Here are some first thoughts. So far I don’t see anything in W-A that affects any of our results. Indeed, I expect that W-A is going to be very positive for this debate, as the points at issue are going to be narrowed sufficiently closely that any people actually interested in the results (as opposed to spin) will be able to finally compare apples and apples. It will require a little cutting through the packaging, but since they’ve commendably provided code for the first time, it won’t be too hard.
As I work through their code in the next few days, I will highlight any points of methodological difference and and will immediately provide notice of the issue. This is what you’re supposed to do in public businesses (and most businesses do) and it’s a good policy.
One of the first questions that readers of this site have had e.g. Paul Gosling – what happens without the bristlecones. Here’s a cutout of their run without bristlecones. In Hockey Team style, they provide a spaghetti graph so that it makes it more difficult to examine the individual calculation. However, you can pretty much see the purple calculation up to the end. The legend says that if the bristlecones are omitted, the calculation is "without merit".
It will be interesting to see how they comment on Mann’s grandiose claims that their reconstruction was "robust" to the presence/absence of dendroclimatic indicators in total (since a very quick inspection of this graphic confirms our report that this representation was false.) It will also be interesting to see their explanation as to how – if a reconstruction without bristlecones is "without merit" – a calculation with bristlecones can possibly have any merit in the face of our many criticisms of bristlecones not only as a flawed proxy, but as a proxy with flaws known to Mann et al.
In any event, the graph below shows that they don’t get a hockey stick without bristlecones either.
Figure 1. Excerpt from figure on Amman’s ucar website.
A first comment on the code. It’s really nice to see a paleoclimate article with code. Good for Wahl and Amman on this. I presume that we can take a little credit for this. It’s provided in a UNIX version. I was able to make a few tweaks to the code to make it Windows compatible quite quickly. I’ll post up these tweaks a little later in the day.
I now have an operational 15th century step result. The advantages of posting code should become obvious very quickly: reconciliations that otherwise take months of inquiry (and can still be inconclusive) can be done in a matter of hours, if not minutes.
Their results are NOT a replication of MBH, but an emulation and only an emulation of the regression-calibration step. For example, there’s nothing in the code to show an implementation of Preisendorfer’s Rule N for retention of PCs in tree ring networks in MBH98 – which was mentioned out of the blue last year for the first time as being a critical step. There’s nothing about estimating confidence intervals.
Although the website says that they "reproduce" MBH98 results, it would be more accurate to say that they emulate the MBH98 reconstruction method. For example, the correlation of their 1400-1450 reconstruction to the 1400-1450 MBH98 reconstruction was only 0.69 in my first check. I’ll plot up their emulation of this step against ours a little later, but at a first blush, they look pretty similar. The similarity is not chance, as in the main steps of their emulation, they use equivalent methodology to ours, the first version of which was archived almost 18 months ago.
Ironically, I’m not entirely sure yet what’s in here that isn’t already in our E&E paper (although with a different slant obviously). We show an emulation of MBH98 as do WA. We report that various permutations and combinations yield, on the one hand, high 15th century results and, on the other hand, low 15th century results, and that these differences pertain to the impact of the bristlecone pine and Gaspé series. Reading between the lines, it looks like Wahl and Amman get virtually the same results. In our respective emulations, it looks like the handling of the regression-calibration steps to make "reconstructed" temperature principal components is identical.
I’ll go through how they go from the reconstructed temperature principal components to NH temperature later today. Mann has carried out some re-scaling here and it’s never been very clear what he did. It looks like W-A had some difficulties here as well (even with input from Mann) as there’s code added in as late as April 2005 modifying a scaling step. I expect to be able to reconcile our results to W-A results step by step without a whole lot of difficulty. I think that the entry of the W-A articles with code is a highly positive development.
Naturally, their presentation will be filled with all sorts of hyperbole and probably even some disinformation, but, for the first time, some one from the Hockey Team has stepped out of the foxhole with some code. We will be able to reconcile calculations in a business-like way to actual Hockey Team calculations and show exact points of difference without shadow boxing.