"Global cooling" hoo-hah doesn't stand up to analysis

[jducoeur]

Here's a good little article in Ars Technica, taking a look at the increasingly-common rumors that have been spreading about "global cooling" -- basically, the way that a lot of people with a vested interest in fighting the idea of global warming have seized on short-term statistics and ignored the long-term ones. Apparently the AP did a simple little experiment of sending the data, context-free, to some statisticians and asking what trends were seen in the numbers: the result (which won't surprise most people here) is that the pattern of warming is quite clear, and the "cooling" is nothing more than statistical blips and preconceptions...

Comments

[antoniseb.livejournal.com]

1998 was a very hot year.

I think it is also the case that we may experience a small reprieve from global warming on account of the strange reduction in solar magnetic activity, resulting in more cosmic rays (and hence high altitude clouds) in the upper atmosphere.

This is NOT going to fix the melting glaciers, but it will give a few consecutive non-record heat years.

[herooftheage.livejournal.com]

What Anton said. I guess there's also the possibility that the sun's about to enter/has entered a cooling cycle, but the data about that is not very clear, and the folks doing that modelling are about on a par with the effects of global warming people - i.e. still not very good.

I found it to be a very poor article. a) I couldn't actually find a link to the statisticians questioned or their methodologies. b) somewhere in there, I'd think one should reasonably talk about Bayesian reasoning, and what sort of weight you are supposed to give the most recent evidence. You'll notice the trick they used to compare the recent data to statistical variation in the past has a glaring fault to it: if it does represent a change in the trend, the fact that it looks like what turned out to be statistical variation in the past is a red herring.

Really, what a good Baysesian reasoner would say is that the recent data would lower the case for global warming by a bit. Not like the official reports that claim that the decline in the last few years is evidence in favor of the theory (if you need the cites for that, chase down David Friedman's blog, where he discusses it extensively.)

[umbran.livejournal.com]

Bayesian or otherwise - a reporter would not say the recent data lowers the case for global warming if the recent data matched previous variations, or was within expected random variation in the sample. If the current cooling blip is of the same duration and magnitude as all the other variations, it does not impact the case in general.

[herooftheage.livejournal.com]

If the current cooling blip is of the same duration and magnitude as all the other variations, it does not impact the case in general.

The thing that makes this notion incorrect is that you don't know that the current case has the same duration and magnitude as the other cases - you only know whether it has that so far. One of the important bits about Bayes Theorem is to get some tools to help figure out the probability that new data weakens or strengthens the probability that the current model is correct.

[umbran.livejournal.com]

...you only know whether it has that so far.

Correct. The true answer of a scientist is, "the current data does not yet support that conclusion". The strict scientist does not speculate on what the future data may or may not be. That would be assuming the conclusion.

I also think that we are talking about slightly different questions. The "case for global warming" is to me a case about a trend in the past - has the average temperature been rising. Even if this cooling trend continues, it does not alter the fact that the past trend was for warming. Nor would it mean that the past warming trend was withing normal variations.