Saturday, January 10, 2009

Wisdom from USENET


USENET has an extremely deserved reputation as a refuge for nasty, cranky, crazy and porn, still there are prescient voices of wisdom and good information to be had. Eli would like to point out that the recently "discovered" idea that one has to take into account the asymmetry of possible outcomes from climate change is an old one to those who used to practice on such groups as sci.environment, and not a recent discover of Weitzman, newly taken up by Anthoff, Tol, Yohe and others. True this offers them an elegant out so they don't have to say Stern was right and we were wrong. Michael Tobis, yes, that guy, put it best:
-----------------------------------------------------------

Newsgroups: sci.environment
From: (Michael Tobis)
Date: 1999/06/23
Subject: Re: Problems with the "man made" global warming debate

My beliefs regarding the costs and benefits of climate change:
1) Global mean temperature is a mere proxy measure of change, and very large climate shifts with very small global mean temperature change are admissible, while the opposite case is meaningless. So focusing on mean temperature shifts already has a neutral to overly optimistic outcome.
2) As everyone is constantly pointing out, current understanding of climate as embodied in dynamic climate models is relatively crude. The result is indeed that the models are unreliable. However, contrary to what I have seen anywhere in the popular press, this means that the models are biased to produce results similar to contemporary climate using contemporary boundary conditions. They therefore tend to be undersensitive to perturbations. Again, this implies that the median modeling result may be an underestimate.
Another problem is that the real world has more degrees of freedom than any model, so modes not accounted for in models may be excited in reality. Again this causes model projections to be conservative regarding change estimates.
3) As has been stated recently, human institutions are built around prevailing conditions. Therefore any change occurring more rapidly than the optimum replacement time for these institutions has considerable costs. The example of building near sea level is the most obvious, though here it seems that the relevant time constants are long (several centuries rather than several decades). Regarding climate change itself, the dominant time constants are those of the atmosphere and the upper ocean, respectively a month or so (effectively instantaneous) and roughly 20 to 30 years. So the non-sea-level related changes are not imponderably distant in time.
4) It is clear enough that a change of 20 C would be cataclysmic, whether that change is a warming or a cooling. It is also clear enough that a change of 0.2 C is of little consequence, and may be on net beneficial. There is no reason to believe that this function is linear. The contemplated changes (~2 C) are large enough that we can not have total confidence that the impacts will not be catastrophic.
5) There is no plausible argument that any particular climate change will have a beneficial impact comparable to the worst plausible case negative impact.
6) The risk-weighted cost of unrestrained anthropogenic perturbation must therefore be dominated by the fact that the plausible worst cases have more cost than the plausible best cases have benefit.
7) I believe that resources should be dedicated not only to best models of current climate but to explorations of plausible worst case scenarios, precisely because these dominate the risk-weighted cost. This is the main way in which science can directly relate to the rational mitigation response strategy, if there's a sensible way to do it. I have heard of no efforts in this direction. Climate science as currently performed can provide much more of use to adaptation strategies, which are local, than to mitigation strategies, which are global.
8) The magnitude of the best estimate temperature climate sensitivity to anthropogenic inputs is not likely to change unless new global scale processes become active. The current global temperature sensitivity is well understood barring such changes, despite what you read in the press.
Point 2 above implies that at longer time scales, model changes are more likely to be underestimates than overestimates. For the foreseeable future the global picture is not likely to change.
9) Most of our knowledge points to the changes being larger than appear in nature. Meanwhile our ignorance prevents us from knowing at what level these changes become intolerable. The potential risk is enormous. The potential benefit is modest.
10) There is no *guarantee* that the climate shift will not be advantageous. There are reasons to suspect otherwise. There are *strong* reasons to suspect that *sufficiently rapid* changes are disadvantageous and it's perfectly obvious that *sufficiently large* changes are disadvantageous. The most important point, though, is that plausible negative outcomes have much larger costs than plausible positive outcomes have benefits.
11) The correct level and strategy of response are difficult to establish, and intelligent and vigorous discussion is warranted. Delay is not. The time constants of the human system that causes the perturbation and the time constants of the climate response don't offer any zone of comfort. Indeed it is difficult to eliminate the prospect of very serious consequences of the perturbation that has already taken place or will unavoidably occur before effective policies can be implemented.
12) The most likely time for catastrophic consequences not involving sea level change is in the next century. This is neither in the typical political time frame nor so far in the future that the matter can be treated as merely theoretical.
----------------------------------

OK, this is sort of cheating, but as far as Eli can see Michael has not put this up on Only in it for the Gold and it needs to be repeated again and again. We can negotiate if he wants to put it up on

14 comments:

Michael Tobis said...

Many thanks Eli.

My career is in such a marginal state that it in some sense hardly matters what I get cited for. But I did say these things long before Weitzman, publicly and often.

This particular effort that you quote seems a bit longwinded and stuffy. I think I was trying hard to be unambiguous. I hope my writing is a but more attractive nowadays. The effort at non-ambiguity may help establish that I was all over these ideas a decade ago.

I simply didn't think it was a deep enough idea to be worthy of submission for peer review. One of many tactical errors, I see. One of many lessons learned too late to be of any advantage.

Anyway, many thanks indeed for standing up for me on this matter and thanks for digging up a good example.

mt

Dave Bath said...

"USENET has an extremely deserved reputation...."

Not if you went down well-self-regulated groups such as "sci.*", "comp.*" (an even "alt.comp.*"). The nastier types were denizens of the "alt.sex.*" tree and left us alone. It was a place most of us trusted personal information. Opening it up outside mainly academia, government, and IT companies led to the much more hostile environment of the modern era.

(And we took our revenge on spammers by mail-bombing their accounts, and blacklisting them).

gravityloss said...

Everyone (that should be all engineers and most scientists at least) who has calculated an expected value E(f) of function f(x) where x has a probability distribution understands that you can't just automatically* use the average x (let's mark it as _x_) and calculate the function at that point f(_x_) and use that as the expected value.

i.e.
E(f(x)) is not always f(E(x))

Even if RPJr does it.

You have to integrate and weigh over all the probabilities:
E(f) = int(p(x)*f(x)) over the whole space of x.

It is interesting how such basic things are "news" for economists and policy analysts etc...

But Michael, your other things in that post show some interesting thoughts as well.

Like the asymmetry that there can be bad local effects even if global temperature doesn't rise, but there won't be local neutral effects if the global rises. Ie in a four field, only one corner is neutral and the rest are bad. (Or one corner is impossible, global change but local statism.)


Meanwhile, James Annan has some good discussion on why the very very small probabilities of huge effects should not weigh in too much on policy, ie they are a mathematical artefact. That the real climate sensitivity is a property of nature that exists, not something fundamentally unpredictable.

*: in some cases you can use the average x to calculate the average f(x), but often those assumptions don't hold.

carl said...

"6) The risk-weighted cost of unrestrained anthropogenic perturbation must therefore be dominated by the fact that the plausible worst cases have more cost than the plausible best cases have benefit."

By this logic, in a circumstance with a 99% outcome of +2 units, and a 1% outcome of -5 units, the correct policy would be to brace for the -10 outcome.

EliRabett said...

Check your math, even better, check your drink

Carl said...

Sorry, -10 is supposed to be -5. But you get my point. Gravityloss says it better than me.

Michael Tobis said...

Thanks, Eli.

Actually Carl has picked an actual nit, although I can't imagine where the -10 comes from. Had he accused me of concluding -5, it would be literally consistent with what I said.

(Is this Lydick's ghost? I thought he had left us lo these many years.)

Of course, in Carl's scenario, I conclude the utterly conventional 0.99 * +2 + 0.01 * -5 = +1.93; while
the quoted statement pretty clearly advocates -5.

Here's what I should have said. The costs are distributed in an asymmetric fashion over the probablities. The 1 sigma benign probably case does not, by now, benefit us at all, but even if it does it clearly doesn't match the damage due to the 1 sigma problematic case. And at 2 sigma benign, the benefits are hardly better at all, while the damages are even worse.

My argument does depend on some conclusions about the shape of the cost function which . It's not an extravagant claim, and while a case for it is built up in points 3, 4, and 5, I would agree that it remains implicit, and the argument would thus stand up better with a little more refinement.

Thanks for the correction.

Michael Tobis said...

Here's the thread:

http://preview.tinyurl.com/7qfq9l

No sign of Lydick on this thread (I believe he had already passed on), but some interesting replies from John McCarthy. It's been a long time since we've seen such cogent opposition.

I clarify further down:

"If you ask well-informed people not about their best estimate
but their 95 % confidence estimate of best and worst cases, I believe that you will find a severe asymmetry. 5 % is not such a small risk as to be
negligible. "

It's interesting to see how McCarthy and I came to a sort of agreement in the discussion. Those were the days... Hard to believe it was only 9 years ago.

Anonymous said...

The insurance industry has been using "expectation" forever to set rates.

it's common is science as well (especially physics).

And gamblers have been using it since Pacal.

So it is more than a little surprising that some consider it a "novel" idea (and even more surprising that many apparently are not even aware of it's existence)

Anonymous said...

Nice thought-provoking post, but those are only opinions, not facts.

Anonymous said...

Anon 9:31 said
"those are only opinions, not facts."

That's precisely the point.

"Expectation" does not deal with facts. It deals with the future -- ie, with potentialities.

Arguably, it is the best way to deal with uncertain risks/costs because it figures in the possibility of high cost/low probability events without knowing whether they will occur or not.

Sometimes "opinions" ("guesses", "hunches") are actually better than "facts".

James Annan said...

An anonymous up there says, the basic idea is not novel, the contribution of the economists is not the idea itself but rather the quantitative application to climate change (with numbers that stand up to peer review even if they are not unimpeachable).

TokyoTom said...

Eli, thanks for bringing this up.

I've done alot of commenting on Weitzman, BTW, including posting a working draft of his response to Nordhaus' criticisms.

Hank Roberts said...

nit:
broken link in the original post above should go to something about Tol and the others: http://sadielou.files.wordpress.com/2008/10/wp252.pdf