Uncertainty and Conservatism

In a comment by Onar Aam on RC, it is alleged that proposed policy responses to anthropogenic climate change are excessive because scientific uncertainty leaves open the possibility that the sensitivity of the system is much smaller than the consensus would have it. This argument is common enough.

For almost fifteen years now I have been (using my unfortunately trivial influence; though for some reason Fergus seems to be singlehandedly trying to change that; thanks Fergus!) pointing out that such an argument is totally wrong, pretty much exactly 180 degrees off the mark.

Here’s my response, verbatim, which you can also read on RC.

Suppose we grant for the sake of argument that the total range of uncertainty (of some quantity) is a factor of 100. Does it follow that the quantity is possibly overestimated by a factor of 100? Perhaps, but surely it follows no more and no less than it follows that there is an equivalent possibility that the quantity is being underestimated by a factor of 100.

Why are people constantly harping on the risk of overestimating climate change when the risk of underestimating it has vastly greater consequences?

Rational policy under uncertainty should be risk-weighted, which implies that the less faith one has in the consensus position, the more vigorous an emissions policy one should support. It is very peculiar and striking to observe how common a position like Aam’s is despite the fact that it is incoherent.

Those people who doubt the consensus in a rational way (e.g., Broecker, Lovelock) advocate for a very vigorous policy. We don’t know how bad it can be, so we really ought to give considerable weight to it being very very bad. The asymmetry arises because we know how good it can be. Climate change can at best amount to a (relatively) very small net gain, if it is modest and slow enough. At worst it can quite conceivably be a threat to civilization.

Most people stressing the uncertainty, though, seem to me to deliberately strive to confuse the policy process, or to echo others who do so. It is discouraging how effective this tactic continues to be, given that it is based on a completely irrational argument. The only remotely sensible way to argue for small or no policy response is not to argue for large uncertainty. A rational argument for policy inaction requires arguing that the consensus position is certainly wrong and oversensitive. A rational, conservative response to uncertainty would be to take more effort to avoid the risk.

My use of the word “conservative” in the concluding sentence is deliberate, of course.

I always find it a tortured use of the word “conservatism” to suggest that monkeying with the biosphere (an astonishing and rare natural phenomenon) is a better idea than tuning the economy (an artifact). I can anticipate the tedious answers of course (cue Mr Duff), but I find myself wondering what, exactly, these so-called conservative people think they are conserving.

Annan vs Hegerl goes Nuclear in a Hurry

Strange day.

I just returned from a day that included a talk by Gabrielle Hegerl to discover that James Annan has a poster criticizing an aspect of her work.

In fact, the point that James makes, was, in essence, covered in her talk. She said the data alone was insufficient to remove the long tail, but she was clear that the long tail was almost certainly an artifact of
the way the problem was set up. I believe that James’ approach amounts to begging the question: if you assert a prior where there are no tails, you can’t say the data has constrained the tail. This seems to me like hair-splitting, though I expect James will disagree.

OK, fine, here I am at Bayesian central, I suppose seeing some technical battles between my virtual friends and my real-world acquaintances is to be expected in the rough and tumble world of science, but it’s a bit of an odd coincidence.

So to see if there is some ongoing longer feud between these two, I googled “Annan Hegerl”, and who should pop up but Lubos Motl! My God, this article was uploaded tomorrow (dateline artifacts). It appears negativeland is already all abuzz with the fact that a climate scientist would dare to criticize another!

(We are damned if we criticize each other and dmaned if we don’t, but still, this is ridiculous.)

And see this on the SEPP site (with similar comments from Motl):

Three weeks ago, Hegerl et al. published a text in Nature that claims that the 95 percent confidence interval for the climate sensitivity is between 1.5 and 6.2 Celsius degrees. James Annan decided to publish a reply (with J.C. Hargreaves). As you might know, James Annan – who likes to gamble and to make bets about global warming – is

an alarmist who believes all kinds of crazy things about the dangerous global warming;

a weird advocate of the Bayesian probabilistic reasoning”.

However, he decided to publish a reply that

the actual sensitivity is about 5 times smaller than the Hegerl et al. upper bound which means that the warming from the carbon dioxide won’t be too interesting;

Now, leaving aside how weird it is for anyone, let alone the ever-rigorous James, to be called “a weird advocate of the Bayesian probabilistic reasoning” (!!!!) doesn’t this read like weird Bayesian James has become almost Lindzenite, putting an upper bound on the sensitivity at 6.2/5 or about 1.25 C?

No. Weird Bayesian James advocates using 20/5 or 4 C as an upper bound. Elsewhere he comes up with 3 C as a best estimate. Which is higher than Gabi’s!!!

I assure you Hegerl knows the long tail is an artifact; that the higher the cutoff the longer the tail; that it tells us nothing.

The only question is how to handle it. It can easily be misinterpreted either way you approach it. James’ approach can be seen as begging the question. Can you really put your hypothesis in as your prior?

And how (to get back to my blog theme) do we go about explaining any of this to the general public now that the engines of apoplectic confusion are gearing up to make a case out of this?

Note to the lay reader; James will correct me if I am wrong, but I am pretty sure he does not claim that the sensitivity is certain to be less than 1.3 C.

Anyone care to make a bet?

[Update: apparently all this denialist noise was from last year. Still…]