2005-09-16

R2 and RE? Err, no.

JF asks for explanation of the R2 vs. RE statistical issue - what's necessary, what's sufficient, and why. This of course is Bartonism - question 7c, to be precise. The honest answer is, I don't know. And of course, neither does Barton. Take him away from his staffers and his notes and he wouldn't have a clue. So why is a US senator allowing himself to be used as a sockpuppet of a canadian mining consultant to parrot questions he doesn't understand? Politics, of course.

Mann responds that R2 isn't so useful because it measures the simple correlation, and doesn't assess the mean or standard deviation (i.e., if series A is 10 times series B, it would have a perfect r2 of 1, but wouldn't be a very good reconstruction). Mann prefers RE, whatever that may be... which sounds entirely plausible.

But basically folks, this is all a waste of time. Its like arguing with creationists about uranium-series dating, there will always be some tedious technical blind alley they will invite you to go down, while the rest of the world does more interestng things. Even the climateaudit groupies have realised this, with comments like "But isn’t it time to use your brain for something new?"

Oh dear... that tempted me into reading some of the CA comments. Try:

"I’m not a politician, but if I was, and saw replies that either ducked direct questions or gave woolly answers to them, then I’d be tempted to ask them for a little face time to discuss this."

This from someone who thought Mann was evasive. Funnily enough, Barton doesn't seem in any great hurry to hold any hearings, having got his fingers badly stomped on by Boehlert (more).

10 comments:

Dano said...

You are judged by the company you keep.

I'm not sure why you'd want to be surrounded by that crowd unless you like yes boys. But you're not going to be legitimate having a site like that.

D

Anonymous said...

For future reference, Barton is a member of the House of Representatives (referred to as a "Congressman), the "lower house" (House of Representatives) composed of 435 voting members. The Senate ("Senators") is the "upper house" with only 100 members. Unlike in Britain, where wholly-owned creatures of the fossil fuel industry seem to be sequestered in the Lords where they can't do too much harm (aside from issuing bogus reports), in the U.S. they can be found in all too influential positions in both houses.

This term "Congressman" is used a bit inconsistently since both bodies together are "Congress" but Senators are not referred to as Congressmen. The strictly proper title of Congressman is U.S. Representative, but that's not commonly used. A little more common is Member of Congress/MC similar to the British Commons practice (but again Senators aren't referred to by this term, although they sometimes use "USS" in place of MC, but much more commonly just writing it out all the way -- I say to avoid confusion with ships).

Members of both houses are referred to as the "Honorable," although this term seems to be resorted to more frequently by Congressmen, who probably have a greater need for it due to their much higher rate of scandal, indictment and conviction (being closest to the people and all). It's an interesting country where the quickest route to honor is to buy an election.

This seems a good place to mention my all-time favorite campaign slogan, a little OT for this post but still broadly topical since it involved a race for governor of Louisiana between a notably-corrupt former governor and a Grand Wizard of the Ku Klux Klan: "Vote for the lizard, not the wizard!" I don't think the shaky state of the levees was an issue in the election. To the relief of all, the lizard won but true to the slogan pretty promptly went to prison for bribe-taking. BTW, those who were wondering why those casinos were in Mississippi rather than the more apt New Orleans will be pleased to know that it's because Louisiana politicians such as the aforementioned Honorable Lizard (governors and former governors get that title too) held out for too much bribe money.

William M. Connolley said...

Oops, I'll accept that correction re his title/status. Apologies.

Anonymous said...

Re: "sounds entirely plausible". Yes nullus in verba is really awfully old hat, isn't it?

William M. Connolley said...

Jo - if you have anything of substance to say, you should probably read the Mann letter I've linked and comment on that.

Anonymous said...

Hi,

On your last comment, if the best evidence that can be mustered is an
assertion by a not-really-disinterested party, one should probably
remember Wittgenstein's advice.

To write your answer to John F for you (briefly, I'm afraid): for a
correlation to have "skill", a good RE and a good R2 are individually
necessary and jointly sufficient. A description of R2 as "not
adequate" is rather too economical. Approximately zero for R2 is, umm,
not very good.

There's a story on from that about why the R2 measure fails in this
case, namely that assumption (1) from MBH98 (p780, column 2) of the
linearity of dendro indicators is false. See the comments
here (and in
particular those by David Stockwell, esp #29) for the full relevant
MBH98 quote and insightful remarks.

Cheers, -- Jo Calder

Anonymous said...

A quick PS. You get an interesting reaction to MBH98 assumption (1) from people who grow things for a living.

Cheers, -- Jo Calder

William M. Connolley said...

I think comment 7 makes more sense. Nigel Persaud has lost the wood for the trees. BTW, if you're actually interested in this, you're better off posting to sci.env: I really doubt that many people read the comments in old posts.

Anonymous said...

This is an odd blog. Comments are invited and then uninvited, questions are invited and then not really answered, and the blogger says that you should really be somewhere else anyway.

In such a parti pris debate, self-parody is always just one step away (on both sides, I hasten to add). Ah, wad some pow'r ...

Later, -- Jo Calder

Anonymous said...

I thought Mann was evasive in his answers. (and I have my union card, not that that should matter). It reminds me of my CO asking someone if they had performed a specific safegaurd in a procedure (accident critique) and them answering that it didn't matter. May or may not be true. But first answer the question.

P.s. Steve may or may not be wrong. But his engagement on the math is serious. He's not some creationist. And obviously the math does matter. this is statistics after all.