It’s not often that blogs come up in congressional hearings, but RealClimate was mentioned yesterday in the Energy and Commerce hearings on the ‘Hockey Stick’ affair. Of course, it was only to accuse us of being part of tight-knit social network of climate scientists, but still, the public recognition is nice.
There is much that could be said about the hearings (and no doubt will be) and many of the participants (Tom Karl, Tom Crowley, Hans von Storch, Gerry North) did a good job in articulating the big picture on climate change independently of the ‘hockey stick’ study as we’ve highlighted before. But it seems to us that there was a missing element in the discussions. That element was the direct implication of the critique that was the principal focus of Wegman’s testimony and that was mentioned periodically throughout the day.
Wegman had been tasked solely to evaluate whether the McIntyre and McKitrick (2005) (MM05) criticism of Mann, Bradley and Hughes (1998) (MBH) had statistical merit. That is, was their narrow point on the impacts of centering on the first principal component (PC) correct? He was pointedly not asked whether it made any difference to the final MBH reconstruction and so he did not attempt to evaluate that. Since no one has ever disputed MM05’s arithmetic (only their inferences), he along with the everyone else found that, yes, centering conventions make a difference to the first PC. This was acknowledged way back when and so should not come as a surprise. From this, Wegman concluded that more statisticians should be consulted in paleo-climate work. Actually, on this point most people would agree – both fields benefit from examining the different kinds of problems that arise in climate data than in standard statistical problems and coming up with novel solutions, and like most good ideas it has already been thought of. For instance, NCAR has run a program on statistical climatology for years and the head of that program (Doug Nychka) was directly consulted for the Wahl and Ammann (2006) paper for instance.
But, and this is where the missing piece comes in, no-one (with sole and impressive exception of Hans von Storch during the Q&A) went on to mention what the effect of the PC centering changes would have had on the final reconstruction – that is, after all the N. American PCs had been put in with the other data and used to make the hemispheric mean temperature estimate. Beacuse, let’s face it, it was the final reconstruction that got everyone’s attention.Von Storch got it absolutely right – it would make no practical difference at all.
This is what MBH would have looked like using centered PC analysis:
Red is the original MBH emulation and green is the calculation using centered PC analysis (and additionally removing one of the less well replicated tree ring series). (Calculations are from Wahl and Amman (2006), after their fig. 5d). Pretty much the same variability and the same ‘hockey stick’. We’d be very surprised if anyone thought that this would have made any difference to either the conclusions or the subsequent use of the MBH results.
In fact, it’s even more simple, Throw out that PC analysis step completely, what do you get?
Again, red is the original MBH98 multiproxy+PC analysis, green is if the raw data are used directly (with no PC analysis at all). (This comes from Rutherford et al (2005) and uses a different methodology – RegEM – to calibrate paleoclimate proxy data against the modern instrumental record, but that doesn’t make any difference for this point).
Why doesn’t it make any difference? It’s because the PC analysis was used to encapsulate all of the statistically relevant information in the N. American tree ring network and so whatever patterns are in there they will always influence the final reconstruction.
So what would have happened to the MBH results if Wegman and his colleagues had been consulted on PC centering conventions at the time? Absolutely nothing.
Can we all get on with something more interesting now?