How Not to Analyze Data, part 3

April 1, 2008 · 139 Comments

Anthony Watts and Basil Copeland have collaborated on another post claiming to establish a connection between solar activity (as proxied by the sunspot cycle) and global temperature (as indicated by the HadCRUT3v data set). Let’s take a close look.

First, they smooth the temperature time series using a Hodrick-Prescott filter. It has a tuneable parameter which enables one to choose the amount of smoothing which the filter imposes; the parameter is somewhat analagous to the “cutoff frequency” in a low-pass filter, or the “characteristic time scale” in other smoothing methods. Watts/Copeland set the tuneable parameter to 7, which effectively smooths the data on a roughly 6-year time scale.

Hodrick-Prescott filtering is an outstanding smoothing method. It’s especially good near the edges of the time span, where many other methods show undesirable “edge effects.” Here, for example, is a comparison of using the H-P filter (labelled HP-7), to using a “running polynomial” smoothing (with a well-chosen weighting function) also on a roughly 6-year time scale (labelled GSR):

Clearly the two methods give nearly identical results except at the very edges of the time span, where the H-P filter performs better. Of course H-P filtering has its drawbacks too, including the facts that it requires an even time spacing and it’s not as useful as other methods for interpolating between the observation times, but for this analysis those aren’t anything to worry about.

They then take the first differences of the smoothed time series; that’s just the difference between a given value, and the preceding value. It’s an estimate of rate of change (i.e., the first time derivative) of the smoothed underlying signal. That’s fine too, except that one has to bear in mind that the values estimated by any smoothing method have an inherent uncertainty. Hence the difference between two consecutive values also has an inherent uncertainty. Here’s their plot of these estimated rates of change (click the graph for a larger, clearer view):

And here’s what I got using the exact same method:

So far, we’re on the same page. But now things start to go wrong. They take the peaks of these rate estimates, and assert that they tend to happen at the same time as the peaks of the sunspot cycle. But they don’t take all the peaks from the estimated rate-of-change plot; they include only the ones that fall near a peak in the sunspot cycle, which are labelled in their plot with numbers indicating which sunspot cycle they correspond to. It’s certainly suspect to include only the peaks that match one’s theory! Even so, there are only 3 peaks which they exclude, so we’ll note that there’s a problem with their methodology, but that it might not be too severe, and move on.

They compare the times of peak of the sunspot cycle to the times of peak warming rate in this graph (click the graph for a larger, clearer view):

They further note that the correlation between the times of sunspot peak and the times of warming-rate peak is a whopping 0.99846454. Now that is mighty impressive; it certainly deserves some exclamation points!!

But there are two problems with this graph. One is that one of the data points is wrong. Look at the points for cycle 11: they plot both sunspot cycle peak and warming-rate peak right on top of each other, happening in the same year, 1870. That’s the right year for the sunspot cycle peak, but not for the warming-rate peak; that doesn’t happen until 1877. Here’s a corrected graph with the erroneous (now fixed) point circled:

It’s curious that the single data point for which the difference between the time of the sunspot cycle peak and of the warming-rate peak is greatest, is the data point which is in error on their graph.

But there’s a far more serious problem. They’re correlating two data series which are both sorted into ascending order. That means that without even knowing what the numbers are, we can be sure that they’ll correlate strongly; as one series gets bigger so will the other, in quite strict relation because they’re two series in the same order. Let me illustrate: I generated 14 random numbers (one for each solar cycle peak) between 0 and 1 (following a uniform distribution), then sorted them into ascending order, then plotted solar-cycle peak year against the random numbers and computed the squared correlation. Here’s the result:

That’s a mighty impressive squared correlation; it deserves some exclamation points too!! Of course, it’s not really impressive when one realizes that two sorted series will always correlate, and that in this case the “impressive” correlation is between sunspot cycle peak years and random numbers. Consider also that this is the squared correlation; the simple correlation coefficient is 0.9941.

Can we apply a valid test to whether or not the sunspot cycle is synchronized with the peaks in the warming-rate calculation? Let’s take the times of peak of the sunspot cycle and compute the differences; these are estimates of the lengths of the sunspot cycle. We can also take the differences between the times of the warming-rate peaks as selected by Watts and Copeland, as estimates of the lengths of the warming-rate “cycle.” Then let’s compare the sunspot cycle lengths to the warming rate “cycle” lengths. If there’s really a correlation, then longer sunspot cycles should correspond to longer warming-rate “cycles.” Here’s the result:

The correlation between them is not statistically significant, but there are very few data points. More noteworthy is the fact that if there is a correlation, these data indicate that it’s going the wrong way; there’s evidence (but hardly conclusive) that longer sunspot cycles correspond to shorter warming-rate “cycles.”

How do we evaluate the analysis of Watts and Copeland? First, they omit three of the peaks in the warming-rate calculation, because they don’t correspond to their theory. Then they plot the wrong value for the cycle which has the biggest difference between sunspot cycle peak year and warming-rate peak year. Then they compute an impressive-looking correlation for two sorted data series, apparently not even realizing that it’s not much different from what you can get correlating sunspot cycle peak times with random numbers. Finally, a comparison of sunspot cycle lengths with warming-rate “cycle” lengths indicates no significant correlation, although the possible relation is in the wrong direction. All in all, the analysis of Watts and Copeland provides no evidence whatever of any relationship between HadCRUT3v global temperature data, and sunspot data.

In responding to a reader comment, Watts states, “I had a number of qualified people including a published climatologist, solar scientist, a statistician, and a certified consulting meteorologist (among others) look at this beforehand.” Which makes you wonder…

UPDATE UPDATE UPDATE

Readers have wondered, what’s a good test to determine whether or not there’s significant coincidence between the times of sunspot cycle maxima and the times of warming-rate maxima? There are many tests; one I’d like to mention is to compute Rayleigh’s R (see Upton & Fingleton 1989, Spatial Data Analysis by Example, vol. 2, Wiley, Chichester, UK).

We wish to know whether warming-rate maxima tend to occur at some particular time during the sunspot cycle. Each warming-rate maximum will be at some particular phase of the sunspot cycle. If it happens right at sunspot cycle maximum, it’s at phase 0. If it’s halfway from sunspot cycle maximum to the next sunspot cycle maximum, it’s at phase 0.5 for that cycle. If it’s a quarter of the way from one maximum to the next one, it’s at phase 0.25. Etc., etc., you get the idea.

To compute Rayleigh’s R, we first compute the phase $\phi$ for each warming-rate maximum; all phases will fall between 0 and 1. Then we compute $(x,y)$ coordinates for each warming-rate maximum by

$x = \cos(2 \pi \phi),~~~ y = \sin(2 \pi \phi)$ .

We can then plot the $(x,y)$ values for each warming-rate maximum on the $x-y$ plane; all these points will fall on the circle of radius 1 (the unit circle).

If all of the warming-rate maxima coincide exactly with the sunspot cycle maxima, then all of the phases will be 0. In that case, all of the warming-rate maxima will be at coordinates $(1,0)$ ; that’s the point on the unit circle toward the right. If all the warming-rate maxima occur at the midpoint of the sunspot cycle, they’ll all have phase 0.5 and they’ll all have coordinates $(-1,0)$ ; in this case they’re all $180^\circ$ out of phase with the sunspot cycle. The plot of $(x,y)$ positions is like plotting points on a “clock” which marks out the sunspot cycle; toward the right (the “3 o’clock” position) is in phase with the sunspot cycle, toward the left (the “9 o’clock” position) is $180^\circ$ out of phase with the sunspot cycle, etc. etc. you get the idea. And as the sunspot cycle goes forward, the clock goes counter-clockwise.

If warming-rate maxima tend to occur at about the same time in the sunspot cycle, then the points when plotted will aggregate near a particular clock position. If they’re happening at the same time as sunspot cycle maximum, they’ll tend to aggregate near the 3 o’clock position; if they’re happening near the midpoint between sunspot cycle maxima they’ll tend to aggregate near the 9 o’clock position, etc. etc. you get the idea.

To test whether or not they’re aggregating, we compute the average $x$ value and the average $y$ value; call these $\bar x$ and $\bar y$ . We then compute Rayleigh’s R as:

$R = \sqrt{\bar x^2 + \bar y^2}$ .

Suppose all the warming-rate maxima exactly coincide with sunspot cycle maxima. Then all the phases are 0, all the coordinates are $(1,0)$ , and the average coordinates are $(1,0)$ . In this case, Rayleigh’s R is $R=1$ . Suppose they all fall at some other (but the same for all) phase in the sunspot cycle; again Rayleigh’s R will be $R=1$ . But if they’re distributed randomly throughout the sunspot cycle, then the average coordinates will be approximately $(0,0)$ , and Rayleigh’s R will be small.

This is the basis for testing whether warming-rate maxima tend to fall at a particular time during the sunspot cycle using Rayleigh’s R. We compute R, compare it to a critical value, and if it’s greater than that critical value we have evidence that the distribution is not random; if it’s less than the critical value, then the available data are consistent with the hypothesis that they’re occuring at random times throughout the sunspot cycle.

Let’s do it! We’ll use the data according to the method of Watts and Coleman. There are 14 data points (although Watts and Coleman omit 3 of the warming-rate maxima, so we’ll do the same, which is cheating in their favor). Also, they only time the events to the nearest year, so some of the phases turn out to be exactly the same; for instance, two of the warming-rate maxima occur during the same year as sunspot-cycle maxima, so two of the points have phase zero and fall exactly at the 3 o’clock position.

The critical value for Rayleigh’s R, given a false-alarm probability $p$ , applied to $N$ data points, is approximately

$R_{crit} \approx \sqrt{ - \ln(p) / N}$ .

This is only approximate and the sample is small, but as we’ll soon see the result isn’t anywhere near close, so we can ignore the exact calculation (which is generally done by Monte Carlo simulation). We’ll use a false-alarm probability of $p = 0.05$ , which is equivalent to 95% confidence, and $N=14$ data points, so the critical value is approximately $R_{crit} \approx 0.46$ . Here’s a plot of the results:

Three of the black dots represent two data points; they’re indicated by a “(x2)”. There’s also a very short red line, from the origin to the average coordinates $(\bar x, \bar y)$ . And there’s a blue circle indicating the radius corresponding to the approximate 95%-confidence critical value. If the end of the red line extended outside that circle, we’d have evidence that the warming-rate maxima are not randomly distributed, rather that they tend to aggregate at a particular phase of the sunspot cycle. If it were a close call, we’d need to compute the precise critical value for Rayleigh’s R for a sample size of 14 data points.

But it’s clearly not a close call. Rayleigh’s R for this data is $R=0.097$ , the critical value is $\approx 0.46$ . Like I said, not even close.

So in fact there’s no evidence that warming-rate maxima, computed according to the method of Watts and Coleman, tend to fall at any particular phase of the sunspot cycle. Not only do they fail to occur at the same time as sunspot cycle maximum (at phase zero), they don’t tend toward any particular phase of the sunspot cycle.

Categories: Global Warming · climate change

139 responses so far ↓

EliRabett // April 1, 2008 at 3:29 am

It’s April Fools Day, right? The thing that drives me nuts is that when I look at HadCRUT what I see is a bunch of jiggle jiggle, maybe even 11 year jiggle, on a large rise whose period is ~ 100 years. This “method” differences the rise away. The clowns have thrown the forest away to look at the trees.

[Response: It's not quite April Fool's day yet. But I guess it's close enough.]
kim // April 1, 2008 at 4:03 am

Longer sunspot cycles correlate with shorter warming cycles. Hmmmm. Very interesting. Maybe longer cycles are weaker? Let’s ask Leif.

Oh yes. April Fools. Laughs hysterically.
=================================
Ian // April 1, 2008 at 4:47 am

Nicely done - hard to find a more cogent demonstation of a flawed analsis than “significant” correlation with random series.
nanny_govt_sucks // April 1, 2008 at 5:00 am

Can we apply a valid test to whether or not the sunspot cycle is synchronized with the peaks in the warming-rate calculation? Let’s take the times of peak of the sunspot cycle and compute the differences; these are estimates of the lengths of the sunspot cycle.

Can you run that again with the 22 year sunspot cycle?
Georg Hoffmann // April 1, 2008 at 6:32 am

Tamino
I dont miss any of your posts. Not that I am particularly interested in Watts “climate analysis” or the question if they are wrong or very wrong, but it’s just a pleasure to read your deconstructions. The best thing that could happen to someone like Watts is that someone like you is taking their studies apart. At least they were worth for something.
Still laughing
Georg
fred // April 1, 2008 at 8:19 am

Seriously, what is generally thought to have caused the rise the smoothed graph shows from 1900 to 1950, and then the subsequent decline or stabilization until the latest rises?
Onanym // April 1, 2008 at 9:24 am

I really can’t see the point of their analysis, except trying to fool somebody. And I can’t see many comments that make critical comments to their work (goes for CA as well).

However, what’s the R² for the straight forward correlation of the timeseries for the differentiated temp. curve and the sunspot curve (not just the maxima)?

Else, your site is really nice and educational!

[Response: For sunspot counts vs smoothed temperature 1st differences, the squared correlation R^2 = 0.025.]
mmghosh // April 1, 2008 at 12:14 pm

Perhaps Mr Watts could get Mr McIntyre to validate his analyses before publishing them.

Mr McIntyre has a good reputation as mathematical analyst, and his imprimatur would carry weight, on this blog as well as others. To do this should be a reasonably straightforward process as well.
blcjr // April 1, 2008 at 12:20 pm

It is true that 2 out of 16 peaks in the data, during solar cycle 16, are anomalous, and invite further scrutiny. But 14 out of 16 peaks still show Schwabe cycle periodicity. And then there’s the matter of how with higher order HP filtering, we no longer see the even numbered solar cycles, but we still see Hale cycle periodicity in the odd numbered ones.

I’ll look into the data plotted in Figure 5, and if there’s an error, we’ll correct it. Your point about not reading too much, if anything, into a correlation from data that is already ordered in some fashion is well taken.

Would you care to share the weighting you use in your GSR plot? I would like to investigate that further.

[Response: Would you be one of Watts' collaborators? One of his reviewers?

You say you'll "look into" figure 5, and correct it IF there's an error? Are you joking? It took me 15 seconds to realize that graph was garbage analysis. According to reader comments, I'm not the only one. If you do correct it, will you pull a switch and try to pretend the ludicrous mistake never happened, or will you admit on the blog that it was just plain wrong?

As for 14 peaks still showing the Schwabe cycle periodicity -- NO THEY DON'T. Don't you get it?

As for my weighting scheme, I'm preparing to share that -- in the peer-reviewed literature.]
Barton Paul Levenson // April 1, 2008 at 12:26 pm

I can’t believe they sorted the series before correlating them. Doesn’t that mean they’ve decoupled the individual points?
kim // April 1, 2008 at 12:49 pm

blcjr is Basil.
=========
Boris // April 1, 2008 at 12:52 pm

Perhaps Mr Watts could get Mr McIntyre to validate his analyses before publishing them.

It makes you wonder why Steve M would want to be associated with Watts. Watts is not just a commenter at Climate Audit, he’s a contributer and authors of dozens of threads.
blcjr // April 1, 2008 at 1:09 pm

I’m “Basil.” I didn’t realize that I was replying under another account (”blcjr”). I still continue to be amazed at how rapidly discourse with you descends into the regions of negative marginal utility. You must be a very angry person, to let things become so personal to you. I would ask you to chill out, but that’s probably not the turn of phrase you want to see or hear in any discussion of climate change.

Basil

[Response: Every time you guys are proved wrong by rigorous application of legitimate analysis, you complain about hostility. I suppose that delusion is the only way you can avoid admitting the truth to yourselves.

You have shown yourself -- repeatedly -- to be the author of not just mistaken, but grossly incompetent data analysis. That's not anger, it's truth. If you really believe in virtue, then you should admit this on Watts' blog and apologize to his readers.]
kim // April 1, 2008 at 1:33 pm

Shouldn’t the point of figure 5 be how close the x’s and crosses are to each other? Does your analysis show that? Why agonize over the slope at all?
====================================

[Response: It's the "correlation" between the x's and crosses that is reported by Watts and Copeland, and which I reported with the crosses replaced by random numbers. The point of *their* figure 5 plotted in that way (their way), is to give a visual impression of stunning agreement.]
J // April 1, 2008 at 1:40 pm

BPL writes: “I can’t believe they sorted the series before correlating them. Doesn’t that mean they’ve decoupled the individual points?”

What’s to decouple in this case? It’s just two lists of dates.

:-)
Joel Shore // April 1, 2008 at 1:47 pm

Excellent post. It is interesting to see how easy it is for people to fool themselves when they know what answer they want to get.

To be fair though, other people (such a Camp and Tung http://www.amath.washington.edu/research/articles/Tung/journals/solar-jgr.pdf ) have found a correlation between the solar cycle and global temperature…so such a correlation may really be out there. However, I agree that this hopelessly flawed method of analysis doesn’t show it.

It is also worth noting that Camp and Tung use the correlation to derive an estimate of climate sensitivity and get good agreement with the estimated range of the IPCC. (This is assuming, presumably, that the same sort of feedbacks apply for warming by changes in total solar irradiance as by changes in greenhouse gases. Of course, the “skeptics” presumably want to believe that the warming is due to some exotic mechanism like cosmic rays so that they can have a high response to the solar changes and still have a low climate sensitivity.)
climatewonk // April 1, 2008 at 1:50 pm

Need I post this?

“Lies, damned lies and statistics.”

April Fools indeed.
P. Lewis // April 1, 2008 at 2:21 pm

As well as their (obvious?) detrending of any global warming signal over the 20th century in their lower panel of their Fig 3 to leave the “cyclical” HadCRUT3, wouldn’t one also want to remove any El Nino (and La Nina) signals before proceeding to do their first-difference plot in their Fig 5 and before assigning peaks to sunspot cycle maxima? The 1877-88 (11), 1986-87 (22) and 1997-98 (23) regions would surely look different at the very least.
Ian // April 1, 2008 at 2:24 pm

kim said:

Shouldn’t the point of figure 5 be how close the x’s and crosses are to each other?

kim, think about the data and look again at how it’s constrained in the graph – how in the world would you get the xs and +s within pairs to be far apart?

Basil, I hope you stay with the thread; we’d like to hear your responses to the criticisms.
kim // April 1, 2008 at 2:37 pm

Ian, I agree the graph is poor, but the correlation they are seeking is how close the x’s and crosses are to each other. Sure, the format makes it look like they are close to each other, but surely there is a method to measure that and analyze its significance.
==================================
Mike B // April 1, 2008 at 2:56 pm

kim said:

Ian, I agree the graph is poor, but the correlation they are seeking is how close the x’s and crosses are to each other. Sure, the format makes it look like they are close to each other, but surely there is a method to measure that and analyze its significance.

Did you even read Tamino’s post? There is no significance in an doing correlations between sorted series, even random numbers produce a near perfect result. Tamino did a relevant analysis comparing the correlations between the lengths of the cycles and it wasn’t statistically significant.
P. Lewis // April 1, 2008 at 3:09 pm

Re my point above. I’ve now got around to reading some of the comments on that thread (Hmm!) over at Watts’house and I see my very point has been sort of raised there (by Pierre Gosselin (aka AGWscoffer)) and is awaiting an answer … although Pierre Gosselin later also says:

I mean just recently a report came out saying the ocean temps have cooled a little over the last 5 years. Could this have led to the current El Nina?

in correcting

(Wikipedia says that the current cooling is caused by the current La Nina. How do we know its not the other way around?)

(Hmm!)
Horatio Algeranon // April 1, 2008 at 3:14 pm

I have a Theory and it is mine
Hank Roberts // April 1, 2008 at 3:37 pm

No, that’s not the correlation they’re seeking. That’s not even the hypothesis they’re testing. That’s a picture displaying some of their work graphically.

Mr. Kim, is it fair to assume you have not had a college statistics class? How about a calculus class? Give us some idea what level of math you’ve reached in school so we can start from there.
Chris Colose // April 1, 2008 at 4:43 pm

Re: fred

[QUOTE]Seriously, what is generally thought to have caused the rise the smoothed graph shows from 1900 to 1950, and then the subsequent decline or stabilization until the latest rises?[/QUOTE]

The earlier warming was mostly solar, some lack of volcanic, a tiny bit of anthropogenic (possibly some black carbon), some internal variations (some El Ninos at the end of the period). Mostly larger fluctuations rather than a trend like today, and more warming is recent decades.

There has been an insignificant secular trend in solar activity from ~1952, no matter how hard the Watts fraud campaign tries to say otherwise. While these deconstructions are nice to see (good work ), it is disappointing that such nonsense is allowed to keep distracting people, which forces people who actually know what they are talking about (e.g., RC, Tamino) to waste time and stray away from the interesting stuff. I refuse to audit poor statistical analysis on my blog because it will never end. Watts is winning the battle, but not because of science, but because he can exploit the ignorance of people who matter– the public. I think it is time to admit this rather than changing the subject to hockey sticks, sunspots, or general counter-arguments I can find on the global warming swindle
luminous beauty // April 1, 2008 at 4:59 pm

darling kim,

“…but surely there is a method to measure that and analyze its significance.”

Surely there is. But surely not via Basil’s bogosity flux diagrams. Try here:

http://publishing.royalsociety.org/media/proceedings_a/rspa20071880.pdf

I have confidence that you may someday cease being a secondary bogon emitter and eventually become a net cluon absorber.

But I’m not going to hold my breath.
Petro // April 1, 2008 at 5:41 pm

kim wondered:

“Surely there is a method to measure that and analyze its significance.”

Indeed there is, did you this one: http://tamino.files.wordpress.com/2008/04/periods.jpg
JCH // April 1, 2008 at 7:14 pm

El Nina?

I don’t even want to speculate.
Dano // April 1, 2008 at 8:04 pm

Did you even read Tamino’s post? There is no significance in an doing correlations between sorted series, even random numbers produce a near perfect result. Tamino did a relevant analysis comparing the correlations between the lengths of the cycles and it wasn’t statistically significant.

Mike, the confirmation bias programmed into the kim bot does not allow your statement to compute. Same thing as Schrödinger’s cat, only different.

Best,

D
Joel Shore // April 1, 2008 at 9:05 pm

kim: I’ll give you a slightly different answer than most of the people here. I think the fact that both the sunspot cycle and the smoothed temperature data oscillate on roughly an 11-year time scale does suggest some sort of correlation in the data. What Tamino’s analysis shows is, however, that the fantastic correlation that Watts and Copeland pulled out was largely an artifact of their method…and he has also shown that any correlation is weak enough that you can’t see it if you look to try to see if variations in the time between the peaks in the first-differenced-smoothed-temperatures are correlated with variations in the solar cycle length.

I would also caution that, to the extent that the smoothed temperature data do have in them an oscillation on roughly an 11-year timescale, one might wonder if it is partly or even largely an artifact of the smoothing process, which Tamino noted is basically smoothing over roughly a 6-year period and thus an 11-year oscillation may be about the shortest that can show up. So, I would want to see the smoothing done for different values of that tunable parameter.

At any rate, there are other papers in the literature, like the one that I referenced above, that do claim to detect the solar cycle in the temperature data…And, in fact, this is not unexpected on the basis of the change in total solar irradiance and the estimated climate sensitivity (although there is some debate about how much this should be suppressed because of the relatively fast nature of these oscillations compared to some slower timescales in the climate system). In fact, that paper by Camp and Tung actually uses the oscillation that they find to show that the IPCC estimate of climate sensitivity seems to be right on.

I’m not sure how the field as a whole feels about the Camp and Tung work. Detecting these sorts of correlations in noisy data is not trivial to do correctly (and is very easy to do wildly incorrectly, as Watts and Copeland have demonstrated).

However, it is important to keep in mind that the AGW theory is not dependent on there being no response of the climate to the sun. Rather, the problem with the idea that the sun is responsible for the warming in the last 30 years or so is that there doesn’t seem to be the necessary secular trend in the solar data to produce the warming. And, of course, in order to make the solar hypothesis negate AGW, one would also have to explain why solar effects (that, as I noted, don’t seem to have the right trend anyway) get amplified and CO2 effects that should be causing warming get de-amplified.
Hansen's Bulldog // April 1, 2008 at 9:09 pm

Readers have wondered what test might give a *correct* estimate of whether or not there’s correspondence between the times of sunspot cycle maxima, and the times of “warming-rate” maxima. So I’ve updated the post, to illustrate the application of a good method.

In case you’re wondering — the hypothesis of correlation fails. Miserably.
Joel Shore // April 1, 2008 at 9:25 pm

Hmmm…My last post was made before Tamino’s update that throws cold water on the idea that there is any phase relationship between the solar cycle oscillations and the smoothed-difference-temperature ones. That seems to make it look even more damning for the hypothesis tying the temperature to the solar cycle.

As a question to Tamino: What do you think of the work of Camp and Tung ( http://www.amath.washington.edu/research/articles/Tung/journals/solar-jgr.pdf ), or other papers in the literature that do claim a relationship between the solar cycle and global temperatures?

[Response: There are lots of papers claiming a connection between global temperature and the solar cycle, but most of them fail statistical significance. Camp & Tung claim both significance, and a larger amplitude of solar-cycle induced temperature variation than most (if I recall correctly, 0.18 deg.C). I suspect there might be a problem with their analysis, which artificially inflates the estimated amplitude of the response. But I'm not sure.

I'd be very surprised if there's *not* some connection; more solar energy in should lead to higher temperature. But the effect is slight, and doesn't seem to rise above the noise level sufficiently to be detected -- unless Camp & Tung are correct, in which case they've managed to identify it. As I say, I'm skeptical but uncertain.]
David B. Benson // April 1, 2008 at 9:39 pm

Tamino aka “Hansen’s Bulldog” — I especially appreciated the update. Exquisitely clear.

Random indeed…
kim // April 1, 2008 at 10:52 pm

Thanks, HB and JS; that is helpful. No thanks, D and HR, but bowling about par. Smooches, lum; that was funny.

Still, solar minima and climate minima coincide. How come?
==============================
Andrew W // April 1, 2008 at 11:22 pm

I tried to make the following comment over on Watt’s blog but he’s not posting it:

I see that Tamino has done an analysis of your post here:
http://tamino.wordpress.com/2008/04/01/how-not-to-analyze-data-part-3/#more-673

Care to comment on the merit or otherwise of his post?”

I guess your work has allowed him to understand just how embarassingly bad his analysis is.
Hank Roberts // April 1, 2008 at 11:26 pm

> Still …

cite, please? Why do you believe this is true? “How come?” following a statement of belief asks the readers here to find evidence to support what you believe.

Can’t you?
Hank Roberts // April 1, 2008 at 11:27 pm

One thing you might want to do is find the terms used, because yours don’t find anything in Google Scholar:

search - “solar minimum” “climate minimum” - did not match any articles.

search - “solar minima” “climate minima” - did not match any articles.
Lee // April 1, 2008 at 11:40 pm

Andrew,

Watts seems to be blocking quite a number of people. He blocks me too. Even better, he has gone back several weeks and systematically removed my old posts questioning his analyses - posts he previously allowed and even commented on in some cases . Sweet!
kim // April 2, 2008 at 12:04 am

Hank, sunspots during the Maunder Minimum were sparse, inordinately large, and all in the Sun’s Southern Hemisphere. During the Dalton Minimum they were virtually absent. How do I know? Leif Svalgaard says so.

There is meaning there. What is it?
==========================
Heretic // April 2, 2008 at 12:07 am

Kudos to HB for exposing Watts for what he is, altough it seems Watts is doing a pretty good job or revealing himself!
That kind of individual gives a bad name to skepticism in general, a time honored virtue indispensable to any scientific undertaking.

The behavior of the individual, deleting posts, responses or the incompetence showed by the “analyses” he posts, which is more telling? What do you think, Kim?
kim // April 2, 2008 at 12:11 am

What do I think, Heretic. I think he’s a soul similar to mine, seeking understanding in a wondrous world, not someone certain and content in his faith. If the shoe fits; wear it.
========================
David B. Benson // April 2, 2008 at 12:43 am

kim // April 2, 2008 at 12:11 am — You would do a lot better to take the time to read S. Weart’s “The Discovery of Global Warming”.

Stick with the physics (and the statistics performed by the competent).
Hank Roberts // April 2, 2008 at 12:55 am

Kim, Leif’s answered your question.
Pointer for you in the relevant thread:
http://tamino.wordpress.com/2007/10/13/solar-cycle-24/#comment-16131
Hansen's Bulldog // April 2, 2008 at 1:10 am

Anthony Watts has posted this comment on the post in question.

We’ll see what happens.
kim // April 2, 2008 at 1:15 am

What do you think of the three independent confirmations of Watts’ and Copeland’s thesis?

Hank, see response on thread titled Solar Cycle #24. I don’t think the question is answered satisfactorily, yet.

Thanks, David, for the pointer to Weart’s work. I’ll peruse it. I may be too skeptical about carbon dioxide and clouds to absorb it though. I’ll try to be critical.
============================

[Response: They're not independent confirmations. They appear to be simply application of filters other than Hodrick-Prescott to estimate the times of warming-rate maxima.

None of them address the real problem: that the method used by Watts and Copeland to show correlation between the two series of timings (sunspot cycle maxima and warming-rate maxima) is completely invalid. They claim correlation of 0.99846454 between those two series, but I managed to get 0.9941 replacing their warming-rate maxima series with random numbers. Application of a proper test shows that the evidence for genuine correlation isn't even close.]
nanny_govt_sucks // April 2, 2008 at 2:29 am

So in fact there’s no evidence that warming-rate maxima, computed according to the method of Watts and Coleman, tend to fall at any particular phase of the sunspot cycle.

Can you run this test taking into account a 22 year sunspot cycle? Can you test odd vs even sunspot cycles? It is well know that the reversal of the sun’s polarity relative to Earth’s magnetic poles every 11 years can have dramatic effects. Just looking at 11 year cycles ignores this aspect of solar effects on Earth’s climate.
climatewonk // April 2, 2008 at 3:03 am

Reading over at Watts’ blog is telling. So many of his readers admit to not understanding the maths/statistics well enough to follow or judge, but still feel confident enough to take him at his word and marvel at the amazingly significant correlation he’s uncovered with his novel approach to statistical analysis!!!

They reject the whole of climate science for the past few decades, not trusting published climate scientists or their work but accept at face value one paper using a methodology they do not understand.

This is what it has come to — rejecting the preponderance of evidence for half-baked and erroneous “papers” published on internet blogs — what a sorry state of affairs.
Jeff C. // April 2, 2008 at 5:25 am

As Tamino mentions, Anthony Watts has posted a comment admitting that the correlation plot has problems. Okay, they screwed up. Let’s see where it goes from here.

As a first time visitor, I was reasonably impressed with your site, write-up and analysis. After wading through your moderated comments, it passed. Probably a third stated in one way or another - you really took that clown Watts down a few notches. Classy. Red meat for the believers won’t bring converts.

[Response: If only it were just a case of "Okay, they screwed up."

I've examined a number of the posts on Watts' site which attempt data analysis. Every one of them is plagued with "screw ups" at least as blatant as that described here; Watts has a pattern of promoting incompetent analysis in an attempt to contradict global warming. These aren't cases of rival methodology, or close calls, or failure to comprehend the subtleties of new and intricate analysis methods. They're blatant, even grotesque examples of incompetence, from Watts himself, Joe D'Aleo, Jim Goodridge, Basil Copeland, and I'd venture to guess, many others. I certainly haven't reviewed Watts' site with any thoroughness, but I have seen a fair number of his posts and I haven't yet seen one data analysis on Watts' site which is correct -- or even close.

This time, perhaps as a result of my previous critiques, he had his and Copeland's post reviewed by "a number of qualified people including a published climatologist, solar scientist, a statistician, and a certified consulting meteorologist (among others)..." Yet none of them caught a glaring problem which it took me about 15 seconds to see (and according to one reader's comment, I'm not the only one to notice it that quickly). There's no getting around the fact that Watts and his collaborators and reviewers simply are not competent to undertake data analysis. Yet he persists in obstinate promotion of falsehoods.

But this is the first time I'm aware of that he's admitted there even *might* be a problem. His response to previous critiques has been suggestive of evasiveness, even to the point of removing an entire post, which was later restored but without any correction or statement that it "has problems." As I said in commentary to an earlier post, if he had simply said "Oops -- my bad" then we would have gained respect for his character. Instead he has let his posts continue to mislead readers, in spite of having been fully informed of the problems. And in commentary here, he has always used polite language but has nonetheless engaged in ad hominem attacks against me personally, his favorite of which seems to be the implication that my choice to blog anonymously is cowardice.

In light of the pattern of grossly incompetent analysis which gives the false impression of discrediting climate science, the lack of corrections and admissions of error, the questionable reaction to criticism, it's no surprise that I and my readers are frustrated to the point that yes, we vent our frustration on Watts.

We'll see how he handles the current situation. He can go a long way toward redeeming himself with a clear and highly visible admission of the error of his ways.]
Charles // April 2, 2008 at 7:12 am

I have to agree with climatewonk and others expressing concern about those not knowing the science being suckered in by seemingly sound analyses like the ones of Watts and Basil. The public will reject the preponderance of scientific evidence available in the peer reviewed literature, largely because the public doesn’t know the the details and depths of the evidence, not being familiar with the literature. So they are easily influenced by blogs–and it looks as though those skeptical about AGW are seizing on this phenomenon. What concerns me are the illiteracies in the public around science and the workings of science.
fred // April 2, 2008 at 7:41 am

Joel Shore: thanks, exactly the right tone.

Lee: he blocks your comments because you have lost him due to their tone. Think about how much more effective Joel’s is - not in terms of releasing feeling, but in terms of getting agreement.

Chris: the tags are “blockquote” and “/blockquote” as in

this is a quote

at least I hope they are since we can’t preview here!
Heretic // April 2, 2008 at 8:30 am

Kim, that is truly funny. I know now how much attention you deserve, and that is how much Watts does, i.e. close to zero. The self righteous faith thing was the icing on the cake.
kim // April 2, 2008 at 11:40 am

Hank, comment #365 in the Svalgaard #4 thread has Leif’s answer. He’s agnostic about the question.
======================
Harold Ramsey // April 2, 2008 at 12:36 pm

1. Random number graph
Your random numbers seem a little too evenly spread to be random. You also choose to use 14 random numbers - why not 10 or 20? It is not random to choose an equal number of random numbers as solar cycles. The fact that there are 14 peaks in warming rate (excluding a couple of minor events) to match with 14 solar cycles over the same time period indicates a pattern in itself.

2. Warming cycle length vs solar cycle length

As you point out, there are not enough data points for this type of graph to be statistically significant - therefore the graph is pointless. It cannot show a correlation or non correlation. Why show it in the first place?
That said, I can clearly see a pattern. For the sunspot cycle lengths of 9, 12 &13 years there are only 1 or 2 data points so there is inadequate information and they can be ignored.
There are 5 data points for the sunspot cycle length of ten years. So if you take an average of these 5 data points you get an average of 10.4 years warming cycle length for solar cycle lengths of 10 years.
There are 4 data points for the sunspot cycle length of 11 years. The average of these 4 data points gives an average of 10.5 years solar cycle length for solar cycle lengths of 11 years.
For a small data set this seams a reasonable correlation. The average is only half a year different between solar cycle length and warming cycle length.!

I believe you are using statistical trickery to discredit some fairly obvious patterns.

[Response: Your criticisms, and your analysis, are so bogus that I'll leave it to others to tear you limb from limb.]
J.Hansford. // April 2, 2008 at 12:42 pm

Tamino… In your response to JeffC… You are a bit like the pot calling the kettle black….

If you were to use your own critique against yourself, you would have, time and again compromised your own integrity by defending Mann’s indefensible hockey stick graph…

… When really you would say that you were only exploring possibilities and interesting areas of significance…. Why is Mr Watts and Co suddenly beasts eating children, when they explore their interesting ideas and areas of significance…? ‘eh?

I suggest you look into the idea that there may be a strong correlation between Solar influences and Temperature…. Why not?

Instead of being scathing of mistakes you could be helpful in exploring other avenues of enquiry…. Considering how no empirical data supports AGW… It remains a Hypothesis based entirely on computer models…

Be less harsh in your criticism and more constructive… After all, as far as you are concerned, you are correct in your own mind…. So why should it be any skin of your nose….?

[Response: Clearly I don't agree with your assessment of Mann's hockey stick. But -- just for the sake of argument -- suppose you're completely right about that and I'm completely wrong. Then, to paraphrase Jeff C., "Okay, I screwed up." How does that compare to Watts & co. who have been wrong about every data analysis I've examined from his blog? How does being wrong about a complex issue involving application of a new and intricate technique, which is still in dispute by many experts and which renowned (perhaps the world's foremost) expert on PCA Jolliffe himself confessed not to understand in sufficient detail to evaluate, compare to repeated blatant and grotesque incompetence?

Answer: it doesn't. Mistakes are inevitable, many issues remain in dispute, but Watts' analytical incompetence is beyond dispute, and in his case mistakes aren't just inevitable, they're the norm. If a physician makes a mistake which leads to the death of a patient, he's likely to be held accountable in a court of law. But if he doesn't know a kidney from a kneecap, and he kills every patient he treats, he's deprived of his license to practice.]
EliRabett // April 2, 2008 at 1:21 pm

Just eyeballing Watt’s #5 I would have thought that the distribution of points in your update would not have been so evenly distributed. Of course, one of the issues to be avoided is which maxima in the temperature you assign to which in the solar output.

[Response: It's not quite as even as it looks, because three of the black dots represent two data points. Also bear in mind that Watts' #5 contains an error, plotting the maxima assigned to cycle 11 as exactly coincident when in fact that's the cycle for which the disagreement between sunspot and warming-rate maxima is greatest.]
climatewonk // April 2, 2008 at 2:01 pm

Considering how no empirical data supports AGW… It remains a Hypothesis based entirely on computer models…

Wrong.

AGW is based on theory and observation and computer model projections.

We have a theory of the greenhouse effect in which CO2 acts to trap heat from the sun that would otherwise escape, leading to a warming of the atmosphere, oceans etc.

We have observations of both CO2 increase and warming over the past thirty years in particular but over the period of the instrumental record and paleoclimate research that informs us about past climate. For observation, think Arctic melting, West Antarctic warming, permafrost melting, glacier retreat, temperature increases, sea level rise, changes in habitat and a host of other observational evidence.

We take the two and create computer models that project what might happen given certain increases in CO2.
fred // April 2, 2008 at 2:23 pm

“I’ll leave it to others to tear you limb from limb.”

No, please do not let us go around doing this. A more constructive suggestion would be that he verify using Excel Tamino’s statement that

two sorted series will always correlate

Try it. Generate two columns of random numbers in Excel, plot them, then sort them, and plot them again. And, everybody, if he comes back with questions, do not sneer. Just answer calmly and factually and constructively.

[Response: I understand the virtue in your approach. But I'm only human, and when someone slings feces at me I have an instinct to respond in the same way as most of my species.

There's also the fact that I get *so much* stuff like this, that it's impossible to keep up with it. Which makes me suspect there's also virtue in another of your suggestions, to be much more ruthless in moderation. That runs the risk, of course, of accusations of censorship, and of the actual fact of censorship.]
Barton Paul Levenson // April 2, 2008 at 2:23 pm

J. Hansford writes:

Considering how no empirical data supports AGW… It remains a Hypothesis based entirely on computer models…

It is based on radiation physics, and the theory was first published by Svante Arrhenius in 1896, some fifty years before the existence of computer models.
dhogaza // April 2, 2008 at 2:28 pm

Considering how no empirical data supports AGW… It remains a Hypothesis based entirely on computer models…

When denialists continue to repeatedly make flat-out incorrect statements like this, they have absolutely no reason to complain about scathing rebuttals.
Eli Rabett // April 2, 2008 at 2:34 pm

You realize that you have just shown Svalgaard is close to right. If the solar insolation difference between max and min is appreciable, there should be a strong forward bias in the Rayleigh chart, once you filter out the dominant greenhouse gas warming.
Lee // April 2, 2008 at 2:49 pm

fred,

Wrong. He blocked me when I asked him when he is going to get to part three of the histogram series, and reminded him that he had open statements that he had promised to support, and that he had made mild attacks on people for not having teh patience to wait..

Its been a month now.

And then, he went back through the old threads, and systematically removed old comments he had approved and often responded to.

He is blocking me because I was not not letting him get away with ignoring his promises to deal with his embarrassing mistakes. Before he removed the posts, he was on record as sayign, in response to one of my comments, that he was not going to let me mention those old promises. He SAID that, in public, and then removed it along with the response that asked him aobut those old promises.

I’ll give you tone now - he’s a dishonest twit, and his rewriting of the history on his own blog demonstrates that.
Hansen's Bulldog // April 2, 2008 at 3:01 pm

On a less contentious note:

I stated that after this post I was taking a hiatus from critiquing the analysis of others, and would post about real science. And so I shall. The very next post will be on a topic fred is eager to learn more about: feedback in the climate system.

It’ll take several days to prepare, perhaps even longer with a heavy work schedule, so please be patient.
Hank Roberts // April 2, 2008 at 3:17 pm

For reference:
http://math.ucr.edu/home/baez/crackpot.html
steven mosher // April 2, 2008 at 3:27 pm

Lee.

If I published that Paris France was located at 42.5N 72.5W and you pointed out my error, would you expect me to correct it?

If I refused what you you say?

If 4 years later, I repeated this same data error what would you think of me?
J // April 2, 2008 at 3:29 pm

Thanks, Tamino. I really, really like the update.

For fun, I duplicated your analysis … then repeated it using a simulated data set, where the temperature peaks were forced to be “correlated” with the solar peaks (plus or minus some random fraction of one year, for “noise”).

The difference between the two was obvious, visually. For the fake data, R was about 0.9, versus R for the real data (approx. 0.1).

I’m convinced — if you convert the dates of Anthony & Basil’s smoothed-temperature-first-difference-peaks to phases of the solar cycle, they’re actually close to random.

Insofar as there is anything useful to be learned from this method (filtering, first-differencing, and identifying local maxima in the temperature data) … it seems to show that there isn’t any significant imprint of solar cycles on climate. Logically, Watts should probably admit either that his method or his conclusions are completely wrong, though I won’t hold my breath.

[Response: There are other issues with the analysis as well, particularly the accuracy with which rates of change can be determined by filtering and 1st-differencing.]
steven mosher // April 2, 2008 at 3:35 pm

HB.

Looking forward to the feedback stuff. Actually what would be nice ( and gentlemanly) would be a comparison of Schwartz and Lucia on lumped parameter models. Lucia has some issues with schwartz, different from yours, and she had some issues with your analysis of schwartz.

Best of all would be a three way discussion between yourself, annan and lucia. Not a war.
I dont think james’ blog gets the attention it deserves, like atmoz.

So. I think a three way discussion between you, lucia and Annan would be a good treat for the math inclined. Not sure how to organize such a discussion/interaction blog wise.

Maybe guest posting. could be more fun than mud wrestling. everyone takes a turn guest posting at the other site. sumthin like that.
J // April 2, 2008 at 3:38 pm

kim writes: “Hank, comment #365 in the Svalgaard #4 thread has Leif’s answer. He’s agnostic about the question.”

Actually, that’s not quite correct. Leif’s answer to your question is in comment #342: “as this is spurious it is hard to make sense of, but I would say that the longer a cycle is, the more it warms. Here I assume that at solar minimum we get a certain base amount and that within each cycle we get a little bit extra due to solar activity, so the longer the cycle the more extra we get. But this is just hand waving, there are people that say we get cooling, warming, nothing, whatever. We are past logic here.”

I agree. The claimed correlations are spurious, hard to make sense of, and mostly just hand-waving. The people who obsessively try to pin down relatively minor connections between solar cycles and climate are “past logic here”.
kim // April 2, 2008 at 4:07 pm

J, different questions. The answer in #365 is about sparsity of sunspots at climate minima.

But, back to the other question. Aren’t longer solar cycles weaker? Leif dodged that one, hiding behind his agnosticism.

There is too much evidence linking the sun’s output to climate changes. There is too little evidence of the causative mechanism.
=================================
JCH // April 2, 2008 at 5:14 pm

“If 4 years later, I repeated this same data error what would you think of me? …” - SM

Is the mistaken location of Paris mentioned in both abstracts?
Hank Roberts // April 2, 2008 at 5:20 pm

Yep. See the thread there for that discussion, it’s off topic here.
Search that thread for “astrology” …
Barton Paul Levenson // April 2, 2008 at 5:32 pm

I don’t think some folks understand why what Watts and Copeland did is wrong. Let me give an example using non-climate data.

Here are the inflation rate (GDP deflator) and money growth rate (new M2) for the 1970s:

YeardPdM
19705.296.57
19715.0013.38
19724.3412.95
19735.586.63
19749.035.45
19759.4312.65
19765.7813.36
19776.3510.27
19787.037.53
19798.297.88

If you take a Pearson’s product-moment correlation coefficient between the two columns of data, you get r = 0.2978. Since variance accounted for goes as the square of r, we can say 9% of the variance of inflation was accounted for by changes in the money supply.

Now, here is the same table sorted in ascending order by dP — from the smallest value of dP to the largest:

YeardPdM
19724.3412.95
19715.0013.38
19705.296.57
19735.586.63
19765.7813.36
19776.3510.27
19787.037.53
19798.297.88
19749.035.45
19759.4312.65

The order of the points doesn’t matter. The correlation is still r = 0.2978.

But now, let’s say you sort dP from smallest to largest, AND sort dM from smallest to largest, independently — matching the lowest dP with the lowest dM, the second-lowest dP to the second-lowest dM, etc. Here’s the data, with the years sorted along with dP but dM sorted independently:

YeardPdM
19724.345.45
19715.006.57
19705.296.63
19735.587.53
19765.787.88
19776.3510.27
19787.0312.65
19798.2912.95
19749.0313.36
19759.4313.38

Now, if you figure the correlation, r = 0.9501, and 90% of variance is accounted for rather than 9%. An astoundingly high correlation!

Except that it isn’t. By sorting the columns independently, you’ve decoupled all the points! The smallest dP occured in 1972, but the smallest dM occured in 1974! It’s no longer a fair comparison because you’re not comparing like with like any more. You’ve spuriously enhanced the fit by pre-matching the orders of the points. In statistical science you’re simply not allowed to do this. It’s bad, it’s wrong, it’s incorrect, and most importantly, it’s meaningless. That’s what Watts and Copeland did in their solar article.
luminous beauty // April 2, 2008 at 5:34 pm

“If 4 years later, I repeated this same data error what would you think of me?

If this ‘data error’ is nothing more than an insignificant typo, I would suspect you might be repeating it just to tick me off.

I’d let it go. Why can’t you?
Barton Paul Levenson // April 2, 2008 at 5:39 pm

Well, that looked awful. Apparently this blog doesn’t take HTML tables, though it does take many other HTML tags. Here are tables 1, 2 and 3 again:

Year   dP   dM
1970   5.29   6.57
1971   5.00   13.38
1972   4.34   12.95
1973   5.58   6.63
1974   9.03   5.45
1975   9.43   12.65
1976   5.78   13.36
1977   6.35   10.27
1978   7.03   7.53
1979   8.29   7.88

Year   dP   dM
1972   4.34   12.95
1971   5.00   13.38
1970   5.29   6.57
1973   5.58   6.63
1976   5.78   13.36
1977   6.35   10.27
1978   7.03   7.53
1979   8.29   7.88
1974   9.03   5.45
1975   9.43   12.65

Year   dP   dM
1972   4.34   5.45
1971   5.00   6.57
1970   5.29   6.63
1973   5.58   7.53
1976   5.78   7.88
1977   6.35   10.27
1978   7.03   12.65
1979   8.29   12.95
1974   9.03   13.36
1975   9.43   13.38
Hansen's Bulldog // April 2, 2008 at 6:03 pm

I received a very conciliatory email from Anthony Watts — an olive branch, if you will. It argues well for his character, especially considering the scathing criticism I’ve had for his posts.

So I urge everyone to set passions aside, and for the time being confine discussion to the facts, dispassionately.

Tamino, aka Hansen’s Bulldog
climatewonk // April 2, 2008 at 6:43 pm

So I urge everyone to set passions aside, and for the time being confine discussion to the facts, dispassionately.

Civility and generosity should be employed in debates, however, where warranted. Sometimes, civility is not warranted and generosity dangerous. I am willing to give Watts the benefit of the doubt if he is willing to amend his post to show that he has made some critical errors in analysis. Otherwise his erroneous analysis will be incorporated into the denialist database of “evidence” against AGHG as a cause of warming and for solar causes of warming.

If he made an honest mistake and the analysis is flawed, he should just come right out and state as much. He should revise his post and acknowledge that his initial findings were premature. He owes it to his readers.

If he is not willing to do so, he should expect critics to be merciless and if that is deemed unacceptable to him and his supporters, he should just give up engaging in public analysis that is apparently beyond his skills .

[Response: I agree with you. But overtures of peace are rare and should be treasured, so please let's set this discussion aside for now. If Watts fails to address important issues to your (or anyone's) satisfaction, there'll be plenty of time to criticize him for that -- but at least let's wait until we find out.]
Brian D // April 2, 2008 at 7:08 pm

Setting aside passions, I should mention that a private “olive branch” (which I sincerely believe he sent) that is kept secret, while leaving the offending analysis public, serves no purpose except to placate those most critical of it while still spreading disinformation.

Watts needs to make his apology public and correct the mistake before I’d be willing to accept his apology as sincere and restore some of his credibility — but then again, I’m just a commenter on a different blog.

Still, out of respect for our host, I’ll silence my criticism for now, giving Watts time to make the necessary move.

[Response: I would think the opposite: that a public olive branch (say, a comment here) might just be a show for readers, while a private one (which Watts could not have expected me to mention) indicates sincerity.]
nanny_govt_sucks // April 2, 2008 at 7:14 pm

So I urge everyone to set passions aside, and for the time being confine discussion to the facts, dispassionately.

Who are you, and what have you done with Tamino?

I wonder if you can run some of your sunspot cycle correlations with some of the other known sunspot cycles, including looking at odd and even 11 year cycles vs temperature. This seems to be where some of the Watts/Basil analysis is going.
steven mosher // April 2, 2008 at 8:18 pm

JCH. is the mistake in the abstracts? Weird question. funny distinction.

But anyway, no the mistake I ask you consider and render an honest opinion about is a mistake in the data used by the study published in the SI.
If Fred Singer published a paper and supplied data supporting his study. AND IF that data showed he thought paris france was located at 42.5N 72.5W, AND IF he were notified of this error, and IF he repeated this same error 4 years later, what would your opinion of Fred Singer be?

What if Tamino made such an error? I suspect he would fix it. What if Anythony Watts made such an error? would he correct it? if he didnt what would you say? What if you made such an error?

Hansen made an error in GISTEMP. Others found it, he fixed it. And he wrote it up.
Here is my error. Here is my fix. here is the impact. It’s simple. we all respect that and the issues goes away, eventually. It would have been nice to credit the people who found the error, but never mind. thats manners not truth.

So. simple question. no one wants to answer.

If I told you that people were evacuating their homeland due to global warming, and that was not true. Would you expect me to correct that mistake? And if I wouldnt, what would you think of me?

Pardon me, I have to duck some sniper fire here in Bosnia.
L Miller // April 2, 2008 at 8:19 pm

I don’t think some folks understand why what Watts and Copeland did is wrong.

I think it’s the fact that there is no clear “sorting” step in his analysis that is confusing people. His final result is basically identical to a sorted random series, he just doesn’t get there using a sort.

Another way you can look at it is that he is comparing a series against itself, and adding a small random variation to one of the terms. Since there are a couple leftover peaks it’s pretty easy to choose the ones that line up with what he wants to see and any temperature change peak can’t be more then a few years from a sunspot peak. This will be a basically random but ultimately small distance from the solar cycle peak assigned to it.

I.E. say x(1), x(2), … are random number between -5 and 5 then he is plotting:
Year Vs Solar Cycle
And
Year + x(n) vs Solar cycle

Which is really no different then comparing a series to itself as far as R^2 correlation is concerned. That’s a little simplified because the range for x won’t necessarily be exactly the same each time.
L Miller // April 2, 2008 at 8:22 pm

bah. I didn’t close the blockquote tag properly

[Response: Fixed]
cohenite // April 3, 2008 at 1:47 am

Before you dismiss the link between Schwabe and Hale solar cycles and global temperature fluctuations completely, may I suggest a peak at this paper which seems to address your complaint about Watt’s susceptibility to randomness:
http://www.fel.duke.edu/~scafetta/pdf/opinion0308.pdf

[Response: Please pay better attention. I did not dismiss the link between solar cycles and global temperature changes, I refuted the analysis of Watts and Copeland. In fact, as I already said quite unambiguously in response to another comment, I'd be surprised if there's *no* connection, but the failure of so many legitimate analyses to establish it argues for its very low level, and I have doubts about even the most plausible detection (due to Camp & Tung).

And as I said before, when someone linked to the exact same opinion piece on another thread, there are serious problems with the work of Scafetta & West, many of which have been addressed on RealClimate (see also the links within that link). Even in the opinion piece you link to their graph already indicates problems, particularly the phrase "smoothed to stress the 11-year modulation."

It's no surprise tha blog posts are vulnerable to substandard data analysis. But what most non-scientists are not aware of, is the journal publications are too. The quality of work is certainly a lot better than blogs, but it still amazes me what sometimes makes it past peer review into the literature.]
Dano // April 3, 2008 at 2:28 am

kim // April 2, 2008 at 12:04 am

There is meaning there. What is it?

The kim bot sows the D portion of FUD. It doesn’t work here. At wingnut sites, surely. Redstate, definitely.

kim bot programmer: new search algorithm and response parameters please! This routine grew stale and tedious about 1999.

Best,

D
EliRabett // April 3, 2008 at 2:45 am

Nanny, I don’t think this sort of analysis is good for what you want. Better some sort of Fourier transform to see if there are significant 11 year/22/60 year cycles in the global temperature and the 60 only would allow for ~3 cycles in the instrumental record, in other words, chancy to dig out.

Another thing to try would be to look at the temperature north of 60 degrees latitude where the swings might be larger.
fred // April 3, 2008 at 7:53 am

Looking forward to the feedback discussion. This is really the central part of the question. If you have space, it would be nice to have some consideration of Spencer’s remarks about sampling and correlation and causation.
Adam // April 3, 2008 at 9:43 am

This is vaguely related so I’ve posted it here:
http://news.bbc.co.uk/1/hi/sci/tech/7327393.stm

See “related links” on the RHS for more information.
steven mosher // April 3, 2008 at 12:45 pm

luminous, if it were an insignificant typo, you’d thank me, fix it and not repeat the error.

what if I got all the locations in my data set wrong?
climatewonk // April 3, 2008 at 1:49 pm

Steven Mosher, can you explain the repeated references to coordinates and Paris? I do not know the background to this and it appears to me to be nothing more than rude behavior on your part to keep repeating it.
fred // April 3, 2008 at 2:28 pm

climatewonk

Mosher is referring to the fact that in one of his datasets, Mann made an error about the location of Paris. A couple other places too if I recall correctly. It was probably a typo, a line feed in the wrong place or something. Obviously he knows where Paris is. I don’t recall anyone saying it made any material difference to the results. Mosher is saying that not only did he do it once, he continued to use the same erroneous data set after it was pointed out, which was a little silly.

This is what the remarks are about. They are not very pertinent to the present discussion, but perhaps Mosher is arguing that standards should be applied to Mann if they are to be applied to Watts. Probably so, but it doesn’t make the Watts charts any more correct.
luminous beauty // April 3, 2008 at 2:47 pm

mosh,

“luminous, if it were an insignificant typo, you’d thank me, fix it and not repeat the error.”

If you were a persistent jackass, ranting excessively about meaningless gobbledygook, I would ignore you, and not make an insignificant correction, just to point out that I am ignoring you.
steven mosher // April 3, 2008 at 4:55 pm

Thanks Fred.

The point is this. We all make mistakes. I made one about what Tamino wrote. I corrected it.
Tamino made one about what lucia wrote. He corrected it. Lucia made a an error about log versus ln She corrected it. Christy and Spencer made and error. they corrected it. GISS made a mistake in GISS temp. They fixed it. Hadcru made a mistake. They fixed it. And the theory of AGW still stands as the best explanation of the warming we see. And now I see people asking AW and Basil to correct what tamino documents as an error. Fair enough.

Then we have this:

Mann not only got Paris wrong, he got EVERY location wrong ( except one) . It was pointed out in 2003, and in 2007 he published another paper using the same mistaken locations. Now,
it’s a simple matter for everyone to say. “He made a mistake, AGW is still the best explanation. ” But nobody, seems able to say that.

I find that odd. I find it odd and noteworthy that you cant even get some people to say simply
” ya Gore made mistakes in AIT” there is always a caveat, always a disclaimer, always a shading..
“err he got the tense of verb wrong ” that was a precious one from RC.

It’s odd.

For the record. I was wrong to compare the hockey stick to the Piltdown man.
Ken // April 3, 2008 at 5:51 pm

HB, sadly I don’t make it to your site daily. When I do, it seems like I’ve already missed out on all the discussion. But I wanted to commend you for these 3 excellent posts, nonetheless. This whole series reminds me of the quote:

“A little knowledge is a dangerous thing”

While I commend anyone that enjoys working with data, making plots, and trying to understand phenomena, I must say that Watts and his blog are an example of the Dunning-Kruger effect.
Mike B // April 3, 2008 at 6:19 pm

It’s not just “a little knowledge is a dangerous thing” although that is certainly part of it. It begins with the fact that they KNOW something other than CO2 must be causing the current warming and therefore their results no matter how contorted and in many cases ridiculous in nature, show that something else is the culprit.
climatewonk // April 3, 2008 at 6:39 pm

Mann not only got Paris wrong, he got EVERY location wrong ( except one) . It was pointed out in 2003, and in 2007 he published another paper using the same mistaken locations. Now, it’s a simple matter for everyone to say. “He made a mistake, AGW is still the best explanation. ” But nobody, seems able to say that.

Ahh, this is the case of the transposition of one line of data that occurred and led to some strange results and interesting speculation, IIRC. It made no difference to the ultimate conclusions, I seem to remember.

I wonder why you keep bringing this up except to flog a favorite dead horse? It has no bearing on whether Watts’ analysis is valid or mistaken and appears rather juvenile.
Boris // April 3, 2008 at 6:44 pm

what if I got all the locations in my data set wrong?

Getting one location wrong seems kind of minor to me. I don’t know exactly what your “gotcha” point is here, mosh. I’d probably, say, “boy, that’s dumb.” and move on.

It’s not like someone is going around the internet using this mistake to argue that Paris is where it is not, right? Not the way the words of Fred Singer, Patrick Michaels, Tim Ball and Roy Spencer are used to support similar misconceptions, right?
Hank Roberts // April 3, 2008 at 6:53 pm

> It was pointed out in 2003, and in 2007
Which journal? Was the criticism published as a letter, comment, or correction in the same science journal? If not was it submitted to the journal?

What’s important is whether there’s a difference in the result section. This isn’t religion where a minor flaw in something attributed to a founder can topple all subsequent work.
TCO // April 4, 2008 at 1:51 am

Agreed. All the more reasons for people on either side to ALWAYS concede true points…because they’re TRUE.
climatewonk // April 4, 2008 at 1:31 pm

TCO, it would be important to always concede true points if this was just another academic debate over methodo

Open Mind