In the NAS press conference, Bloomfield said that you could get a Hockey Stick from an average of the proxies. This was a pretty misleading comment. You CAN’T get a HS from averaging the MBH98 proxies. We showed this to the NAS panel on our presentation as follows:
The simple mean of the Mann et al. data is in the top panel of Figure 2. One notes that the 20th century is unexceptional and, for what it is worth, that there is a downward trend over the 20th century. The final reconstruction, shown in the bottom panel, yields a remarkably different story, in which the historical values were low prior to the 20th century, and the data have a strong upward trend after 1900.
Original Caption: Figure 2: Top — Average of 415 series in MBH98 “dataall” dataset archived in July 2004. Bottom — MBH98 reconstruction.
This idea that you can get a HS by averaging the Mannian proxies seems to be an urban myth among climate scientists. Marcel Crok told me that Nanne Weber of KNMI in Holland told him the same thing.
The only way that you can get a HS by averaging MBH98 proxies is if you previously pick proxies. Arguably a process like this takes place in the “other” studies that don’t use principal components. We’ve never suggested that Mannian principal components, plus partial least squares regression, is the only way to cherry pick data. It’s just a mechanized method of data mining, which has the appearance of objectivity. There’s always the old-fashioned method – pick series with HS by hand. For example, in the North American tree ring network that we’ve discussed so much, the average of the 14 most HS-shaped series (all bristlecones) is pretty much identical to the PC1 (naturally so, since the weights in the first eigenvector essentially wipe out contributions from all other series.)