted Sep 25, 2006 at 8:21 PM | Permalink | Reply

    Ken and bender: I was referring to the bulleted points in 32.

    Dave: Agreed that these are not documented specifics, but more subareas. I have in the past weighed in on the specifics to support them. The info is on this blog. Do you really contest or not know it? I guess I forget that not everybody has read every word of this thing as I have.

    P.s. If you require me to buckle down and write some essay with footnotes and such, I better not hear any whining about how I keep pursuing topics in a tedious manner…

  49. James Lane
    Posted Sep 25, 2006 at 8:24 PM | Permalink | Reply

    TCO, I think Steve’s position on Preisendorfer’s N is pretty clear. What’s your beef there?

  50. TCO
    Posted Sep 25, 2006 at 8:25 PM | Permalink | Reply

    Dave, I really do think that Steve conflates issues. I have tried pursuing a single one and seen him respond with other ones or multiple ones. The topic adressed had already been restricted to one and he would not answer with just that one.

  51. TCO
    Posted Sep 25, 2006 at 8:35 PM | Permalink | Reply

    49. My beef is that the impact of HS mining (by off-centering*) is small for a reconstruction (as opposed to PC1) that is made up of Preisendorfer’s n of PCs. Steve can say that Mann did not adequately document Preisendorfer’s n or even that he suspects it is after the fact rationalization, but he has no proof of that and it is definitely arguable that n is a condition of the work (there is a sentence on it in one section of MBH.) At a minimum, to be “Feynman fair”, Steve needs to clarify what his offcentering mining impact is with preisendorfer’s n. I don’t like how Steve wants to have his cake and eat it too: not taking a firm position on the Preisendorfer issue and then not showing the impact if it is retained.

    *only, not standard deviation dividing. And I hold it to that as Steve has not staked out a position against correlation matrix. Off centering is his crown jewel of rhetorical effect, but it’s effects are smaller then one would suppose from all the noise.

  52. Dave Dardinger
    Posted Sep 25, 2006 at 8:53 PM | Permalink | Reply

    re: #50

    I have tried pursuing a single one and seen him respond with other ones or multiple ones.

    Steve says, “The sun rose in the east this morning.” You complain, “But you forgot to mention that it isn’t an equinox and so it didn’t rise exactly in the east. This could be important and your failure to mention it is dishonest.” Steve trys justifying his position by pointing out that EE05 described the sunrise measurement procedures and give a historical summary dating back to Stonehenge. You reply, “Why did you post in EE rather than Science? If you’d simply written better they’d have accepted it.” Steve mentions that both the Bible and the Communist Mannifesto agree that the sun rises in the East and you reply, “That that doesn’t matter. Stop changing the subject and keep genuflecting before me!”

    As you may have noticed, about all of us regulars have had it with your antics. A single person getting on your case could just be a personality conflict. When everyone is on your case, introspection is called for.

  53. TCO
    Posted Sep 25, 2006 at 8:56 PM | Permalink | Reply

    DD: I have pursued a specific issue of math/methodology and seen Steve wanting to involve other issues. I really have. Would you like to bet? He lists x1-5. I want to pursue X3 in depth (and perhaps to some discomfort for him) and he wants to shift to discussing the others rather then laser siting in on the one that I’ve already predicated a question on.

  54. James Lane
    Posted Sep 25, 2006 at 9:25 PM | Permalink | Reply


    Steve can say that Mann did not adequately document Preisendorfer’s n or even that he suspects it is after the fact rationalization, but he has no proof of that and it is definitely arguable that n is a condition of the work (there is a sentence on it in one section of MBH.)

    Steve has shown that MBH did not apply Preisendorfer’s N to the retention of proxy PCs in the original paper. That’s pretty good evidence that subsequently resorting to Preisendorfer was a post hoc effort to avoid the implications of the centering issue. I anticipate that you’ll now say “that doesn’t change the fact that Preisendorfer says the PC4 should be retained”. That’s true up to a point. There is no hard and fast rule about retention of PCs – Preisendorfer, eigenvalue rule, scree test etc. Mann would be on much stronger ground if he had applied any of these rules consistently, but, in fact, PC retention for the proxies appears to be completely arbitrary in the original paper.

    There is also a big difference between a PC which explains a large proportion of the variance and a lower order PC that explains 8% of the variance. A related conceptual issue is that if you claim that the PC4 represents the temperature signal, what then do the higher order PCs represent?

    At a minimum, to be “Feynman fair”, Steve needs to clarify what his offcentering mining impact is with preisendorfer’s n.

    Steve doesn’t need to do this. Realclimate has done it. You use Preisendorfer to get the PC4 into the regression stage, and bingo, the reconstruction is unaffected (but on much shakier conceptual foundations).

    Re the correlation/covariance matter, I have used PCA extensively, and I can assure you that the “convention” is to use covariance matricies. There are situations where the correlation matrix might be preferred, but MBH doesn’t appear to be one of them. MBH claimed that “conventional PCA” was used. It’s not incumbent on Steve to show why covariance should be used, but rather for MBH to make a case for using correlation. In any case, it’s a moot point, as the analysis should be robust to either choice.

    I doubt that the above will satisfy you, but it does it for me.

  55. bender
    Posted Sep 25, 2006 at 10:29 PM | Permalink | Reply

    It won’t satisy him because I’ve told him this before. And I wasn’t the first either.

  56. bruce
    Posted Sep 26, 2006 at 1:45 AM | Permalink | Reply

    TCO, I don’t particularly want to pile in to the discussion here re your approach to the blog, but here goes anyhow.

    I have no doubt that you are well intentioned, and knowledgeable, but it seems to me, FWIW, that rather than continually needling Steve M, that you might do better to use your undoubted talents, and wide reading of the subject, to address the points that you make, and even write papers. Perhaps do that on your own, or alternatively team up with someone with complementary viewpoints and interests.

    Remember, it wasn’t that long ago that there were no climate scientists, so most of us with a scientific bent should feel free to pitch in and help Steve M with some of the heavy lifting.

    I do enjoy your post prandial posts – you seem to have been much more moderate lately, or have you (as I have) given up posting after I’ve had a drink or two!

  57. Hans Erren
    Posted Sep 26, 2006 at 3:28 PM | Permalink | Reply

    The wine was not some ordinary Burgundy it was a Pinot Noir 2003 proxy

  58. Kevin
    Posted Sep 26, 2006 at 3:46 PM | Permalink | Reply

    #54: James, I have been using PCA for 25 years and it is in fact rare to run it off a covariance matrix. The correlation matrix is the default in every stats package I have have seen and using the covariance is rarely useful. Multi-group SEM, which is conceptually related to PCA, would be an exception.

    In any event, PCA is not appropriate for time-series data whichever matrix you use.


  59. Willis Eschenbach
    Posted Sep 26, 2006 at 4:02 PM | Permalink | Reply

    Re #58, Keven, it would be great if you could expand on the reassons for your statement that:

    In any event, PCA is not appropriate for time-series data whichever matrix you use


  60. bender
    Posted Sep 26, 2006 at 4:46 PM | Permalink | Reply

    Re #59
    In part because the PCs are in no way contrained to contain the same kind of complex autocorrelation structure that is inherent in the original time-series. Because the original observations are non-independent the PCs are bound to be highly idealized representations of a highly fictitious ensemble. That might matter, or it might not. Depends on context.

    Not sure if this was Kevin’s primary concern, but it is a legitimate concern.

  61. Willis Eschenbach
    Posted Sep 26, 2006 at 7:13 PM | Permalink | Reply

    Re 60, thanks, bender, very clear. If kevin has other concerns I’d like to hear from them as well.


  62. James Lane
    Posted Sep 26, 2006 at 8:09 PM | Permalink | Reply

    #58 Kevin, that’s interesting. I was taught to prefer covariance way back when. I can’t recall why, it might have had something to do with the nature of the data I was using. The package I used, Systat, didn’t have a default setting as far as I can recall.

    It was actually my experience with PCA that got me interested in MBH. I was astounded to see it used in the way it was.

  63. bender
    Posted Sep 26, 2006 at 8:41 PM | Permalink | Reply

    Recalling that Cor(X,Y)=Cov(X,Y)/(Var(X)(Var(Y)) …

    If the variances among variables differs much (e.g. if the proxies are expressed in different units) you should definitely use the correlation method as this will standardize inter-proxy covariances by proxy variances.

  64. James Lane
    Posted Sep 26, 2006 at 8:54 PM | Permalink | Reply

    #63 Yes, I just realised that. In the applications I was using PCA, the variables were invariably in common units. (Tree ring series are also in common units).

    There’s a discussion of correlation vs covariance (in relation to Huybers) here:

  65. Steve McIntyre
    Posted Sep 26, 2006 at 10:02 PM | Permalink | Reply

    Posted for Kevin stuck behind a firewall:

    59,#60: PCA on TS data is very risky for the same reason it is ill-advise to correlate two time series: both series will be contaminated by their ARIMA structure and this can cause spurious correlations. This is why Econometricians frequently use pre-whitening in Transfer Function (causal) modeling. I think it was Yule in 1927 who first pointed out the problem. As you can imagine, a correlation matrix only compounds matters. Think of it as latent variables we are ignoring.

    A further difficulty is the lack of contemporaneous structure. There will nearly always be lags in the data and, while identifying lags between two series can be done (after pre-whitening), with multiple series this is very difficult.

    This is not nitpicking by a statistician; this is a real world headache in time series modeling with no easy solution I know of.

    I suspect Wegman et. al. did not comment because they already had their plates full!

  66. Steve McIntyre
    Posted Sep 26, 2006 at 10:18 PM | Permalink | Reply

    The NAS panel discussed covariance versus correlation as follows (North et al 1982 presumed that covariance was appropriate for common units):

    This exercise demonstrates that the baseline with respect to which anomalies are calculated can influence principal components in unanticipated ways. Huybers (2005),
    commenting on McIntyre and McKitrick (2005a), points out that normalization also affects
    results, a point that is reinforced by McIntyre and McKitrick (2005b) in their response to
    Huybers. Principal components calculations are often carried out on a correlation matrix
    obtained by normalizing each variable by its sample standard deviation. Variables in different physical units clearly require some kind of normalization to bring them to a common scale, but even variables that are physically equivalent or normalized to a common scale may have widely different variances. Huybers comments on tree ring densities, which have much lower variances than widths, even after conversion to dimensionless “standardized” form. In this case, an argument can be made for using the variables without further normalization. However, the higher-variance variables tend to make correspondingly higher contributions to the principal components, so the decision whether to equalize variances or not should be based on the scientific considerations of the climate information represented in each of the proxies.

    They should really have added a sentence that the decision whether to use PC or not needed to be established on scientific grounds. In experimenting with pseudoproxy networks, I haven’t found any situations where PCs lower than the PC1 contain relevant information about an overall “signal”. In the pseudoproxy networks, these lower order PCs tend to be local oscillations which cancel out in the average. You don’t get distinct loadings on a subset (with negligible loadings otherwise) as in the MBH covariance PC4 or correlation PC2. PC methods are extraordinarily “non-robust” in a statistical sense in that they preserve contaminants for a regression as distinct entities rather than downweighting them. This is true even for conventional PC methods.

  67. fFreddy
    Posted Sep 27, 2006 at 12:52 AM | Permalink | Reply

    Re #63, bender
    Sorry, shouldn’t there be a square root on the bottom of the RHS there ?

  68. bender
    Posted Sep 27, 2006 at 1:32 AM | Permalink | Reply

    PC methods are extraordinarily “non-robust” in a statistical sense in that they preserve contaminants for a regression as distinct entities rather than downweighting them. This is true even for conventional PC methods.

    Not sure exactly what you mean by “contaminants for a regression”, but (assuming I follow) … isn’t this non-robustness because these multiproxy networks are so incredibly sparse, with signal carrier and contamination effects highly regionally confounded (bcps ONLY in CA, corals ONLY in AUS, etc., with huge gaps between both samples & signal carrier types)? Thus you have isolated regional signal (+comtaminants) arrayed against isolated regional signal (+comtaminants), with no opportunity for separating common climate signal (large PCs) from uncommon proxy contaminant (small PCs). If so, then what you really need is spatial intermingling of signal carriers: corals in Arizona, bristlecone pines in Australia, etc.

    That’s not PCA’s fault; it’s the experimenters’.

  69. bender
    Posted Sep 27, 2006 at 1:36 AM | Permalink | Reply

    Re #67
    Yes. Thanks.

  70. James Lane
    Posted Sep 27, 2006 at 1:44 AM | Permalink | Reply

    bender, I think what Steve is talking about is that the regression stage in,say MBH, doesn’t “care” about the ordering of the PCs or the proportion of variance explained. Thus PC4 is treated as equally as PC1.

    I’d love to see some of those Australian bristlecone pines! There are some old trees in Tasmania, but not in the league of the BCPs. Perhaps TCO would like to come down and section them!

  71. bender
    Posted Sep 27, 2006 at 2:15 AM | Permalink | Reply

    I think I see what you’re saying, James, You’re critizing the logic: “PC4 is small, but who cares – let’s feed it into the regression model anyways because it’s interpretible. Meanwhile let’s not bother trying to interpret PC1, even though it’s way bigger than PC4.”

    If that’s what you’re saying, I agree. But I don’t think that’s what Steve is referring to. Steve doesn’t mention regression at all in the bit about PCA non-robustness. (Let’s let him answer. Maybe you’re right.)

    (You know I was joking about the AUS bcps, right? ok. You never know.)

  72. James Lane
    Posted Sep 27, 2006 at 3:08 AM | Permalink | Reply

    Yes bender, that’s exactly what I was saying, maybe Steve has a different take. And I know you were joking, no worries, mate! My joke was really about TCO trying to section a Tasmanian tree – down here that would be as acceptable as strangling the Pope, maybe less acceptable. Weak joke, oh well.

  73. Steve McIntyre
    Posted Sep 27, 2006 at 8:33 AM | Permalink | Reply

    JAmes Lane has my sense right. Let’s suppose that you have a non-HS climatic signal and a nonclimatic “contaminant” HS in the network going into PCA. “Robust” statistics worries even about averages in a network – preferring things like trimmed means or medians or things like that. But if you do PCA on a network with (N-1) series being signal plus white or low-order red noise and one-two HS-shaped non-signal series, then the HS series will stamp a high-order PC and seem “significant”. PC analysis preserves the non-signal series because it’s a distinct pattern.

    Now the assumption that there is a “signal” in the majority of MBH proxies is not a correct assumption. If you take the non-BCP proxies in the early network and do MBH regression (partial least squares), you don’t out-perform a similar operation using low-order red noise series. You get calibration r2 over 0.5; verification r2 of ~0.

    If you add 1-2 HS shaped series (BCP plus Gaspe) into MBH regression, you get a tuned reconstruction with a HS shape. The bend in the HS is not as big as the bend in the PC1 but the PC1s tend to “overshoot” in RE terms. So by blending the HS with noise, you can tune the reconstruction to fit the NH temperature history and get a high RE result. So if you mix 1-2 synthetic HS series with networks of low-order red noise, you get the MBH verification statistic package – high calibration r2, verification r2 of ~0 and RE of about 0.4 or so.

    MBH regression on large networks is an amusing little can of worms that I’ve alluded to here and really need to write up formally. IT comes nicely into focus in the Wahl and Ammann variation of MBH where (without knowing that this is what they’re doing) they do partial least squares regression of one temperature series of length 79 on 95 series with minimal common signal. Wahl and Ammann argue that their “fit” from this procedure entitles them to extend the proxies using the instrumental record. It’s pretty comic once you see what they’re doing. The Team are always full of such jolly pranks.

  74. bender
    Posted Sep 27, 2006 at 9:38 AM | Permalink | Reply

    Re #73
    Then my #68 take is quite different, but still relevant. Each proxy type is going to be uniquely subject to its own peculiar brand of bias/error/contamination. If proxies vary from one data location to the next, then you have no way of separating those errors from the regional climatic signal. They become part of the regional signal. Whereas the intent of PCA is to factor out the common signal as large PCs, relegating the bias/error/contamination to small PCs. If there’s no spatial intermingling of proxy types (because the grid is sparse and proxy types are regionally unique), that isn’t going to happen. PCA becomes trivialized.

    I’m sure the multiproxy people have figured this out – it’s kind of obvious. There’s not much you can do about it – except use caution in interpreting results. Which doesn’t necessarily happen when your modus operandi is alarmism.

  75. Steve McIntyre
    Posted Sep 27, 2006 at 9:49 AM | Permalink | Reply

    From my experiments with pseudoproxy networks, it seems to me that, if you’ve got a network with an actual signal, it doesn’t matter too much what multivariate method you use, you get something that looks like the signal – whether you use PCA, averaging,.. unless you go to extreme overfitting through inverse OLS. PLS (or Mannian regression) isn’t the worst, but does overfit. The issue is when you insert records that have nothing to do with the signal and then try to determine how the multivariate method performs – that’s where MAnnian methods really fall apart.

  76. bender
    Posted Sep 27, 2006 at 11:43 AM | Permalink | Reply

    Setting aside the issue of Mannian methods I believe there is still a problem with PCA itself when the multiproxy data network is sparse and the various proxy types regionally concentrated, not spatially intermingled.

    I have a hunch that proxy-based reconstruction is highly proxy-dependent, and that that is why the regional reconstructions tend to be fairly orthogonal and thus preserved through the PCA process. If the signal-carriers were true proxies, they should be highly non-orthogonal, i.e. correlated, under global cliamte change. That they’re not tells you something. Either the climate signals are truly regional (and there are millions of subtle teleconnections all around the globe), or the contamination is strong, such that all you can extract are regionally impure signals; the pure global (or hemispheric-scale) signal is non-extractable.

    Remember that even the very best proxies have relatively low skill. I have a hunch that that’s the hole that the MWP has disappeared into.

  77. Phil B.
    Posted Sep 27, 2006 at 1:39 PM | Permalink | Reply

    Re #76, bender in reality, using optimal estimation techniques under certain assumptions, one should always get a hockey stick temperature reconstruction with poorly correlated proxies and the existing instrumental temperature rise in the 20th century.

    The limiting case for the temp reconstruction is no proxies. So if one assumes the NH temperature series is ergodic gaussian wide sense stationary, the optimal estimate for the years 1500-1850AD is the sample mean (-.17C) of the 150 samples (1850-2000AD) of existing instrumental data with an estimation error variance equal to the sample variance for the 150 years of data (~.25C^2). Noting that this optimal estimate uses equations 1-3 from T. Schneider JC 2001 paper which I have seen called the LS method.

    Now if one adds a standardized proxy with low r, than equation 1 justs adds some wiggles to the hockeystick and reduces the estimation error variance to (.25^2)*(1-r^2). Just a few thoughts.

  78. TCO
    Posted Sep 28, 2006 at 7:51 PM | Permalink | Reply


    I finally read the presentation. It is a very nice presentation in general. Some very pretty slides at the beggining that deliver a great story and allow you to talk without too much words on the ppt. Just using a visual aid.

    Further on, there are some that are too technical (in keeping with audience and rest of talk).

    Also, it seems it would take 3 hours to cover all this. In all seriousness, what was your time limit?

    There is also one slide where you talk about PCA “mining for hockey sticks”. Since you have called the recon itself, “the hockey stick” and have been criticized for muddling pc1 versus recon to overemphasize the impact of mining, you really ought to stop such remarks, Steve. It is a slight dishonesty. I would say it to your face…

  79. rafa
    Posted Jun 13, 2009 at 10:06 AM | Permalink | Reply

    “…I offered to bet 500 Euros that their CPD submission would use the Yamal series, rather than the equally “available” Polar Urals. She didn’t take the bet”

    Why am I not surprised?

Post a Comment

Required fields are marked *



Get every new post delivered to your Inbox.

Join 2,881 other followers

%d bloggers like this: