Sunday, April 5, 2009

"It Hasn't Warmed Since 1998"

Although it's been shown to be unsupported by the data, the claim that global temperatures have declined since 1998 continues to be high on the climate skeptics' hit parade. This canard reared its disreputable head again last week in George Will's latest bag of climate lies.

Even if this discredited claim were true, it still wouldn't mean anything by itself, since it's a classic case of cherry-picking. Andrew Revkin's April 3 post at the NYT's DotEarth blog picks up on that theme with a discussion of a paper to be published in the American Geophysical Union's peer-reviewed journal Geophysical Research Letters. Despite the clear evidence that out-of-context short-term trends are meaningless at best in the presence of random natural variations, the comments to the post contain many of the typical knee-jerk reactions from the Usual Suspects.

John Mashey, however, suggests a little experiment with the Swiss Army Knife of Data Analysis, Microsoft Excel. The graph at the top of this post shows an example result using Mashey's recipe. The blue line shows a series of 30 points with an overall upward trend (black line). The red line, however, shows the trend of the first 10 points alone. Notice that this trend is downward, and in fact, the value of point 10 is actually less than the value of point 1. For an animated display of a series of such randomly fluctuating trends, click here. In the example, the short-term trend may be much above, much below, or nearly equal to the longer trend, but the overall trend is always upward.

Image: Trend with random fluctuations based on DotEarth comment by John Mashey; graph created using Microsoft Excel and animation from


John Mashey said...


chrisd said...

Nicely done.

Stevo said...

It's a good experiment. Extend it a little.

Define a sequence T(i) by setting T(0)=0 and T(i+1)=0.99*T(i)+Random. (Your random numbers should have a zero-centred Normal distribution. You can do that in Excel with NORMSINV(RAND()).) Extend it for a few thousand points, and plot it.

Over short periods, you get apparent trends up and trends down, but these are all spurious - there is no trend, as can be seen if the series is extended far enough. It's an experiment everyone should do.

How could you tell, just looking at the data, whether any apparent trend was genuine, or the result of this sort of randomness?

Seasonal Outlook

Latest seasonal forecast: Click here.

Latest 3-month temperature outlook from Climate Prediction Center/NWS/NOAA.