Here it is in a different form.

Although slightly repetitive, in the style of the times, the message is kept simple and is notable for predicting certain language traits that have become especially popular in recent times (“It was later than I thought ….”)

References to ships and boats clearly speak to issues about sea levels and unresolved issues about flotsam and jetsam or contentious papers are discussed. (“… all that could not sink or swim was just left there to float.”)

Matters concerning appeals to authority, the paucity of useful content as discovery over time erodes the initial hypothesis and the need to make ones concerns public are also subjects about which the authors wax lyrical.

I have found both text and the spoken word forms of the commentary and offer these for review by readers.

Note that the graphics used for the spoken word version linked here are not the best but it was the shortest version of the commentary I could find in the time available … readers are of course free to indulge their passion for search engines and seek alternative presentations – of which there are a few.

In Steves original post the match wasn’t just too perfect, it was exact. The reason for this is that RegEM masks data which are available and infills the missing values. This makes sense for data which are sparsely missing or cases where the infilling data is of equal quality to the data being replaced. As in typical expectation maximization where only a few values are missing from data of the equal quality.

In the Steig recon, the sat data is lousy for trend and horribly noisy in comparison to the surface stations so Steve’s approach of replacing the entire dataset with the infilled values is the only way to go. I think in the case of fitting proxies to temperature the same kind of condition exists where there is a mismatch in fidelity of the proxies to the instrumental temperature.

If you look at the RegEM ttls code there is a line where the available data is masked and only the missing data is replaced.

Ahhh. So we now have de facto splicing. In all the Briffa/Kaufman stuff recently, someone (Rattus non Readus?) said (essentially): I don’t care about reconstructions or divergence after 1880 or so because we have the instrument record.

Prediction: Team paleos will “move on” to de jure splicing. We’ll see recon after recon showing the 1,000 years (2,000? 5,000? mill-yun?) prior to 1880 as flat as a pancake, followed by a CO2 induced blade from the instrument record thereafter.

Look on the bright side, it’s not as nutty as “teleconnections”, which was so bizarre it was hard to know where to start.

infilling gridded temperature data is a very different animal than reconstructions using proxy data. Since we don’t have actual data and runs, it would be more productive of my time (after analyzing the new Briffa data) to revisit RegEM. Plus it looks 99% certain to me that his BARSAT method has important elements in common somewhere with RegEM which is what causes the “too perfect” calibration period reconstructions. These sure look like hidden splicing is an effect of the algorithm (as Jeff Id also agrees.) In my emulation of RegEM, I tweaked the code to keep unspliced results (that’s the Xmis that Jeff mentioned) and maybe this can be used to perform statistical tests that the authors neglect.

I agree completely. The EIV method has become pretty clear as to what it is but it’s taken time to figure out. I don’t think we’ll hear Mann saying -noone splices temp records on proxies anymore. It would be interesting to see the Xmis matrix behind the recon for the last iteration though. While I expect it correlates fine, there is a chance that it doesn’t do very well.

The code is even more time consuming that RegEM so beware. Also we don’t have the actual data used in his 600-year reconstruction yet.

Also before people spend a lot of time on this method, it would be worthwhile working through the algebra both for this and RegEM. We noticed a year ago that Mannian RegEM recons yielded too-perfect reconstructions in the instrumental period i.e. the recon was in effect splicing the target instruments. This needs to be revisited with the additional knowledge of the method developed in the Steig analysis – it always takes a while to refresh this file and some clear time is needed – something that I won’t have until the Briffa thing is analyzed.