Bad statistics
I’m fed up of the news today; we’re having a bit of warm weather, and the temperature today was forecast to be the “hottest September for more than 100 years”. See for example, The Daily Mirror.
But is this all that unusual? I am assuming that the data being used is the Hadley Central England Temperature data ; I’ve no real interest in weather or climate statistics, but this is the longest series available in the world, and I think well respected. I’ve not done the calculations to check, but this daily detailed data goes back goes back to 1878, or 133 years before this one.
Now, let’s assume that we had 133 years worth of data, and we get a new one for this year, so we have 134 data. What is the probability that this year’s is the highest of the 133? Assuming we know nothing else, under this null hypothesis, the best guess we can make on this is 1/134.
So any day in the year chosen at random should have a probability of 1/134 of being the hottest day with that date on record. Thus, every year, we should expect to see the hottest particular day of the year approximately 3 times; in other words, it’s not that surprising.If we broaden it to the hottest or the coldest day on record, something “remarkable” we should see this headline 6 times a year, or once every two months.
Nor is it particularly surprising that if day x is the hottest, day x+1 is the hottest on record, although the calculations here would require some more science to work out. Now the cynic in me seems to think that it might be an excuse for newspapers to publish lots of pictures of fruity girls in bikinis; but surely The Daily Mail would never do that?
Non-news story. How about “Weather patterns still random”. Move on!
Originally published at Ben Parker's Blog. You can comment here or there.