Let me preface this with the following: I think anthropogenic climate change is happening, and we need to do something to limit our effect on the environment.
As you might tell from this, this post is going to involve climate change. As with some of my other posts though, it's going to be of a statistical bent.
The motivation for this one is last night's Question Time. Leaving aside the current media mess over this, and the vile Melanie Phillips and her grossly incorrect views, one of the members of the audience referred to the events of Cockermouth and that as evidence of global warming.
This, I think, is a bad thing to do. Its something people do a lot of - they blame extremes of weather on global warming. Now, I have no reason to doubt that extremes of weather are not affected by global warming, but I am quite confident that people can't look at local extremes of weather and just go "Oh, that's due to climate change".
Extremes happen. They happen in lots of situations - the vast majority of distributions of any sort of value, whether climatic, economic, astronomical, biological or whatever, tend to have most samples drawn from them sitting somewhere near an average, with less common values out on the extremes, and really far out very extreme values.
The study of how to analyse these is quite well-developed as extreme value theory, and for a statistics geek I know very little about it. Its an interesting subject to go Googling on though - it turns out for example that it came about from someone studying the strength of cotton. The strength of a thread depends on extreme values. A chain is only as strong as the weakest link, so to study the strength of something like that you need to understand how extreme values are drawn from a wider sample (a thread is a bit more complex than a chain in that sense, but you get the idea).
I can reasonably comfortably make the following points though.
First off, extreme values are, as said above, uncommon. So looking at extreme values in order to judge the behaviour of the wider distribution is going to suffer from small number statistics. If you look at the temperatures of the hottest day of the year over the last hundred years you have a hundred days to look at. Looking at all the days naturally gives you many more. This might be offset by other effects from the shape of the distribution however.
Second, and following on from that, extreme values do not tell you how the shape of a distribution is changing. In a case where there's a hard limit on how extreme a value can be, then a change in the mean has to change the shape of the distribution. An example of this might be a Poisson distribution with low means. A Poisson distribution describes the occurrence of events in certain simple circumstances, and you can't have fewer than zero events. So your measurements from a Poisson distribution will never be below zero, and if you're already seeing lots of zeroes its hard to draw an inference from them happening. Similarly the same might happen with climate - if something tends to curb off the temperature the hottest or coldest day can reach then you can't necessarily draw much of a conclusion from how hot or cold that day was.
Thirdly, people remember extreme events. This biases our perceptions of what actually might be happening. Do you remember the 43rd coldest day last year? Do you remember the coldest day last year? You might not remember either, but I can bet which one will be remembered by more readers of this blog. Also, you as sure as hell can't point out one extreme event that happens to coincide with a hacking event and a global warming coincidence - you should sample things a little bit more carefully than that. You can't pull stuff out that happens to be in the forefront of your mind when the issues end up in the news.
Fourthly, climate change is complicated. I've studied atmospheric physics at a fairly high level - about as high as you can get without starting a PhD in it - and I'd never claim to have a good understanding of climate science. It's not clear, certainly in the case of somewhere like the UK which is heavily affected by the Gulf Stream, how global warming affects weather on a small scale. There are also complicating factors which climate scientists no doubt have to spend a lot of time worrying about.
I just want to make the point that it's a really bad idea to go blaming global warming for one-off events, unless you're actually a climate scientist and understand what's going on and the issues surrounding the interpretations of the measurements and how extreme events happen in the context of weather. The average person cannot point at Cockermouth and blame global warming.
Half the fuss over the emails that were (rather shamefully I think) released by someone cracking the University of East Anglia email system was exactly about how climate scientists have to carefully control their data. A colloquial recollection of heavy rain in November is not carefully controlled data, and Nicola Sturgeon stating that
"I'm not a scientist but I know there are many people sitting there in Cumbria and in Dumfries and Galloway whose lives have been wrecked by the flooding we've seen over the past week or so who know from their own experience that something is going wrong with our climate."
is pretty disgraceful. I'm afraid to say that they don't, and it's pretty clear Nicola Sturgeon isn't a scientist. That's not to say I have anything but great sympathy for the people affected by the flooding, but our knowledge of climate change comes from a wealth of careful measurements and study of the facts, and not instinctive feelings about a small number of extreme events. We can see from the continuing debate that this is not a simple matter to get to grips with and touting anecdotes like this does not help.
