AP IMPACT: Statisticians reject global cooling?

There is a slightly weird AP news article by Seth Borenstein which purports to show that In a blind test, the AP gave temperature data to four independent statisticians and asked them to look for trends, without telling them what the numbers represented. The experts found no true temperature declines over time. All very well, taken literally, but I thought the world was supposed to be warming, not not statistically cooling.

But it gets worse:

The AP sent expert statisticians NOAA's year-to-year ground temperature changes over 130 years and the 30 years of satellite-measured temperatures preferred by skeptics and gathered by scientists at the University of Alabama in Huntsville. Statisticians who analyzed the data found...

Are we really supposed to believe that the statisticians who received this data didn't recognise it? You have to be fairly out of touch not to know what the global temperature record looks like.

[Update: a bit more detail here. See-also the comments re the "blind" aspect. This story has tickled a number of fancies: for example CM, DSB, JR. And Eli, naturally.

I've struck "slightly weird" - on reflection, that wasn't called for -W]

More like this

Why do you assume the statisticians would know (or figure out) what the data was? Not every statistician works in the climate field. We have excellent statisticians at my cancer center, but I doubt if I showed them such data that they would identify it for what it is.

The AP put up more details on how temperature data was [sic] analyzed. It not only names names (I don't recognize any of them, but I'm not in this field), but also says "None of the experts detected a downward, or cooling, trend in the numbers. All saw a distinct upward trend." (i.e. they found warming, not not statistically cooling").

Jumping the gun a bit on the snark, are we?

[My snark is on a hair-trigger -W]

As the reporter behind the story let me also explain this: The spreadsheets I sent the statisticians showed temperature anamolies in degrees C, thus fractions of 1 or so. Thus it did not look like global temperatures. Only one had indicated that he had figured out it was temperatures when I revealed to them what they were. The other three said they didn't know. One said he was tempted to do some snooping, but said he resisted.

[Thus it did not look like global temperatures - but the first thing they are going to do with a new dataset is plot it out. What units it is in won't matter - the look will be obvious. That aside, perhaps only one did spot it, but it is also possible that all did and are being polite. And - forgive me - the story says it was a blind test for 4. At best, as you've said, it was a blind test for 3.

I'd say that the satellite record for the last 30 years is far less familiar and more likelt to go unrecognised. Unles you give it to people at the sametime as the full T record -W]

By Seth Borenstein (not verified) on 27 Oct 2009 #permalink

That must be some sort of record - straight to the source in 3 comments!

talk about street cred.

By thomas hine (not verified) on 27 Oct 2009 #permalink

The spreadsheets I sent the statisticians showed temperature anamolies in degrees C, thus fractions of 1 or so. Thus it did not look like global temperatures.

Yeah, I gotta say that doesn't really do a lot to hide it. Anyone with a tangential knowledge of the subject will probably recognize that data set for what it is.

[It might have helped to reverse the data series. Or perhaps de-trend it (or add some other fake trend to it) first. It would also be nice to see exactly what dataset and what text the statisticians were given (ha: Brian has already linked to that. Oops), and exactly what they said in reply -W]

However, since only one of them recognized it, it seems you chose your statisticians well for this purpose, Seth.

But seriously, we know that the cooling trends are cherry-picked from 1998 due to the extremely powerful El Nino that year and the corresponding warm sea surface temperatures. I don't see the problem with pointing this out by doing the sort of "blind taste test" with three people presumably unfamiliar with the brand of cola they're trying out.

By Moderately Unb… (not verified) on 27 Oct 2009 #permalink

It was a blind test for all four. Because the one who thought he figured it out, wasn't told until afterwards. He wasn't certain, just in an email response he alluded to it, saying if these were temperatures. Think of it this way, it's just like how most recipients of new medications in drug trials think they are getting the real drug and about half are right? They're not certain and they're not told. Same thing here. The other three were not being polite. They said they were surprised by the data. It was blind for all four. Period.

[I think you're being a touch generous to yourself. If #4 asked if these were temperatures, then it seems fair to think he guessed. Since this is becoming a bit touchy, I may as well throw into the mix: it is interesting how uncritically this is being jumped on -W]

By Seth Borenstein (not verified) on 27 Oct 2009 #permalink

Too bad Mr. Borenstein simply did not just add 523 or some other arbitrary amount to all the temperatures in the data set. Likewise add some number to the time. It would not change the trend whatsoever and it would have virtually eliminated any chance of recognition. For extra security change (time, temperature) to (temperature, time) and again in it would not change any trend. Just ask if the change in the x value from y=whatever to y=whatever is significant.

But then again, hindsight is 20-20 and the basic point has been correctly made: an unbiased look at the data disproves the denialist claim.

And even if they recognized the data set, whether or statistical significance at x confidence exists for a precisely defined trend for a very particular data set is hardly something different statisticians could get a different answers for assuming no careless errors were made. Determining statistical significance is something that undergraduates are often asked to do -- even those who are not math majors.

(Of course once you have the answer of yes or no for statistical significance you might have a disagreement over the interpretation of that result, but that is another story. But is clear that the denialists would expect statistical significance assuming they both understood the concept and are honest with their decreasing temperature claims.)

[Adding a constant doesn't help much, since the shape stays unaltered. I'd hope that he munged the time axis away from calendar years otherwise it makes the task even easier, though now you mention it I don't think he did. I think reversing it is the easiest obfustication; adding a random trend is better and still quite easy -W]

Actually, ol Stoat, Eli finds this quite believable having had some recent interaction with a whole bunch of people who you would think were much more in on climate stuff and who turned out to be totally naive.

You probably could do the whole thing over again, twist the data how you like (for example turn the years into cardinal numbers, list em backwards, etc) and get the same result. At some point skepticism turns into cynicism. Watch out.

[Well yes. Maybe I've stared at these things too long. But it does look like people are picking this up somewhat uncirtically. Perhaps it would be better to emphasise the obvious: this study tells us nothing new. We already know it is warming, and that it hasn't been cooling over the last 10 years -W]

Geoff Wexler

Not not cooling?
Were they just classical statisticians who decided to reject the null hypothesis that it was cooling? Would the result have been different if they had been Bayesians? David McKay's impressive book * (e.g. p.459 section 37.1) suggests that it might well have been:

http://www.inference.phy.cam.ac.uk/itprnn/book.html

Incidentally, if you were to tell Bayesians the meaning of the data, they might have found it easier to improve their asessments.

*. [Not to be confused with his book on sustainable energy.]. By the way I am only guessing, since it is not my subject.

By Geoff Wexler (not verified) on 27 Oct 2009 #permalink

Perhaps it would be better to emphasise the obvious: this study tells us nothing new. We already know it is warming, and that it hasn't been cooling over the last 10 years

Perhaps it *would* be better to emphasize the obvious:

The article wasn't written for us.

It's written for those who've been sucked into denialist claims that recent cooling is significant and that those fraudulent hucksters in the climate science community are lying when they say it's not significant.

Whether or not he succeeded in getting a blind sample of 3 statisticians his idea had merit: run the numbers by statisticians who aren't part of the "fraudulent climate science community" we're always hearing about (leading Republican members of Congress make this claim, remember), don't tell them what the numbers represent, and ask them what they find.

[I agree entirely that the idea had merit. Unfortunately it also had obvious flaws which are easily spotted by anyone who thinks about it; it would have been easy to improve -W]

@WC:
the look will be obvious

This seems incredibly presumptuous and not a little bit naive. Why on earth should it look obvious to those not concerned with this or related subjects? It's not as though data sets with similar shapes are terribly uncommon.

@dh
The article wasn't written for us.

Just so.

Of course, if Borenstein had done a transform on the data (flipping it upside-down, for instance), the denialists would have accused him of manipulating the data to get the result he wanted, no matter what "adjustment" he used. It'd be like the SurfaceStations crowd all over again, except without easy mockery.

That, plus what dhogaza said.

Re: #13
Its a pity about Clive James. Clever, but climatologically naive people, are good at amplifying their raw material (e.g from rogue BBC articles) with lots of damp hot air. With all this positive feedback, public opinion will continue to be very unstable.

By Geoff Wexler (not verified) on 28 Oct 2009 #permalink

I agree entirely that the idea had merit. Unfortunately it also had obvious flaws which are easily spotted by anyone who thinks about it; it would have been easy to improve

He might've consulted a statistician beforehand, I guess - "help me make this a real blind test". I'll cut him some slack, though, because, since he's a journalist, I don't expect him to execute it as skillfully as you wish, perhaps.

And blind test or not, four independent statisticians did verify what we all know to be true, and which so many have been lying about. That's a good take-home point for readers regardless of whether or not he succeeded in getting a truly blind analysis.

[It isn't quite clear that 4 did. If you read SB's comments here, it looks like at least one guessed, so that is only 3 who did it blind at best -W]

The fact is, much of this discussion is moot. "Statistics" are a property of the data. "Conclusions" are a property of the statistician. **It doesn't matter how disguised the data was.**

Consider this: after any (good) statistician analyzes the (possibly disguised) data and you ask him if there is a trend, the first thing he is going to ask you is: "well, over what interval in the independent variable?"

The answer to THIS question is really the thing at issue here. And while Borenstein could have been a little more explicit in properly stating the problem, I thought he did a decent job at answering it.

It isn't quite clear that 4 did. If you read SB's comments here, it looks like at least one guessed, so that is only 3 who did it blind at best

Above in my previous post I said 3, essentially conceding the 4th guy.

As far as what I meant with this

And blind test or not, four independent statisticians did verify ...

(well, I thought it was clear, guess not. :)

[Oh I'm sorry - I missread your comment. I think I transmogrified "indep" into "blind". OK, we agree now -W]

was this - regardless of whether or not the test was blind, four independent statisticians did verify ...

Independent from the "fraudulent climate warming alarmists", working independent of each other thus not influencing each other's work, etc etc.

This is mildly interesting:

The AP contacted University of South Carolina statistics professor John Grego, a longtime reliable statistics source. In addition, the American Statistical Association sent an e-mail request from the AP seeking statisticians willing to examine certain sets of numbers and look for trends without being told what those numbers represented.

Three professors of statistics agreed...

I'd say it's a pretty fair bet that Grego was the guy who guessed, since he's been a "longtime reliable statistics source" for the AP. Probably a bit more tuned in to how newspaper types think and operate ...

W.,

I don't think you're giving Seth Borenstein enough credit here.

A lot of the "cooling" or "flat" global temperature stories don't even mention decadal trends or moving averages. Borenstein refers to both, and presents the specific example of comparing 1999-2008 to 1989-1998.

This article is a welcome antidote to the confusion of Andy Revkin who got badly spun by the likes of Patrick Michaels.

http://deepclimate.org/2009/09/25/nyts-andy-revkin-backtracks-but-not-n…

Was Borenstein's article perfect? No, but if you can point to a better explanation of temperature trends (and debunking of the persistent "no warming since ..." meme) in the mainstream press, I'd like to know about it.

Besides, RP jr didn't write about it, so it must be right. (RP jr's current preferred source on trends is ... Chip Knappenburger, who wrote "an informative and fair post" entitled A Cherry-Picker's Guide to Global Temperature Trends.

http://rogerpielkejr.blogspot.com/2009/10/cherry-pickers-guide-to-globa… )

Yikes!