28 October 2009

Trends

Ars Technica | John Timmer | Talk of "global cooling" based on bogus statistics

There's an inevitable problem with trying to find trends in data that is subject to a great deal of random variability: unless the most recent point was a record high, it will always look like there's a downward trend.
Isn't it just as true that unless the most recent data point was a record low, it will always look like there's an upward trend?

Both of our statements are only true with appropriately loose definitions of the word "trend."

The problem isn't even the degree of variability, it's the degree of variability in relation to length of the time period you're describing the trend over.

In Timmer's defense, he does get into the matter of period length later in his post, but that's still a dangerous sentence to put out there as an opening line.

He also says this:
But, as we noted in February, the reent drop in temperatures has been so small that 2008 was still the 10th warmest on record. Other recent years were equally warm or warmer, while the hottest year on record, 1998, was unusually warm compared to the surrounding years. In fact, if you started tracking trends in either 1997 or 1999, you saw a general increase in global temperatures.
So if you cherry-pick your end points you can make your data say whatever you want. Right on.

But let's keep in mind that the only really accurate global temperature records we have start 30 years ago, so saying a year is the 10th warmest on record puts it firmly in the interquartile range, or what I might describe technically as warmish.

No comments:

Post a Comment