It often feels like shoveling out a swamp in dealing with the misrepresentations of our stuff. Someone over at Tim Lambert has said that I “originated incorrect information” about Mann’s CENSORED directory:
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.
Here are the facts.
The uncensored directory ftp://holocene.evsc.virginia.edu/pub/MBH98/TREE/ITRDB/NOAMER/BACKTO_1400 (“url/BACKTO_1400”) contains 70 North American tree ring sites. The ITRDB codes for 70 sites are listed in the file url/BACKTO_1400/noamer-itrdb-ad1400.txt sites (see here). The url/BACKTO_1400-CENSORED directory does not contain a file with a listing of sites, but the file of eigenvalues BACKTO_1400-CENSORED/eigen.out (see here) shows that the PC calculations in this directory were done with 50 series. What are the 20 censored series?
If one compares the two files url/noamer.inf (here), which has 212 series, and url/noamer-censored.inf (here), which has 192 series, one can obtain the codes for 20 excluded series. All 20 excluded series are present in the 70 sites in the uncensored BACKTO_1400 network.
What are the 20 censored series? 19 of the 20 are were collected by Donald Graybill and 1 by Connie Woodhouse. Graybill was specifically trying to identify CO2 fertilization in this data set, which was reported on in Graybill and Idso [1993]. Of the 20 censored series, the majority are listed in Graybill and Idso [1993]. 16 of 20 are bristlecone/foxtail series; 4 are high-altitude strip-bark limber pine. I collectively refer to these strip-bark sites as “bristlecones” (with appropriate definition in published material.) A listing of the censored files is here.
I don’t know why people make such categorical statements about things where they lack knowledge.
Climate Audit 
16 Comments
Interesting priorities.
I myself would be reflexively suspicious of reasons why files were are tagged CENSORED.
But that’s me..
So Lambert is wrong? It’s 20 series removed, not 70?
Thanks for the info, Steve. I hadn’t actually checked myself, and simply assumed the claim (by one of the commentators at Deltoid, not Tim Lambert himself) was correct. I get the feeling most of the claims on Deltoid are simply uncritical repetitions of Mann’s responses, which (as we know) are mainly straw men.
I’ve found it is virtually impossible to get across the point that if the “temperature” reconstructions are not robust, then they are statistically meaningless. They continually use both your plots and Burger and Cubasch’s plots as alternative temperature reconstructions, rather than illustrations of problems with data and methodology. Unfortunately, bad papers such as W+A help to encourage this kind of misunderstanding.
I suppose I will get yelled at, but isn’t it fair to say that, in certain respects, Ross’and Steve’s work does show an alternative temperature reconstruction–one without the influence of bristle cone pines? I think I can see why there is a propensity to treat the work as an “alternative” construction. Of the “reconstruction” is completely non significant, and therefore worthless.
That’s why I suspected, possiblly incorrectly, that Steve could get to 0.2 degC at the last millenium.
=================
JAE,
It’s all about what one claims. Steve and Ross did not claim to present an alternative reconstruction. I mean, how could they when they are doubtfull that tree-rings are even valid temperature proxies? They made it clear that they were testing a specific claim made by MBH98: that the results would not change if some proxies were removed. That’s perfectly valid, but characterizing that test as an alternative reconstruction is dishonest because that was not the stated purpose of the exercise.
Agreed Paul. But it is relevant to drill down and see the extent/accuracy of methodological kvetches if MM choose to make them. And it’s not right if they want to dodge and say “well tree rings suck anway since moisture is not accounted for” when we are trying to do an examination of their claims on a methodological concern that they have raised. (I’m not saying they went quite this far, but one of their “defenders” did try to do this, when he was chiding me.)
TCO,
While I agree that to falsify a paper like MBH98 you have to not only show that it is flawed, but you have to quantify the results of the flaws so that the reader can come to a conclusion as to whether these flaws are serious enough to render the entire paper useless or not. However, I don’t think you need to show where the flaws come from, in a theoritical sense. It’s nice if you can, but it’s not necessary. I think this is possibly where we disagree.
Steve has made the argument that there are multiple things which damn the study. For instance even if BCPs are “good” proxies, that they are overweighted. When he says things like that, it is reasonable to drill down on the issue of overweighting. How overweighted? In the actual HS, versus “PC1” (and I want the info in HS)? If I ask questions like that and the answer is “sorry, we don’t know*, but really who cares since the BCPs are lousy proxies from CO2/soil fertilization”, then I feel sorta cheated. Why cite the initial complaint if you won’t back it up and run to a different one when pushed? The only thing that would make the situation more vexing would be if I pushed on the details of the CO2/soil blather and got a redirect to weighting concerns!
I actually think my (painful to watch, I imagine) interrogation is that we find out things that I bet lots of you here didn’t realize…like the “PC1” caveat of Ross, when what matters is weighting in the HS. Or that the major damage is done from lack of geo-weighting rather than from PC methods. Or that PC methods themselves do way more damage than “offcenteredness”. The good thing about learning things like that is that we can then concentrate the debate and the learning on the key issue. For instance we can come to grips with “geographic weighting”. I’d like to hear both sides of that debate. I’m not sure that the opposition will concede Steve’s view or that Steve has established his view on this in detail. And I don’t like it coming in the backdoor, when I ask about “offcenteredness”. That’s not fair. It’s shifty…or at least it’s poor thought disipline.
*which would be the straightforward response…rather than the Mannian attempts to tell me why I shouldn’t ask the question or to answer a different question, without even doing me the courtesy of saying “I don’t know”.
Re #9, TCO
Overweighted is a relative term. Your question “How overweighted?” assumes there is some “correct” value at which they should be weighted. To assign such a “correct” value would require some sort of solid theory of the correlation of tree rings to temperatures, which doesn’t really exist.
About the only thing you can say is that, on the standard theory that tree rings are a linear function of temperature, and given that nearby weather station observations show that 20th century BCPs are totally uncorrelated with temperature, then the “correct” weight for BCPs should be zero.
Au contraire mon frere. Arguments have been advanced that BOTH, the proxies themselves as well as the algorithm are in error. I will fight tooth and nail attempts to play shifty games to say that there are two errors and then when we try to examine one, to deflect conversation to the other. If you really believe what you say, then what’s the point of talking about the off-center PCA. All the discussion would merely be on the proxy validity (and if that argument ended in Mann’s favor, he would win the match).
Or…to keep this from being a semantic argument. How about just saying THE weighting. We can argue about how much is “over”, but at least we know then explicitly how different methodologies affect the “percent BCP character”.
Think about what Burger and Cubasch did. Was their work pointless, even though they made no commentary on the substudy validities in the meta-analysis? No. It was still useful to see that different methodologies give different answers FOR THE SAME DATA.
If it helps, I think of these problems as multiplicative (?) where you seem to be regarding them as additive. To be less opaque :
Problem A – BCPs as dodgy temperature proxies
Problem B – dodgy statistical procedures like the “Mannomatic”
Am I right in saying that you think of the Total Problem as being A+B ? So if you set A to zero, then you still have B, and vice versa.
In fact, I think the Total Problem is A x B. If either A or B is zero, then the Total Problem is zero. You only have a Total Problem when both A and B are non-zero.
Oops. My 13 responds to your 11. Ref 12, well yes, different statistical approaches to the same data are likely to give different results, or there wouldn’t be much point in thinking up new methodologies. The question is whether it gets you any closer to some useful level of understanding about the world.
In this case, you could get to a table of “Methodology A, Resulting Hockey Stickness of X; Methodology B, Resulting Hockey Stickness of Y; …”
If you are a zealot – at either end of the spectrum – you then start looking for justifications for whichever Methodology gives the Hockey Stickness you prefer.
If you are looking for understanding, you have to decide on a basis for selecting the Methodology independently of its result in this case, then accept whatever Hockeystickiness it gives you.
Is this any use ?
Fred, I’m well aware that there may be an interaction effect. However, I think the “interaction” of an overweighted method with an invalid sample is a rather trivial concept. I can think of much more counterintuitve interactions. And the incessant train of people who want to essentially shift the topic to BCP proxies or to say “it doesn’t matter because tree-rings are evil” is almost Mannian.
Steve made comments on the method. That it mined for hockey sticks out of noise, that it overweighted what hockey sticks there were in the samples. That the BCPs carry the load while the rest of the stuff is a dog’s breakfast. I’m trying to get those critiques qantified and seperated. And they are seperable. Even if there are interactions, we can still qauntify how the story is changed if Steve is right/Mann is wrong on criticism x1, crit x2, crit x3, etc. and all the combinations of them. And it’s actually relevant to know this for the general tone of the Mannian debate.
If Mann has minor errors that don’t change the answer, and we go running around thinking/acting/saying that they change the story, then that’s just wrong and inaccurate and crappy. Plus, if we push ourselves to understand extents of effects, I think we end up having a much better feel for the whole complicated mechanism that is the Mannomatic.
Heck, I think it was already relevant to clarify between weighting issues in the PC versus the hockey stick overall (and I’ve actually asked this months ago). It’s relevant that “off-centering” versus “centering” PCA is a smaller issue than weighting versus unweighting by geography (and if we compare an off-centered non-geoweighted to a geo-weighted centered, we lose that understanding. EVEN if there is a non-trivial interaction, we shouldn’t compare (a.b) to (1,2) and then say that we’ve expressed the differnce between a and 1 (since we were changing another variable at the same time.)
I really don’t think Steve or Ross have any issue with that point of view. (And will even say…that they never did…blablabla…fine, that’s not the point.) And the support crew shouldn’t either.
TCO, let’s remember where we’re coming from. MBH is this big famous study that says that it has remarkable statistical skill, is "robust" to everything under the sun, has the world’s best proxies etc. etc. Our take on this is that: the statistical skill does not exist when usual statistics are considered; the "robustness" does not exist; and the key proxies are contested by the specialists.
That’s where I started. I found it amazing that these claims were false. Of course, all kinds of local color was added into these issues in terms of detective work and all the other nonsense. But the main point for me was that the representations in the prospectus were false. Also that there was evidence that the authors knew that the representations were false.
THe allocations that you request are all sensible question – how much of the problem relates to the Mannomatic relative to ordinary PCs; to PCs relative to means; to means relative to medians. To geographic weighting as opposed to data mining. To proxy validity. All good questinos. But these allocations were not material to the truth or falsity of the initial representations.
In the aftermath of our articles, it seems that nobody in climate science seems to care whether the representations were true or false or honest or dishonest; but only whether they can "get" a HS in some other way or not. In this environment, allocation issues become relvant and I’ll deal with them.
2 Trackbacks
[…] own work is subject to “scrutiny”, he has to resort to FTP data folders labeled “censored“. Look at what happens when that “censored” data is […]
[…] this broader statement is that the censorship seems to influence the same period as the aptly named “censored” directory by Mann that was ultimately erased from […]