The significance of 5 year trends

Posted by William M. Connolley on May 17, 2007
(10)
More »

5 year trends from surface temperature are not very significant and are a bad measure of anything. As everyone should know. But it seems that some people don’t. So in tedious detail…

Pick up the HadCRU temperature series from here. Compute 5, 10 and 15 year trends running along the data since 1970 and get (black lines data, thicker black same but smoothed, thin straight lines non-sig trends; thick straight blue lines sig trends):

i-f52290941a698ea6e0aa2af09525f70d-5-year-trends.png

From which you can see (I hope) that the series is definitely going up; that 15 year trends are pretty well all sig and all about the same; that about 1/2 the 10 year trends are sig; and that very few of the 5 year trends are sig.

From which the motto is: 5 year trends are not useful with this level of natural variability. They tell you nothing about the long-term change.

(10)
More »

Comments

  1. #1 uBeR
    2007/05/17

    What about 9 years? ;-)

  2. #2 SCM
    2007/05/17

    It perturbs me when supposedly respected scientists make statistically unsound claims like Pielke did about these 5 years series when they really should know better. The “It has been cooling since ’98″ meme is another example propagated by Lindzen & Bob Carter among others.

    You don’t even need stats to see the claims are unsound – its pretty obvious just from eyeballing the relevant graphs.

    What is going on here? Have they really forgotten all they learned in Stats 101 or are they playing fast and loose with data to push an agenda or preserve their ego? It surprises me because it can only damage the credibility as scientists – but perhaps scientists aren’t the peer group they are trying to impress.

  3. #3 Hank Roberts
    2007/05/17

    Thank you.

    Making a picture of it really does help convey what the numbers mean.

  5. #5 Hank Roberts
    2008/02/12

    The bloggers against the idea of warming have a new statistician they’re starting to point to:
    http://wmbriggs.com/blog/2008/01/27/best-statistical-scientific-talk-on-global-warming/

    You’ll be glad to know he says he’s a Bayesian.

  6. #6 Hank Roberts
    2009/05/31

    cherrypicking from trend lines, splendidly illustrated:

    http://www.skepticalscience.com/global-warming-stopped-in-1998.htm

  7. #7 Hank Roberts
    2009/05/31

    And the video, wonderfully illustrated and narrated:
    http://www.youtube.com/watch?v=y15UGhhRd6M
    Climate Denial Crock of the Week – Party like it’s 1998

    Hat tip for both to ‘Perps’ at bravenewclimate (Barry Brooks):

    http://bravenewclimate.com/2009/05/22/al-gore-on-nuclear-power/#comment-15285

    (PS, the Atmoz thread is locked, and the ‘comment link at the top of the page’ is beyond my limited capacity to discern. This will have to suffice as an xref.

  8. #8 Hank Roberts
    2010/04/14

    belatedly:

    see also:
    http://hot-topic.co.nz/keep-out-of-the-kitchen/
    (page down for the Javascript slider thingy)

  9. #9 Hank Roberts
    2010/05/21

    A lovely example of picking a timespan, as featured at a climate change obfuscation event:
    http://scienceblogs.com/deltoid/2010/05/don_easterbrook_hides_the_incl.php?

  10. #10 Brian Schmidt
    2010/05/21

    Significant as in 95% significant? Certainly important for scientific reasons, but in a potential bet I’m willing to put down money even when the odds are less than 95% in my favor. Still, five year periods look pretty dicey. Someone’s trying to get me into a five year bet, but I think I’ll need a longer period.