2012-02-02 17:01:32Updating The Escalator
Dana Nuccitelli

Want to strike on this one while the iron is hot - please get comments in ASAP

Updating The Escalator

2012-02-02 17:32:08
Kevin C


I'd leave out this comment:

Unfortunately I've lost my GIF program, so the NCDC version is a bit choppier than the original Escalator.  Any graphics wizards are invited to create a prettier version, and to test out different short-term cooling periods.

I wish I could evaluate Briggs' claims. Unfrotunately I can't work out what he is talking about. Ideally a real statistican with cerdentials to argue with Briggs would step in at this point. Anyone got any contacts?

2012-02-02 17:39:15
Brian Purdue


I suggest heading change “John Cook, Denial Clairvoyant” – sounds less like he belongs on the funny farm.

Graph doesn’t look as bad so will this incite denial attack of cherry picking BEST data?

A truism - get them an inch and they will take a mile.

2012-02-02 17:54:32
Kevin C


Here's the list of usable trends, assuming start and end point on year boundaries to keep it short. I can do a longer list if you want. Columns are start year, start month (Jan=0), number of months, trend.

1970   0   84  -0.017969
1972   0   60  -0.044599
1977   0  108  -0.000365
1978   0  108  -0.000274
1979   0   96  -0.009919
1980   0   72  -0.026588
1980   0   84  -0.017320
1981   0   60  -0.036706
1987   0   84  -0.013745
1987   0   96  -0.008003
1988   0   72  -0.019238
1989   0   60  -0.023177
1990   0   60  -0.030848
1990   0   84  -0.002834
1997   0   60  -0.015970
2001   0  132  -0.001854
2002   0   84  -0.011463
2002   0   96  -0.007463
2002   0  120  -0.005178
2003   0   72  -0.014032
2003   0  108  -0.005427
2005   0   84  -0.008803

A quick inspection suggests:

1970(7yr), 1977(9yr), 1986(8yr), 1990(7yr), 1997(5yr), 2001(11yr)

Comparison with your version suggests that you got it right first time.

2012-02-02 18:08:35
Kevin C


The third frame in the current animation isn't aligned with the rest.

2012-02-02 18:18:57
Tom Curtis


It is quite noticable that the last down step is the longest of those shown.  The risk in that is deniers will pick on that additional length as being significant.  If you shift the start of the second downstep about 10 months earlier, that would help break up that impression.


Also, if you switch to GISS for your data source, you could use this image by Tamino showing the error estimates of the trends for trends starting in a particular year and terminating in June, 2011:


That would be helpfull in the discussion of error estimates.  Alternatively, Foster may tell you the technique so that you can draw a similar graph for the NCDC data.

Tamino's discussion is here.

For comparison, here is the equivalent figure for the adjusted data from Foster and Rahmstorf, in case you believe it to be more usefull for this discussion (although I do not):

2012-02-02 18:24:54
Tom Curtis


Did you really intend your last link to be to Bishop Hill?

2012-02-02 18:47:43
Brian Purdue


Is there no way of presenting the same clean cut image? The new one gives the impression of a significant change to the graph.  If not a “brief” explanation is required.

Edit: I think the post should be held back until the same image can be produced because our "friends" may not be happy either. 

2012-02-02 19:12:41
Rob Painting

"I don’t know what the prediction uncertainty is for Plait’s picture. Neither does he. I’d be willing to bet it’s large enough so that we can’t tell with certainty greater than 90% whether temperatures in the 1940s were cooler than in the 2000s"

This is just frickin ridiculous. Does it need to be in the post? You are just reinforcing a pathetic bit of look-squirrel!

And the guy is supposed to be a statistician?

2012-02-02 21:10:56
Kevin C


Here's the data for gistemp (just annual intervals this time):

1970   0   84  -0.017600
1972   0   60  -0.046910
1972   0   72  -0.005205
1973   0   60  -0.013037
1973   0   72  -0.002720
1977   0  120  -0.000945
1979   0   84  -0.013515
1979   0   96  -0.011211
1980   0   72  -0.032585
1980   0   84  -0.024265
1980   0   96  -0.005654
1981   0   60  -0.042205
1981   0   72  -0.028379
1986   0   96  -0.008792
1986   0  108  -0.006108
1987   0   84  -0.027938
1987   0   96  -0.019890
1987   0  108  -0.005203
1987   0  120  -0.002344
1988   0   72  -0.040625
1988   0   84  -0.026606
1988   0   96  -0.006671
1988   0  108  -0.002848
1989   0   60  -0.044979
1990   0   60  -0.051098
1990   0   72  -0.009797
1990   0   84  -0.003141
1995   0   72  -0.000698
1997   0   60  -0.009303
2001   0   96  -0.000934
2002   0   84  -0.009054
2002   0   96  -0.003533
2002   0  120  -0.000177
2004   0   60  -0.010940
2005   0   60  -0.018148
2005   0   84  -0.005908

Looks like

In fact, you could chop the long run at the end of the NOAA data in two to get steeper declines: something like 2001(8yr) 2005(7yr)

Tom: Does Tamino's graph help? To a knowledgable reader, the error bar doesn't cross 0.2, but if this article is going to be widely read by people unfamiliar with statistics, the dropoff at the end looks awfully convincing.

2012-02-02 21:21:24
Brian Purdue


Dana – you need to slow animation speed down to same as old graph. There is now 5 steps instead of six with no overlaps. Is this due to changed data?

I agree with Glenn that it must not look like a reaction to deniers jumping up and down.

I think you should say the old graph is still applicable because the data used has not changed. It’s not a good idea to say old graph has “drawbacks” but say new graph has been produced to reflect data from other sources that have been updated still old graph was draw.

I think this could turn into walking through broken glass if we are not very careful.

2012-02-02 21:40:27
Tom Curtis


Kevin C, if we are going to discuss Briggs' comments on uncertainty, then the graph is usefull.  In the first instance, it shows that the errors of the trends have been calculated.  In the second, it shows that contrary to his claim of errors of the order of +/- 0.1-0.5 degrees C, the largest error in the trend analysis is plus or minus 0.025 C/yr, or one one quarter to one twentieth of the error Briggs claims is likely.  More directly, the error in the trend from 1979 is +/- 0.0035 C/yr on a trend of 0.017 C/yr, or 20%.  Where Dana writes, "In the BEST data in recent years, the uncertainty (95% confidence interval) is in the ballpark of 0.06-0.1°C, as compared to the warming trend of approximately 0.3°C per decade (land-only)", the range is 20-33% error, and the error used in the error in a year, not the error in a trend.

I am not convinced that including Tamino's graph would improve the post, but if we are going to discuss error I suspect it would. 


Edited to note:  Gistemp has an error of plus or minus 0.1 C for comparison of nearby years at the beginning of the twentieth century, declining to 0.05 C by the end.

2012-02-02 21:49:38
Kevin C


How about replacing the paragraph before the figure1 with this:

The original Escalator was based on the Berkeley Earth Surface Temperature (BEST) data, which incorporates more temperature station data than any other data set, but is limited to land-only data; additionally the record terminates in early 2010. It is interesting to apply the same analysis to a current global (land-ocean) temperature record to determine whether short term trends in the global data can be equally misleading. A global version of the Escalator graphic has therefore been prepared using the NOAA NCDC global (land and ocean combined) data through December 2011 (Figure 1).

Tom: Yes, that makes sense. I guess the questions are: a) is Briggs worth responding too, and b) should it be done in a separe post.

2012-02-02 21:58:01Find the gif program
Pete Dunkelberg


is it here? [Major Geeks search on gif]



2012-02-02 21:59:39
John Mason


I agree with Kevin C above about editing out the bit on losing the GIF program but otherwise it's pretty clear and concise. Good stuff!

Cheers - John

2012-02-02 22:03:36
Pete Dunkelberg


I still like the idea of adding a contest to find the most / best escalators. Let 1000 escalators bloom.

2012-02-03 00:35:12Denial psychic
John Cook

To be honest, I don't think it's that impressive anticipating the attack on the escalator before the event. Seems a no-brainer when everyone was using it all over the place and Andy Dessler was the final straw, everything seemed to be coming to a head. it was only a matter of time before deniers realized the escalator was gaining traction and needed to be taken down a peg.

Did I really say "the next hockey stick"? Maybe we shouldn't overstate it and be accused of hubris. While I think the escalator is fantastic, powerful communication, the hockey stick was an unprecedented analysis, the first reconstruction of hemispheric temperature ever to be done, that took the world by storm and furthered our understanding of past climate. We might be overstepping our bounds in putting ourselves in such company. Perhaps qualify the language, rather than say "the next hockey stick", tone it down to something like "might start getting attacked like the hockey stick graph". Well, more eloquently than that.

Still think its worth saying that the uncertainty in the trend is much higher for the short cooling trends compared to the longer warming trend. Maybe a follow up post with the actual numbers? Anyone with some stats chops want to take that on?

2012-02-03 01:02:17
Kevin C


Alex C was asking how to calculate uncertainties on trends using Tamino's method a few days back, and the same would be required for doing an uncertainty graph for NCDC as Tom suggested above. I posted a summary of the method with refs here.

The uncertainty in a trend goes as 1/T3/2. That's a factor of T1/2 for increasing the number of observations, and T for the increasing 'leverage' as the period gets longer.

I can help with the stats, but don't have any inspiration at the moment. A while back I tried calculating the period over which you'd expect to see 95% significance in the warming trend for the major indices, and it's not far off 11 years. Which is inconvenient. But if the underlying forcings have changed (e.g. solar and aerosols), then the calculation becomes meaningless.

2012-02-03 03:03:07
Dikran Marsupial
Gavin Cawley

I suspect that Briggs may simply be asking for the uncertainty on the regression to be shown on graphs rather than simply the maximum likelihood regression line.  IMHO that wouldn't be a bad thing, and would be to our advantage if it were standard rather than the skeptics.  I suspect that Briggs has rather over-estimated the uncertainties.

Essentially there is nothing special about the maximum likelihood regression line.  There are a multitude of similar regression lines that explain the data nearly as well, and the "true" regression is almost as likely to be one of them than it is the maximum likelihood one.  If you compute the 95% envelope of those regression lines you get a quadratic error bars, which sort of say "there is a 95% chance that the true regression line lies in here somewhere" (or at least it would if it were a Bayesian analysis.

However Briggs' obfuscatory waffle is so opaque that it takes quite a lot of thought to realise that is what he is asking for (a picture would have made it obvious).  I have left a comment on his blog asking if that is what he actually wants (if he says "yes", I will suggest he make his point better by computing it!)

2012-02-03 03:09:30
Dikran Marsupial
Gavin Cawley

update Briggs says "I don’t know what the prediction uncertainty is for Plait’s picture. Neither does he. I’d be willing to bet it’s large enough so that we can’t tell with certainty greater than 90% whether temperatures in the 1940s were cooler than in the 2000s."

It was this phrase that game me a clue what Briggs was going on about with predictions.  The error bars on the regression are quadratic, so the further back or forward you go, the broader they get, so if you were to hindcast 1940s temperatures using the long term trend from the escalator (1970-onwards) the error bars would be huge, and hence you wouldn't be able to tell for sure if they were warmer or not.  However, the escalator isn't used to predict 1940s temperatures, we would use 1940s temperatre data for that, so the point he is making is specious.  The escalator isn't being used to make predictions about future temperatures either, climatologists use GCMs for that.  All the regression line is being used for is to estimate the long term trend, it isn't being used to predict anything,

It may be worth waiting to find out if this is what he is talking about to we can point out the error in his reasoning mode clearly.

2012-02-03 03:30:12
Dana Nuccitelli

Dang, that's a lot of comments.  I should put my draft posts in red every time! :-)

So let's see, I tweaked the discussion above the new escalator as suggested by Kevin and others.  I added tamino's graph and a discussion of the uncertainty in the trends.  I noted as Dikran pointed out that Briggs may have been treating our linear trend as a tool to hindcast past temps and predict future temps.  I took out the hockey stick reference.

I can try to make the new escalator look a bit better this evening, though I'd like to get the post up sometime today.  Let me know what y'all think about the revised version of the post.

2012-02-03 04:46:06
Tom Curtis


Given that Briggs tries to make a point about the 1940s, I did up this graphic comparing 1940's and 2000's temperatures:



As should be clear, I simply took 2008 minus the 95% error margin and compared it to 1944 plus the 95% error margin.  So within error,  there is no overlap between even the hottest year of the 1940s and the coolest year of the 2000s.  So much for his 90%.

I have primarilly done this for my own use in comments, but thought I'ld show it because its an effective filing system, and there is a slight chance it is useful (though I don't thinks so as it is unnecessary to the post, and not of sufficient quality as a graphic).

2012-02-03 05:19:41
Dana Nuccitelli

Cool Tom.  It's starting to sound like Briggs' 1940s comment was totally irrelevant (operating under the assumption that we would use the 1973-2010 trend to hindcast the temperature in the 1940s, or something bizarre like that).  But it would be worth showing the graphic in the comments to show the reality of the situation.

Frankly I suspect Briggs was just trying to come up with a statistical gobbledygook argument to confuse the issue and give the denialists a talking point against the temperature trend ("Briggs said uncertainty is large").  If so, he clearly succeeded, as Watts and Bishop Hill (and undoubtedly many others) bought into it hook, line, and sinker.

2012-02-03 05:31:46
Dikran Marsupial
Gavin Cawley

BTW, having a discussion with Briggs at his blog, I am not impressed by his behaviour.  He seems keen to make the statistics as little informed by climatology as possible, I wonder why


will have to go home soon, but I think he is just an obfuscator.

2012-02-03 06:21:43
John Mason


Dana, I'd ditch the John Cook, Denial Clairvoyant bit. Instead, replace it, I'd like to suggest, with "The reaction? All too predictable". Others may have a better phrase. But a typically "gawd - here we go aqgain" serves to convey the tedium peddled by this crowd.


Don't give the buggers anything they could at a later stage attempt to use against us.

Cheers - John

2012-02-03 06:41:30
Dana Nuccitelli

You mean just the section title or the discussion of John's prediction altogether, John M?  I think the discussion is useful to show how predictable these empty denier attacks have become, but I don't mind changing the section title.

2012-02-03 08:13:35
Brian Purdue


Now looks like not too much deniers can get their sticky fingers on and our friends who have published the first graph should be happy with the reasons given for the second one. I presume current graph hasn’t been changed yet because looks the same? Speed is still the same.

Is the first graph going to be left as the main escalator on SkS?

2012-02-03 08:27:07
Dana Nuccitelli

The new escalator hasn't changed yet - will tweak it when I get home this evening.  The Escalator link and right margin will stay as the original escalator.

2012-02-03 08:29:12
John Mason


Dana - the title. The discussion is worthwhile. I was just concerned about John C acquiring a title that might be used against him in future.

Cheers - John

2012-02-03 11:20:17


Dana1981 - the images aren't registered (jump a bit), and change resolution visibly. I think this distracts from the content considerably.

The circles at the data points should be removed. That would clean up the image considerably.

There are plenty of Mac GIF creators:


And even more Windows versions.

2012-02-03 11:31:30Late Suggestion
Brian Purdue


Change the post name to "Updating The Up Escalator". Has a ring to it and leaves no doubt about the message.

I concur with KR’s comment about the graph


2012-02-03 13:12:11
Tom Curtis


Link to article is currently broken.

2012-02-03 13:31:38
Dana Nuccitelli

The URL changed along with the title.  It's now published.