Anthony Watts has maintained a blog on Climate Change, What’s Up With That? that I’ve enjoyed for some time. He not only presents good science analysis, of his own and his guest bloggers, but also maintains a light touch. Sometimes a bit too light, I think, obscuring the data and analysis in popular rhetoric.
However, this article: AMO PDO= temperature variation – one graph says it all | Watts Up With That? is bang on!
In my other life as an archaeologist, I worked with dendrochronological time series a great deal, attempting to sift prehistoric climate change data out of tree rings from drift wood and archaeological wood around the Bering Strait and the Yukon River. My experience with these statistical analyses matched Anthony warnings about the limitations of time series smoothing and the dangers of using smoothed data as inputs in further analysis.
Smoothing data introduces intentional error into the datatset, with the expectation that the simplified dataset will more clearly reveal meaningful underlying patterns. In dendrochronology, we seek to eliminate the normal growth curves of different tree species in order to reveal the climate influenced growth variations.
It’s important to keep in mind that a smoothed dataset is not real and cannot be used for further data manipulation. Each iteration of data analysis must begin with the raw data, and analysed with an increasingly complex set of parameters in order to achieve more complex flights of analysis. When a smoothed dataset is statistically manipulated, all manner of artifacts and false patterns can be introduced into the results, which cannot be distinguished from “real” outcomes.
I see the same types of results occurring in climate data that I witnessed in over manipulated tree ring data in dendroclimatology.
In statistics, as in all walks of life the KISS admonition applies: Keep It Simple, Stupid!