David Appell writes in Scientific American of a number of recent errors in climate science data identified by bloggers, and how, though largely trivial, they are undermining the faith in AGW.

While any error in science is important, and those identified should be swiftly corrected, my concern has always been non-trivial errors of statistics. My beef is that large tranches of AGW orthodoxy are supported by claims that do not pass standard tests of significance. Why is significance important? Wikipedia states that a result is called statistically significant if it is unlikely to have occurred by chance. Therefore many of the claims may be simple change occurrences that scientists are being fooled into believing by their own prejudices. For example:

  1. I called Rahmstorf on the claim that ‘the climate is more sensitive than we thought’, based on his untested, mangled and misunderstood methodology before he subsequently recanted.
  2. I have forthcoming submissions showing that sea level is not accelerating – its not, the models are non-significant in the acceleration terms, and based on the most recent satellite altimetry, sea level is significantly decelerating.
  3. I have shown that claim that current temperature levels are ‘unprecedented in the last 1000 years’, is based on faulty calculation of confidence limits that fails to account for selection bias in the proxies.
  4. I have submitted an article showing that water vapour feedback is strongly positive is based on a statistical methodology with an infinite variance, that is, it has no inherent confidence.
  5. The claim that droughts are going to increase enormously is based on projections of unvalidated climate models at variance with actual recorded significant decreases in droughts in the last 100 years.

The list goes on.

What possesses usually rational people to believe in things that fly in the face of standard testing of claims they learn in under-graduate school (one hopes)? To some extent the answer is; the claims are consistent with expectations based on the projections of the climate models. But the climate models are largely used without validation, even though it is acknowledged they have large errors in many areas, particularly precipitation. Either because the phenomenon is so subtle that it can’t be seen yet, or the models are wrong, there is a troubling disagreement between the models and reality across a range of phenomena.

And for one reason or another, climate scientists choose to believe their models, and forgo basic testing against reality. Many global warming papers do not report significance tests because if they did, they would show that there is no statistical significance to the phenomena they report on.

The general public are easily able to appreciate trivial errors in data, but less able to appreciate significant errors in statistics. Nevertheless, the errors in the data draw attention to the fact that climate science is not the squeaky clean paragon of science it is claimed to be.

I don’t expect everyone to be perfect. But if a paper contains no statistical tests to show the result is not a chance event, I assume it is a chance event and reject it. And every so-called scientist should do the same.

Read on in the Wiki article and find that tests of statistical significance are not perfect, but they are much more preferable to no tests at all, which seems to be the norm in climate science.