The kerfuffle over my intemperate denunciation of Chris Mooney refers. Check the comments.
As CM observes, "There have now been eight Category 5 Atlantic hurricanes in the past five years (Isabel, Ivan, Emily, Katrina, Rita, Wilma, Dean, Felix); There have been two Atlantic Category 5s so far this year; only three other seasons have had more than one (1960, 1961, 2005); There have been eight Category 5 Atlantic hurricanes so far in the 2000s; no other decade has had so many. The closest runner up is the 1960s with six (Donna, Ethel, Carla, Hattie, Beulah, Camille)."
I dislike the way CM is approaching discussing this, because I consider it misleading. However, I'm struggling to explain what I mean, so the below tries to discuss this more. Its something of a series of random jottings rather than a coherent post, which is my attempt at a caveat, but -cough- see below on that :-)
[Update: woo-hoo: Nature has picked up on this - and inadvertently provided a fine example of what I meant in my reply to this comment. Oh dear]
[Update: a good place to look for the numbers is
CM properly stuffs in lots of caveats into his post, but he you and I all know that people just blip over those. Putting in caveats is necessary when you're speculating, but it isn't a defence against leading writing. And I think his stuff is excessively leading. And vague. CM is clearly trying to talk around the issue of hurricanes and GW, and is pushing the recent numbers as exceptional, but doesn't seem to have formulated his questions in an answerable form. Or indeed, really formulated any questions at all.
I'm not familiar with the hurricane record, so I'm relying entirely on CM's data above. Lets try a few questions:
Q: Does having 2 cat 5's in a season imply a worrying trend towards more cat 5's? A: clearly not. 1961 disproves this.
Q: well how about 6 cat 5's in a decade? A: no, the 60's disprove this.
At which point you might step up and say "well wot about 8 in a decade? the 60's don't disprove that!" Which is true but cheating. You're only using 8 to exclude the 60's. And what about *landfalling* hurricanes? There are fewer of these so you just get noisier statistics.
Q: does the record of cat 5 hurricane numbers tell you anything useful, statistically? A: I don't know. But I suspect not, because if it did people would have published it and CM would be referring to it. CM has been in close touch with lots of hurricane folk recently so would be aware of this, if it existed.
Q: OK, so it doesn't say anything rigourous statistically, but come on look at those numbers! Surely its pretty suggestive A: Careful, you're sounding like the solar variation causes global warming folk; they can find you any number of "suggestive" correlations that aren't quite significant.
But... its also true that statistics aren't everything. In fact a mechanistic explanation is better. Consider, for example, a biased die. Suppose I know that a 6 has a 1/3 chance of showing up, and the others adjusted accordingly. Then I'd be happy to bet, on 1/6th odds, on the 6. But I wouldn't be very surprised if a 6 didn't turn up first, second or third go. Conversely, if given two die, one loaded as above and one fair, I wouldn't be confident of telling them apart by just rolling them and looking at the distribution without a lot of throws (cue JA to tell me how many, at 5%). And since we're on JA... you can probably do better that simple cat-5-#-counting by some clever Bayesian combining of different sorts of anomalies (that may be what CM is trying to do, in an informal way, by finding records in different aspects. But post-hoc record finding isn't good).
A mechanistic explanation is better, if you've got one. And indeed we do have the beginnings of one.
But what is CM actually saying? However, the more Category 5 storms we get in the 2000s -- which are hardly over yet -- the more this decade will appear anomalous when compared with previous ones. And at some point, it seems to me that people will simply have to throw up their hands and say: We are in a new place now.
Does this predict *more* cat 5 storms this decade? I can't tell. Is the current decade anomalous, at the moment? I can't tell. If yes, in what sense is it anomalous? Are we in a new place now? Presumably not, because the sentence is phrased in the future.
Lets try another approach. Prediction. Often regarded as the hallmark of science, even though observational astronomy still counts as science. Anyway. The assertion seems to be some connection between GW and h#'s, though clearly not a linear one.
Q: does this theory of a connection allow the prediction of h5#'s from SSTs? A: I'm not sure, but I doubt it, because no-one is trumpeting their predictions of 1 or 2 cat 5's this season (if people did make such predictions, please let me know).
Q: how about prediction h5#'s for the rest of the season? A: well funny you should ask that...
Because CM says In my opinion, we also have a right to be very worried about what else this hurricane season might bring... Felix is only our sixth storm of the year in the Atlantic. The forecasts suggest there will be 13-16. Could that conceivably mean yet another Category 5? Is this a prediction? No, except in the Piers Corbyns sense: if there aren't any more it will be forgotten; if there are it will be remembered. Yes, I know, thats a bit unfair because of course it wasn't clearly labelled as a prediction. But its something we need to be worried about. How worried? A bit? Not much?
So... how about a challenge:
Q: how many more cat 5's will there be this season, and/or in the rest of this decade?
A: (your turn now): 0, 1, 2, .. n? Or: at least 1, 2, ..; Or: at most 0, 1,..; Or: the available data and theory doesn't allow me to make a confident prediction?
- Log in to post comments
It's good that we can get the mortgage loans moreover, it opens up new possibilities.
This is just silly. Of course I cannot make a confident prediction of the number of category 5s this season. Forecasters predicted 3-5 major hurricanes this year, categories 3 to 5, in the Atlantic. So far we have seen 2.
[OK, thats fine, though I'll be disappointed if thats your only comment.
So you definitely think the evidence isn't good enough to confidently predict at least one more c5? There may or may not be, you make no prediction at all? -W]
I will predict there will be 1 more category 5 in the Atlantic, by 31 Dec 2009. But I'm not a scientist of any kind.
[I presume you mean "at least one more"? -W]
About statistics and category 5 hurricanes, see my comment to your previous blog entry (someone please check the code). In summary, the numbers would almost tell something if hurricanes were independent events. But they are not. And if one really wants to know about potential increase in hurricane power, there are better numbers to look at.
Janne, can you just clarify? Is your calculation showing that the number of cat 5 hurricanes in the current decade is bordering on being significantly more than those in the preceding four decades?
Sorry - my statistics aren't up to saying whether you're calculation is right or wrong, but it seems to me to be the right sort of calculation to make. Although as you point out, numbers of cat 5 hurricanes may not be the best measure - since they are rather rare.
I think it's pretty obvious what game Chris Mooney is playing. It's the same one Bush did with the WTC bombs, Al Quaeda, and Iraq. Irrespective of its validity, the association gets firmly embedded in the public consciousness, but if anyone calls him out on it, he can point out that he never actually said it...
OTOH it is quite clear that the whole concept of "detecting" changes is fundamentally misguided anyway (as I've posted about repeatedly).
Prediction:
Seven more hurricanes to make landfall in 2007.
Four more in the Gulf and three Florida/East Coast. At least two more Cat 5s and two Cat 4s.
This is an excellent learning experience.
Wm wonders what knowledge individuals can learn from the numbers Chris gives. Chris wonders what knowledge society can learn from the numbers he gives. IOW: Wm wants to inform scientific knowledge, Chris wants to inform decision-making knowledge.
All the rest is trivial.
Best,
D
> trivial
Or missing. I think Chris could set an example for science writing by _always_ including these words in every piece, as I noted in the other thread:
"... some connection ... is entirely possible. But is it possible to read anything meaningful into these statistics?" -WMC
And then answering that question. Every piece about science should mention that our knowledge is statistical, in science.
Annan:
WTC bombs? The bombing of the WTC occurred in 1993. It failed, and I don't recall then Governer of Texas Bush doing anything in response. I'll assume you mean the airliners which were crashed into the WTC towers in 2001.
It was known during the run-up to the Iraq war that Saddam and Al-Qaeda had a history of animosity. In particular, Al-Qaeda representatives visited Saddam's Kurdish rivals several times. Bush was careful to never mention this. There is no similarily strong reason to reject an association between category 5 hurricanes and global warming. The closest is the association between more frequent El Ninos and AGW, but it has not been shown that this in turn results in fewer Atlantic category 5s. In particular, the proposed mechanism is shear, and there is so far no upward trend in shear over the Atlantic. In any case - Mooney has many mentioned the association between AGW and stronger / more frequent El Ninos as a reason to be skeptical about predictions that AGW will result in more hurricanes. Finally, the invasion of Iraq was also strongly pushed by the claim that Saddam had weapons of mass destruction, about which Bush said:
Mooney did no equivalent (such as claiming a Category 5 hurricane could bring 160 mph winds to Denver, Colorado) .
"There is no similarily strong reason to reject an association between category 5 hurricanes and global warming"
Indeed, it is of course inevitable that there is an association, which is why null hypothesis significance testing is such an intrinsically vacuous (and in practice frequently misleading) activity. Nevertheless, none of Mooney's wide-eyed hysteria actually does anything to illuminate such association. I think this is also William's main point.
Ian: yes, you got it right, it's borderline significant at p=0.011. But there is a caveat: The test assumes independent events, while hurricane occurrence probably comes in chunks because there are slowly-varying environmental variables that affect the birth and intensification of the storms. (Technically, the Poisson rate parameter varies.) Dependence between hurricanes can only lead to the direction that the test becomes less significant. p=0.011 is therefore a kind of lower limit from this data.
In general, a more serious problem may be that even p=0.01 gives you one false positive in a set of 100 cases with no real effect. Therefore, mostly spurious changes are found if we test everything in sight. This becomes tricky, for obviously we want to detect early changes in important processes such as hurricanes, Greenland ice, etc., and there are quite many of these "important" parameters, and ways to quantify them.
On the other hand, I'm all for action against climate change because it may turn serious. All feedbacks and our sensitivity to changes are not well understood. And biodiversity will suffer (sorry, decrease) anyway, although most do not want to put much price on it.
In a democracy, action does not follow unless there is a concrete threat to deliver. p-values are certainly not threatening enough. Probably nothing short of water or hurricane on your backyard is.
Information about harmfulness of smoking has been around for decades, and the papers, doctors, and the packages all say it kills. But people are still smoking. We are biased by our motives and beliefs, discount far future very strongly, and have problems on taking abstract stuff seriously.
James, I don't see much point in calculating anything out of this data (category 5 counts). Anyway, if the effect was significant even in such a ridiculous data, it would likely be large. I think the critique of significance testing as you presented it applies better to cases where the effect is negligible in practice.
My criticism of NHST is that if no-one thinks the NH is plausible in the first place (let alone likely) then we have in principle learnt precisely nothing by "rejecting" it. And note that according to the frequentist approach on which this whole line of analysis is based, we cannot even say that we have shown that the null hypothesis is (very) likely false, only the substantially different statement that the observed data are very unlikely if it is true. Of course that doesn't stop everyone mixing up those statements in practice.
...but I'm digressing a bit from the main point.
James, I agree with you about NHST (null hypothesis significance testing). It is widely misused. Often confidence intervals would reveal the whole misery, which I guess is a common reason to not use them. :)
[For my toy calculation around here, the 98% confidence interval of rate intensification multiplier is about 1.0-8.3, with posterior median at 3.0.]
Since Chris made no formal claims I wonder what the fuss is about. We look at what is around us and try to perceive patterns. Is there a pattern of increase in hurricane intensity? That's agreed. How severe is it? Well, the time series of Cat 5 storms per decade is one meaasure of that.
Irene and I talk about this in the context of psychology, where there is a distinction between 'data' and 'lore' (yes, data's evil twin!) Sometimes there is no data, but you need to come up with a treatment plan anyway, so you have to use the lore.
I am also willing to bet that Chris is a baseball fan. Americans love to play with what they call "statistics" which are very different from what James means! That doesn't mean there should be some sort of law against this sort of speculation.
The comparison to the 9/11 - Saddam association is unfair both as a near-violation of Godwin (it's probably not conducive to rational thought to bring up an occasion where emotions run high as well as reason being abandoned) and as an analogy (there really is a widely accepted prior belief that high SST is conducive to tropical storm formation).
I think we are all agreed that the SSTs are increasing, the storm intensities are increasing, and the observations are consistent with a causal connection between the one and the other. So I don't see the basis for the critique.
Every piece about science should mention that our knowledge is statistical, in science.
and
I think we are all agreed that the SSTs are increasing, the storm intensities are increasing, and the observations are consistent with a causal connection between the one and the other.
Again, I think this discussion arises from POV. If you're motivating scientific inquiry, you use Wm's caution. If you're motivating reducing building near the shore, you use Chris' framing.
Both points are valid. The two comments I italicized -- both from commenters I respect -- are perfectly reasonable in my view, and are relevant for different situations and different scales (putting my ecology hat on).
But the real point is: if a decision-maker asks me for a brief, I'm going to end it with: insurance companies don't want to write new policies near the shore. Why do you think that is?
Best,
D
[I disagree that all our knowledge is statistical. Andinsurance companies should want to write policies for appropriate premiums. If they don't want to, they are silly -W]
As with anything else for NHST you have to know a lot to set a useful NH.
William,
FWIW, back when Webster et al. first came out, we took the opportunity to extend Webster's analysis for the North Atlantic back to 1945 so you could get a better look at the longer term (note the cat4+5 rise in the 1960s).
-Chip
[I disagree that all our knowledge is statistical. Andinsurance companies should want to write policies for appropriate premiums. If they don't want to, they are silly -W]
We didn't say all, Wm. But science is about chance and predictability so statistics often helps us express scientific findings.
Nonetheless, my overarching point - now less delicately put - is that you are approaching this from a certain point of view, convinced your view is the best. It may be, but not everyone shares it, hence my juxtaposing the two italicized comments.
Aside: I just made comments on a grant proposal for a green infrastructure modeling project. I suggested a few things to say to get out of academic mode to be more useful for decision-makers, as the author is an academic and I'm applied on the ground. Her view was great in academia, but not quite good enough for those of us who don't choose academia and instead practice out here.
Ahem.
Best,
D
[I'm definitely approaching it from my pov. But no, I'm not *convinced* its best - as I thought I'd made clear, this is all unclear. I'm not sure what you're trying to say - please expand -W]
But no, I'm not *convinced* its best - as I thought I'd made clear, this is all unclear. I'm not sure what you're trying to say - please expand -W]
Apologies for the half comments. Lots of deadlines making the time to do a good job go away.
It takes a long time to turn society's boat. Chris is interested in getting the societal conversation started on which direction (or whether) to turn.
I read comment threads like this and they read to me like not wanting to start the discussion without first having fantastic knowledge. That's not the way it works, #1, and #2 the unclarity is part of the discussion and shouldn't preclude it starting.
That is: sure there's uncertainty and unclarity but they are minor course corrections to, say, 125 when you know you are heading somewhere between 120 and 130, then when you see the rocks, you know you have to turn to a nor'easterly heading. The uncertainty doesn't mean you are stayed for while the wind sits in the shoulder of your sail and the tide's in.
Lastly, this sort of thing is exactly what scenario analysis is for, and IMHO the IPCC is not doing a good job explaining how to manage societal turns while using their scenarios.
Best,
D
[I'm still unconvinced. Unclarirty is indeed part of the discussion; adding to that unclarity isn't good. Of course simply repeating the bit that *is* clear - that the vast majority of the increase in damage is due to our building habits, not hurricane change - is (a) boring; (b) playing second fiddle to RP Jr and (c) unsexy. Who wants to tell people not to live on the beach? So we'll continue to get over-excited about variations in hurricane numbers that we struggle to resolve from noise -W]