2005-10-24

von S on the Hockey Stick

There has been a certain amount of argy-bargy between von Storch and MBH which has lead some people, I think, to mistake von S's position. A recent GRL paper goes some way to clarifying things. Read the RC post on this for refs, and the pdf etc.

As far as I can see, Von S and Zorita basically re-do the stuff from their Science paper, but trying the different centering techniques. And they discover... that it doesn't make much difference. I could have told them that (here though obviously I never quite finished it...). They say:

Our results, derived in the artificial world of an
extended historical climate simulation, indicate therefore
that the AHS does not have a significant impact but leads
only to very minor deviations.


All is, of course, not sweetness and light. There is more to come, I think.

18 comments:

Luboš Motl said...

Why don't you show your readers the paper itself? Do you think that everyone is a Steve Bloom who can't understand it, and your readers should just pray to you as a divine force who reveals us the truth? I am kind of baffled by your approach.

Luboš Motl said...

Sorry, I linked to the abstract. The full article is here.

Luboš Motl said...

Let me summarize their paper very differently:

First of all, they agree with MM05 that the PCA-centering was a wrong step of MBH and it should never be done again.

Second of all, von Storch and Zorita announce that they only addressed one of many concerns of MM05, namely the PCA-centering. Other concernes of MM05 "have not yet been dealt with".

Third of all, while their redone analysis has the blade, it does not confirm that "shaft" of the "hockey stick" - simply because the true oscillations were much greater.

Your summary is bogus - and the reason why you did not show your readers the article itself is simply that you wanted them not to recognize that your summary is as twisted as the stalinist presentation of the history of capitalism.

Best
Lubos

Luboš Motl said...

Let me also mention that Steve McIntyre made a detailed answer here. Besides the climateaudit article, a 11-page publishable answer of MM is here.

Luboš Motl said...

Finally, I add the following summary of MM's reply to von Storch and Zorita:

MM never claimed that PCA-centering affected all situations. It affected simulations like MBH98 in which there are a few "bad apples" that are picked by the PCA-centering method, giving rise to a spuriously high RE statistics but low R^2.

Von Storch and Zorita present an example that is different from MBH98 because the correlations are kept in 0.3-0.7 with the gridcell temperature data, which by definition eliminates bad apples with non-climatic spurious signals.

William M. Connolley said...

Well Lubos, since you've said "finally" I can get in the last word...

M&M have a changing story. Originally the claim (as picked up and repeated by Muller) was that you got a hockey stick out of random data. That would now seem to have been abandoned. Now you need special data they say... but vonS&Z show quite clearly that in the context of *their* experiments (which you will recall were sufficiently realistic for them to be trumpted by all the septics when published in Science) the normalisation doesn't matter.

And... it would be nice to see you have your own opinion. There is no need to parrot M&M here... anyone who wants to read their stuff can find it easily enough.

Luboš Motl said...

Dear our beloved and respected climate modeller William,

(I was told that this is how the comments must start otherwise they're deleted instantly),

M&M still claim that you get the
hockeystick out of random data, but by random data, they mean something that is really as random as the sources of
MBH98.

On the other hand, von Storch and Zorita did not use quite random data in this sense because they engineered various correlations to be small between 0.3-0.7 for all of input. It's about the measure that determines "randomness", and M&M have the same measure as the MBH input data but different from von Storch and Zorita.

You are more than welcome to publish another "last word".

Thankfully Yours
Lubos

William M. Connolley said...

Ah... thats more like it.

Quite what sort of data M&M are insiting on this week I don't know. The MBH data clearly isn't random, because it has autocorrelation and it averages to produce the tradiational hockey stick shape, as Huybers clearly shows.

You are evading the point: von S's data was good enough for Science, and it was good enough for you when it looked like it was going against MBH. Now its going the other way, suddenly vonS's data is no good.

Eagerly awaiting your next last words...

Anonymous said...

Lubos, regarding your first comment, let me just say that I try to carefully think my posts over so that unlike you I'm not forced to go back and correct myself more than once in the same thread. But I think I understand -- string theory is hard work, and since participating in these discussions is a sort of primal scream therapy for you, stopping and thinking before you post would be contraindicated.

William M. Connolley said...

Um, Steve, I think that was a bit off... please folks, no snarky comments about other posters (except from me, on occaision... but I run this mad house).

William M. Connolley said...

Hmmm... now I've just noticed that Lubos's very first comment was an attack on you. Sigh. I slap his wrist too.

Luboš Motl said...

Dear William,

von Storch's article is interesting because it studies a particular issue connected with one of the very important recent climate articles, namely M&M.

Von Storch and Zorita demonstrated that a generalization of a statement of M&M - namely the statement that PCA-centering *always* destroys the analysis - would be incorrect.

However, VZ do not show that PCA-centering did not affect the analysis in the particular example of MBH. Of course they could not prove and publish something like that because it is not true, and the paper M&M already did demonstrate that one could generate the hockey sticks from random data of the similar type and distributions as the MBH98 input data.

VZ only consider input data that satisfy certain conditions about the limited/forbidden presence of misleading proxies. These conditions were not satisfied by MBH98.

I know exactly what Steve McIntyre insists on not only this week, but this afternoon. I hope it's OK to republish one of his statements from today 3PM:

"VZ also did not replicate the hockey stick algorithm. They did PC on the
correlation matrix of the de-centered data; Mann did SVD on de-centered data
and took the left matrix. There is a material difference."

"Secondly, our position on getting hockey sticks from random data remains
unchanged. The amount of bend depends on the amount of persistence. VZ
argued that this effect was particular to random data and would not exist
with an actual signal. In such circumstances, the signal competes with the
hockey stick mining operation. If the signal is strong enough, it gets
picked up rather than pure mining. But the mining strongly interferes with
finding an actual signal in high-noise environments. This is not resiling
from our earlier analyses, merely applying the analysis to the different
situation posited by VZ."

All the best
Lubos

EliRabett said...

The Huybers comment concentrates on the North American Tree Ring data set although the point (at least to me) appears to hold for similar PCA in general. His data is the seventy NOAMER Tree Ring records discussed by MM05, not the complete set used by MBH98. The "hockey stick" as such is not discussed, although it would be straightforward to apply his normalization to the entire set of proxys.

vS&Z use a climate model to construct temperature fields across space and time, and then use the "various" PCA normalizations to build a "global" temperature record.

Does anyone read these things??

Anonymous said...

thanks for having a debate with your "opponents" vice censoring the discussion.

William M. Connolley said...

Bear in mind that one of the reasons you're here is that the discussion *is* moderated. Moderation (ie, deleting certain posts) is a virtue, as long as its not overdone. Unsurprisingly, my line-in-the-sand will not be exactly the same as yours, though. Go to sci.env if you want unmoderated discussion (I post there too, but its noisy).

Anonymous said...

I don't know how to post on usenet. And I haven't seen you doing any moderation. Are we allowed to debate? To troll? To drunkpost?

William M. Connolley said...

Trolling is out. Repeating the same question twice gets you dangerously close to trolling.

If you have no proper new service, then groups.google.com will allow you to post. Since you're a newbie, I can advise you on such matters...

Anonymous said...

Thanks, dude. Sorry about the repeat.