Explaining the Flaw in Kevin Drum's (and Apparently Science Magazine's) Climate Chart

I won't repeat the analysis, you need to see it here.  Here is the chart in question:


My argument is that the smoothing and relatively low sampling intervals in the early data very likely mask variations similar to what we are seeing in the last 100 years -- ie they greatly exaggerate the smoothness of history and create a false impression that recent temperature changes are unprecedented (also the grey range bands are self-evidently garbage, but that is another story).

Drum's response was that "it was published in Science."  Apparently, this sort of appeal to authority is what passes for data analysis in the climate world.

Well, maybe I did not explain the issue well.  So I found a political analysis that may help Kevin Drum see the problem.  This is from an actual blog post by Dave Manuel (this seems to be such a common data analysis fallacy that I found an example on the first page of my first Google search).  It is an analysis of average GDP growth by President.  I don't know this Dave Manuel guy and can't comment on the data quality, but let's assume the data is correct for a moment.  Quoting from his post:

Here are the individual performances of each president since 1948:

1948-1952 (Harry S. Truman, Democrat), +4.82%

1953-1960 (Dwight D. Eisenhower, Republican), +3%

1961-1964 (John F. Kennedy / Lyndon B. Johnson, Democrat), +4.65%

1965-1968 (Lyndon B. Johnson, Democrat), +5.05%

1969-1972 (Richard Nixon, Republican), +3%

1973-1976 (Richard Nixon / Gerald Ford, Republican), +2.6%

1977-1980 (Jimmy Carter, Democrat), +3.25%

1981-1988 (Ronald Reagan, Republican), 3.4%

1989-1992 (George H. W. Bush, Republican), 2.17%

1993-2000 (Bill Clinton, Democrat), 3.88%

2001-2008 (George W. Bush, Republican), +2.09%

2009 (Barack Obama, Democrat), -2.6%

Let's put this data in a chart:

click to enlarge


Look, a hockey stick , right?   Obama is the worst, right?

In fact there is a big problem with this analysis, even if the data is correct.  And I bet Kevin Drum can get it right away, even though it is the exact same problem as on his climate chart.

The problem is that a single year of Obama's is compared to four or eight years for other presidents.  These earlier presidents may well have had individual down economic years - in fact, Reagan's first year was almost certainly a down year for GDP.  But that kind of volatility is masked because the data points for the other presidents represent much more time, effectively smoothing variability.

Now, this chart has a difference in sampling frequency of 4-8x between the previous presidents and Obama.  This made a huge difference here, but it is a trivial difference compared to the 1 million times greater sampling frequency of modern temperature data vs. historical data obtained by looking at proxies (such as ice cores and tree rings).  And, unlike this chart, the method of sampling is very different across time with temperature - thermometers today are far more reliable and linear measurement devices than trees or ice.  In our GDP example, this problem roughly equates to trying to compare the GDP under Obama (with all the economic data we collate today) to, say, the economic growth rate under Henry the VIII.  Or perhaps under Ramses II.   If I showed that GDP growth in a single month under Obama was less than the average over 66 years under Ramses II, and tried to draw some conclusion from that, I think someone might challenge my analysis.  Unless of course it appears in Science, then it must be beyond question.


  1. RambleOnDude:

    I apologize for repeating myself, but there is no need to "refute" this graph, it's been handled. Well. Re-posting a few links:


    Instapundit drive by.

  2. marque2:

    I was thinking the same thing about lack of data points when I first saw the chart yesterday.
    But a second reason is that NOAA and NASA et al, are monkeying with the ground temperature data. If you look at temperature graphs from 1999 and current you will see that the temps in the later years of the 20th century were increased by about 1/3 degC and in the early part reduced by abuot 1/3degC I would guess that about 1/2 that rise, at the end, is due to monkeying with the data.

  3. Douglas2:

    Although I am aware of the excellent work by Steve McIntyre and Prof. Pielke, I don't read their blogs daily because I find the writing style dry and boring. Nor do I read the opinion columns of the Financial Post. But I do read Coyoteblog.
    As a teacher, I find it useful to watch the lectures of other people teaching the same stuff, because different people have different ways of explaining things. I find it really helpful if a student comes to me and says "I don't understand" to have a pool of different examples and illustrations.
    I think this illustration is a good one.

  4. Another_Brian:

    Perhaps an even better way to demonstrate the flaw would be to average everything from 1948-2000 (3.582%), then 2000-2008 under GWB (2.09%), and finally the single year of Obama (-2.6%).

    If you could find figures for GDP from the founding of the country, you could further average everything from 1789-1900, 1900-2008, and the data point for 2009 and offer the same analysis. Look how steady our GDP rose from 1789-1900, how it plateaued from 1900 to 2008, and then Obama came along and tanked 230 years of progress in a single year.

  5. obloodyhell:

    }}}} Drum's response was that "it was published in Science."

    Drum makes me think of ducks.

    Quack. Quack. Quack.

  6. obloodyhell:

    }}} I would guess that about 1/2 that rise, at the end, is due to monkeying with the data.

    Or getting unrequested anal sex from James Hansen and/or Michael Mann, more accurately.

  7. HenryBowman419:

    One should always keep in mind that, at least since Phil Abelson was the editor, Science (that is, Science Magazine) has been an oxymoron.

  8. johncunningham:

    I have yet to see any warmalist explain why they have gone back into temp records from 1900-1940 and lowered them by about 2 degrees C. why the lowering? and how was the 2 degree figure arrived at? actual temp records from stations that have not been moved in rural areas in the US show no pattern of warming.

  9. Don:

    Dammit, Warren! You're not supposed to look behind the curtain! Didn't you get the memo?

  10. RightThinking1:

    Wolfe Simon's famous research regarding bacteria that substituted arsenic for phosphorus was published in 'Science' also. Her work was supported by NASA, and her findings were touted with a huge roll-out by NASA. The problem? It simply wasn't true. It was a matter (among other things) of someone hoping for an outcome, and doing sloppy work.

    All 'peer reviewed'. I would like to link to the NASA roll-out videos and interviews, but, scrubbed.

  11. Harry:

    QED. Merry Christmas to Coyote and his pack,

  12. Zachriel:

    Coyote Blog: My argument is that the smoothing and relatively low sampling intervals in the early data very likely mask variations similar to what we are seeing in the last 100 years

    If you are referring to Marcott et al. (2013), then they use Monte Carlo techniques to help determine the range of plausible short-term variability. There are important limitations to this technique, but they were still able to conclude that temperatures are higher now than at any point in the last 5000 years.

  13. Zachriel:

    The McKitrick paper is off the mark.

    a remarkable admission: “[The] 20th-century portion of our
    paleotemperature stack is not statistically robust, cannot be considered
    representative of glbal temperature changes, and therefore is not the
    basis of any of our conclusions.”

    Now you tell us! The 20th-century uptick was the focus of
    worldwide media attention during which the authors made very strong
    claims about the implications of their findings regarding 20th-century
    warming. Yet at no point did they mention the fact that the 20th
    century portion of their proxy reconstruction is garbage.

    The resolution of most proxies are not valid in the near term. Uncertainties and limitations can be found in the individual papers from which the proxies were pulled, and are often discussed in Marcott et al. (2013).

    Haven't read the other papers in detail, but they seem to make the same error.

  14. Joel McDade:

    Warren, if you look back at McIntyre's posts from March-April you will see that the problem really isn't about time resolution per se (though nothing can be resolved under about 300 years -- as you say). It was about Marcott's (sp?) redating of samples and about the drop-off in the number of samples in the modern period. The two combined created the modern spike (and they subtly, if not sufficiently, retracted this in an RC FAQ stating that it was not statistically robust, lol).

    Most interesting, there was no spike in his original thesis (presumably a doctoral dissertation), yet it appeared plain as day in the article published in Science. This hints at an obvious cherry picking scam. So in a sense it is worse than you claim, since it involves a statistical sleight of hand that most would not see.

    It also had nothing to do with the instrumental record. In the graphs shown above, the instrumental record only appears in the smooth of Mann's reconstruction. This is a photoshop. I will never defend Mann, but at least his original splice of the instrumental record appeared in a red color to distinguish it from his tree ring data. There is no such distinction above.

  15. deano32:

    Computer models are always wrong; however, some are useful. This is a common phrase used by people that simulate everything from semiconductor physics to photon behavior inside sub-wavelength structures. Monte Carlo techniques involve running numerous computer simulations around possible combinations, and then using the results to generate some statistic inference on the output. If your computer model is wrong, or flawed, or incomplete, then Monte Carlo results are flawed as well.