## Mann to speak at UC Santa Cruz

As one of our commenters has helpfully pointed out, Michael Mann will be giving a presentation at the University of California, Santa Cruz, this Wednesday.

Michael Mann, director of the Earth System Science Center at Pennsylvania State University, will give a lecture on global climate change on Wednesday, May 10, at UC Santa Cruz. His talk–"Global Climate Change: Past and Future"–will take place at 7 p.m. at the Seymour Center at UCSC’s Long Marine Laboratory. The event is free and open to the public.

He’s going to be speaking on topics very close to our hearts:

Mann is one of the leading authorities on global climate change. His research has been central to establishing the growing human influence on climate and, as a result, has been the target of criticism from skeptics of global warming. Mann will present the evidence for a human influence on the climate of recent decades. Such evidence includes instrumental measurements available for the past two centuries, paleoclimate observations spanning more than a millennium, and comparisons of the predictions from computer models with observed patterns of climate change.

If, by any chance, someone is in the area and can give a report on what transpires, I’m sure Steve will be most interested. Hopefully, Dr Mann will not have to dash off for a flight back to Philly and sticks around for a few questions at the end, you never know…

1. Steve McIntyre
Posted May 8, 2006 at 6:09 PM | Permalink

It’s remarkable how the hockey stick doesn’t “matter” while, at the same time, Mann’s research is “central” to establishing influence.

2. Pat Frank
Posted May 8, 2006 at 7:15 PM | Permalink

Mann is likely to get a very pro-partisan audience at Santa Cruz. It’s 1.5 hrs from here and I know the area fairly well. Both Santa Cruz itself — gov’t and population — and UCSC are hotbeds of environmental activism. Any audience is are likely to be completely convinced of AGW, anxious to do something about it, and ready to denounce oil company shills and corporatist stooges. Anyone critically questioning Mann is likely to end up tarred and feathered. OK, that last might be an overstatement. … Maybe … but then …

3. nanny_govt_sucks
Posted May 9, 2006 at 12:05 AM | Permalink

It looks like he’s on a West Coast tour. Here he was at UC Irvine on May 3rd: http://www.ocregister.com/ocregister/homepage/abox/article_1124603.php

Don’t know if he stopped at Cal Poly SLO on the way North… I wonder if Ticketmaster will have the goods on the rest of his stops on this leg.

4. Posted May 9, 2006 at 4:35 AM | Permalink

Glad to know that Michael Mann was going to speak on these interesting topics. I’m really looking forward to see some reviews about it here.

5. Louis Hissink
Posted May 9, 2006 at 8:46 AM | Permalink

“Such evidence includes instrumental measurements available for the past two centuries,”

???

I feel a very serious JORC coming on.

6. Louis Hissink
Posted May 9, 2006 at 9:04 AM | Permalink

“paleoclimate observations spanning more than a millennium”

Observations?

These people make me look like a schmuck! Scientfically that is.

7. Brooks Hurd
Posted May 9, 2006 at 10:05 AM | Permalink

Re: 5, Louis,

Don’t we all believe Jones’ instrument graphs in spite of the fact that he will not release the data?

8. Brooks Hurd
Posted May 9, 2006 at 10:10 AM | Permalink

Re: 2
Pat,
I agree with you about the likely audience at UCSC. Politically, Santa Cruz makes San Francisco look middle-of-the-road.

Although I am confident that the cheering choir in the audience would make even RC seem reserved. I wish that I were back in SLO this week so that I could drive up to UCSC to hear his talk and report on it.

9. Gerald Machnee
Posted May 9, 2006 at 10:58 AM | Permalink

Will anyone ask how accurate the temperatures according to paleoclimate observations were a thousand years ago?

10. Posted May 10, 2006 at 12:36 AM | Permalink

Has anyone confirmed attendance? I’m about 45 to 60 minutes away, and if I do some “spousal negotiation” I might be able to find my way there.

Also, what would be the one question to ask? Being tarred and feathered is my specialty ;-)

11. nanny_govt_sucks
Posted May 10, 2006 at 1:25 AM | Permalink

Maybe you should ask him for the Caramilk secret. How DID he compute the confidence intervals in MBH98/99? That question would be just technical enough that you might not appear to be a wolf in the fold, and could probably escape the wrath of the true believers gathered there. Can you bring a tape recorder to catch his complete reply?

12. John A
Posted May 10, 2006 at 1:52 AM | Permalink

Justin,

It might be helpful if you could take notes and/or get a copy of Mann’s presentation. It would be fascinating to know whether Dr Mann has altered his position on any point.

As for a question: “Why do practically all reconstructions of past climate over the last 1000 years show much greater variability than yours?” might get you booed. On the other hand mentioning Steve McIntyre might get you lynched. There’s a happy medium somewhere…

13. Jean S
Posted May 10, 2006 at 2:21 AM | Permalink

re #11: We know how he did MBH98, but MBH99 is still somewhat mystery (or actually the exact procedure). But I’m sure if you asked that question, you would get an answer equivalent to “read the f**king papers, stupid, it’s all there”.

I find it curious that Mann has at least three (MBH98/MBH99/Gerber et al,2003) uncertainties for the SAME reconstruction. Seems like his uncertainty varies depending on the mood he wakes up ;)

14. Jean S
Posted May 10, 2006 at 2:35 AM | Permalink

re #10: An “innocent” question to ask, which certainly upsets Mann, but the audience does not have a clue what you are asking, is to ask about his “principal component analysis” method. Ask something like: what are benefits of your unconventional PCA used in MBH98 over the standard PCA? ;)

15. Jean S
Posted May 10, 2006 at 3:21 AM | Permalink

Justin (or anyone else going there), please take some pictures. At least I’m curious to see if he really looks like in the picture on his web page. Also, it would be nice to see the audience…

Oh, and remember to ask if it is necessary to hurt ancient trees when collecting those “paleoclimate observations” ;) On the other hand, after reading #2 and #8, I think someone may already have asked this question before it’s your turn!

16. Posted May 10, 2006 at 10:07 AM | Permalink

I have a 50/50 chance of going right now, so no promises. Still having a hard time convinving my wife that the future of the global environment is at stake and more important than changing diapers for an evening ;-) In all fairness, she read Crichton’s “State of Fear”, and while previously being a bit of an environmentalist, a little digging in to his references left her disillusioned. Now she thinks it is all a big waste of time…

IF I go (a big IF) I will bring a camera. Probably not a tape recorder, though if they allow video cameras, I will have one along. As far as the question, nothing too technical as I don’t want to get shot down with a pseudo-technical reponse that I cannot refute – I have a background in economics and understand regression analysis, r^2, etc, but things start to get hazy at the autocorrelation point.

Here were my potential questions:
1. “Given you efforts to bring to light the dangers of global warming, why did you wait seven years to release your raw data and code so that others might contribute to your findings?

2. “Many of your critics site the fact that you have never released your r^2 numbers. Given your position that these numbers are meaningless in your type of analysis, why not just release the numbers and quiet the critics?

3. More of a rhetorical approach. I don’t know if these are valid, so please let me know:

** If global warming is determined (from a layman’s perspective)by an average, isn’t it possible that calculations are influenced by temp spikes in certain areas and therefore the “average” number isn’t really indicative of true GLOBAL trends?

** The graph of Mann’s data is for the northern hemisphere only. If we chart Mann’s 2003 data for the southern hemisphere (200AD to 1980AD), we actually see a cooling trend (see the front page of my blog). Wouldn’t it be appropriate to always show both of these together?

4. I haven’t been following climateaudit for a month or so (this whole day job thing is a real downer), but is there a clear explanation of R^2 versus RE, and a good argument that R^2 also needs to be provided? My understanding is that R^2 is correlation based upon each discrete data point (year to year comparison, if you have yearly data points), wherease RE looks at correlation over several periods (comparing 50 year cuts). The only thing I can think of is that looking at a 50 year cut admits that your data is weak, given that correlation should increase as you smooth your data over a number of time periods. (I may be WAY off on this).

Others?

17. Doug L
Posted May 10, 2006 at 10:51 AM | Permalink

Trying to trap him sounds like fun, but will probably be disappointing, unless perhaps there are others there who would be encouraged to do the same. I’d rather see him encouraged to agree with some outrageous statement, but have no ideas for that.

Possible fun ideas:

Ask if he’s involved in preventing Steve M from accessing a web site from his Steve’s home computer.

18. Posted May 10, 2006 at 11:07 AM | Permalink

Yeah, as they say in the legal world, never ask a question you don’t know the answer to. Anyway, I assume anything written here is read by those at RealClimate anyway….

Perhaps I’ll just sit quietly and take notes and avoid becoming a lightening rod (worked in graduate school!)

Justin

19. John A
Posted May 10, 2006 at 12:50 PM | Permalink

Copious notes and a copy of his presentation would be nice. There’s no sense antagonizing the crowd, however much they deserve it.

20. Jean S
Posted May 10, 2006 at 12:52 PM | Permalink

[John A/Steve, please correct the mistakes (formulae, bad spelling/English) and format as you wish]

re #16: About 4: Suppose you do a standard linear multivariate regression (normality of errors). Now it is a standard theorem stating that RE (=coefficient of determination, and usually denoted in statistics as $R^2$) increases if an additional regressor is incorporated into to the model irrespective of its values. That is, in layman’s terms, if you have $N$ proxies, and you add an another proxy (eg. noise), you get a better RE value! Therefore, RE is not that good alone. In fact, high RE values and low squared sample correlation coefficient (usually denoted in statistics with $r^2$) sign exactly what you can read from the above theorem: overfitting! This is why you need to have $r^2$ to determine if your RE values are of any good.

And while I’m on to it, I’ll put the formulas here and some explenation for those interested (see here for better explanation). In regression (you can think it as "prediction"), one has $N$ variables $X_1,\ldots,X_N$ called regressors, and (in this case) a variable $Y$ called response variable. Each of these are just vectors of length $T$ (i.e., just a collection of $T$ numbers), so we have $T$ observations. It may be helpful think here each of $X$‘s as a "proxy" (e.g., treering widths) for the time period 1902-1980, and $Y$ as the CRU instrumental record for the same time.

Now suppose we somehow think that $X$‘s and $Y$ are related through an equation

$Y=f(X_1,\ldots,X_N)+\epsilon,$ where $\epsilon$ is a normal (Gaussian) random variable (error), and $f$ is our magic function.

If $f$ is linear, then we talk about linear regression. So in the simplest linear case ($N=1$), the above equation is written as

$Y=aX_1+b+\epsilon,$, where $a$ and $b$ are some constants.

Now in regression we simply find the "best" values for the parameters of $f$, i.e., in the above simple linear case the best values for $a$ and $b$. Usually "the best" means that we minimize the squared error (i.e., this is ordinary least squares) between $\hat{Y}=f(X_1,\ldots,X_N)$ (think $\hat{Y}$ as your temperature reconstruction) and $Y$, i.e. minimize

(e.g., in the above simple case, for $a$ and $b$) the sum of squared residuals ($\epsilon_t=Y_t-\hat{Y}_t$, where $t$ denotes "time"):

$RSS=\sum_{t=1}^T\epsilon_t^2=\sum_{t=1}^T(Y_t-\hat{Y}_t)^2.$

Now suppose we are only given $Y$ and somehow obtained $\hat{Y}$, and we are asked how "good" prediction $\hat{Y}$ is. The simple way is to calculate the reduction of error (RE) $R^2$ (so called mainly in climate circles) given by

$R^2=1-\frac{RSS}{SYY}$, where $SYY$ is obtained from

$SZY=\sum_{t=1}^T(Z_t-\mu_1)(Y_t-\mu_2)$

by setting $Z=Y$ ($\mu_1$ is the average of $Z$ and ($\mu_2$ is the average of $Y$).

This value ($R^2$, RE) is always smaller or equal to one, and one denotes the "perfect fit". However, it always denotes the fit relative to the model assumed (e.g., linear model of degree 5). As explained in the beginning of this post, you get higher RE values simply by adding more regressors regardless of the fact if they have anything to do with $Y$! Therefore, it is always good to calculate some other "measures of fits", especially if the model is not explicitly stated (as in the case of MBH98 although the model is linear there). Another "fit measure" is the squared sample correlation coefficient given by

$r^2=\frac{S\hat{Y}Y^2}{S\hat{Y}\hat{Y}*SYY}.$

These are the values Mann is not reporting, and are low in MBH98.

21. jae
Posted May 10, 2006 at 1:13 PM | Permalink

22. Posted May 10, 2006 at 2:50 PM | Permalink

Ditto jae’s comment.

One clarifying question: for SZY, you say Z is set to equal Y, so essential we have: SZY = SYY = the Sigma(Y_t = Y-avg)^2

Correct?

23. Posted May 10, 2006 at 3:11 PM | Permalink

And Mann hasn’t given a clear answer to the question: “Would you prove you didn’t overfit your model”? Like I said, I haven’t had time to keep up to date on ClimateAudit….

24. John A
Posted May 10, 2006 at 3:12 PM | Permalink

Justin,

Just ask the organizers if you can video the thing and/or take copious notes. I’m sure it’ll be interesting.

25. Steve McIntyre
Posted May 10, 2006 at 4:04 PM | Permalink

If you’re brave, one question to ask: “Recent studies show that the MBH98 reconstruction failed the verification r2 statistic. In your original publication, you said that you calculated the correlation and r2 statistics for the verification period and even illustrated the verification r2 statistic for the 1820 step, but at the NAS panel, you said that you did not calculate the verification r2 statistic. How do you explain these seemingly contradictory positions?”

26. Posted May 10, 2006 at 5:10 PM | Permalink

Steve –

Playing devil’s advocate, he could just say:
1) I didn’t say that
2) I wasn’t referring to MBH98
3) MBH98 is old hat, we admitted our errors, we’ve moved on

27. Howard
Posted May 10, 2006 at 6:25 PM | Permalink

Mr. Mann, now that active global warming has been confirmed on Mars, Neptune, Pluto and most recently Jupiter, do you think that this has confirmed your “Hockey Stick” and will quiet your detractors?

28. Ross McKitrick
Posted May 10, 2006 at 7:01 PM | Permalink

I’d re-phrase Steve’s question as follows. “In your 1998 paper where you introduced the hockey stick graph, you used the RE score to test significance, but you also said you compared it to the r2 score, and you said that “Significance levels for r2 were determined from standard one-sided tables” (p. 786). There’s no mention of it being an inappropriate test. However you have never published the r2 scores. Yes or No, were the r2 values significant when you compared them against standard one-sided tables?”

His answer will be something like, “Well, my critics keep harping on this only because they don’t understand the statistical methodologies used by the recognized experts in the field. We evaluated significance using the RE – Reduction of Error – statistic, which is the measure used by all competent climatologists for these kinds of analyses. The r2 statistic is quite simply inappropriate because it fails to accurately evaluate low-frequency skill in settings where the mean is nonstationary, which is the case in long-term paleoclimatological studies. No matter how many times I and others have explained these things, there is a small but noisy band of mostly industry-funded skeptics, who aren’t even climate scientists, who don’t understand the science and keep trying to drag the debate back a decade to re-hash old issues, like r2 versus RE, which no one in the field is paying attention to anymore, since they’ve long been settled in the peer-reviewed literature.”

To which you can reply: “So in other words the answer’s No, the r2 values weren’t significant.”

29. Ed Snack
Posted May 10, 2006 at 9:58 PM | Permalink

Why not ask why ask how he can justify the use of the Bristlecone/Foxtail pine records when the original researchers who provided the data clearly indicated that the ring widths were not correlated to temperature. And as an extra, ask what happens when those records are omitted from the calculation. To any answer state that this removal was in fact modelled and the rsults are still available, and the outcome is no Hockeystick.

Or you could ask about the r^2 issue, remember (thanks Danny boy) that Mann did publish some r^2 results in MBH98, those for the 1820 step, ask why he denied calculating r^2 when he published same, and what were the r^2 results for the 15th Century step.

On the same lines, Wahl & Amman did calculate r^2 for the 15th century, say: W&A’s emulation of MBH98 produces the same results and has been supported as doing so, and they have published r^2 statistics for the 15th century showing that the result has no merit, does he therefore agree that the same holds for MBH98, or does he repudiate W&A.

Pity it is just a bit too far for me to attend, though I guess I would feel a bit out of place.

30. Jean S
Posted May 11, 2006 at 4:29 AM | Permalink

re #22: I defined SYY through SZY because I needed the formula also later in $r^2$. Yes, SYY is $N$ times the sample variance (sometimes you want to use $(N-1)$ but that is immaterial here).

Notice also that the above formulation for $R^2$ gives the possibility to check the MBH98 "uncertainty limits" as Steve explained earlier. With simple manipulation, you get
$\sqrt{\frac{1}{N}RSS}=\sqrt{(1-R^2)\frac{1}{N}SYY}$
Now $\frac{1}{N}SYY$ is simply the sample variance, which can be calculated from the available data, and $R^2$s were reported, so we can calculate the left hand side which is usually known in regression as the standard error. These values match almost perfectly to uncertainty sigmas in MBH98. Since "uncertainty limits" in MBH98 are given as $2\sigma$, we can say that those are simply twice the standard error.

The weird thing is that the standard error is calculated in the caliberation period. This means that making a temperature reconstruction that matches almost perfectly to the CRU record in the caliberation period (this can be done pretty well since there are relatively many proxies with respect to the length of the caliberation period), would give a reconstruction that in Mann’s opinion has almost zero "uncertainty" also in the past! Or to put in even harder way, you can generate 79 (the length of the caliberation period) or more white noise "proxies" and find a projection that matches perfectly in the caliberation period to the CRU record, so then you have a "temperature reconstruction" that has zero "uncertainty" in the past! This reconstruction would also have perfect values in the caliberation period ($R^2=r^2=1$) (of course it would be "stupid and foolish" to report the values in the verification period). The best part of this reconstruction is, however, that you can extend it back in time (or to the future!) as far as you wish!

31. Jean S
Posted May 11, 2006 at 5:03 AM | Permalink

[I think I forgot some end-tags above in #30, hopefully it will be fixed. Wouldn’t it be nice if you could somehow edit your own posts?!]

Re #29: Ed, I’ve missed this one. Where did Mann published r^2 for the temperature (in the 1820-step)? (Of course we can actually calculate that from the available data).

32. Posted May 11, 2006 at 4:02 PM | Permalink

Just to let everyone know, I made it to Santa Cruz last night for the Mann presentation, but haven’t had a chance to write up all of my notes yet (should finish this tonight). And yes, I did ask him about r2 – I’ll leave you in suspense for now :-)

(but don’t count on anything like: “he admitted that his models have no statistical value and that he is publishing an article shortly recanting all of his earlier claims regarding global warming / climate change)

33. Steve McIntyre
Posted May 11, 2006 at 4:34 PM | Permalink

#31. Jean S. Look at Figure 3 in MBH98. He shows verification r2 for the AD1820 step by individual gridcell. Look at our NAS Panel submission for a summary on the verification r2 issue.

34. miniTAX
Posted May 13, 2006 at 10:46 AM | Permalink

Hi all,
Didn’t you know that after Gore’s film, another prominent actor in the GW filmography entered the scene : Michael Mann now has his own plot of a warm place : want a review ? Look here
http://filmforce.ign.com/articles/589/589674p1.html