let x be the known values and y readings on a scale (temperature and proxy/thermometer, for example). Calibration problem is how to estimate new value X by reading Y, using calibration data (x,y). Assuming data is zero-mean (just saving space), we have

CCE=Classical Calibration Estimator

ICE= Inverse Calibration Estimator

CVM= Composite Plus Variance Matching

those lines are plotted in the figure. (BTW, ICE minimizes calibration residuals)

I made a scatter diagram of Juckes et al MBH calibration data. The plot includes both regression lines. And these are actually ICE and CCE – take a new observation y (proxy) and use the regression line to estimate x (temperature). CVM is in the middle. I think ICE is not suitable, it is very unlikely that the calibration data is a sample from binormal distribution (check the residuals..). CVM lacks theoretical justification (30 000 hits per day, some of the authors read this blog, if it exists why it is not written down in here??). CCE is left (if I would be mean and provocative I would say that the only problem with CCE is that you can’t make alarming conclusions with the results, and climate science wouldn’t get very far.)

Comment TAC: “Is this necessarily a bad thing? The blogs are fast and accessible.”

In contrast to the CA blog, which is fast and accessible, when RC exhibits their heavy-handed approval process (not infrequently), RC is neither.

John,

With open comments, Steve is not the only one who can post critical comments on HT work. It remanins to be seen whether anyone will get such a chance in the future. I hope that this sort of review becomes the standard, but I would have to assume that the HT will fight it.

I tried to get a user name and password on Copernicus so that I could comment on COP. Two attempts and both failed. Perhaps I should not have used my real name.

The problem of A2 is that the given statistical model is not sufficient for deciding what would be the optimal calibration estimator. You can choose ICE or CCE, depending how you interpret past temperature statistically (wrote it down ). PCA is somewhere between those two. And this paper introduced CVM (tilts the regression line towards the observation axis). Four different ways to calibrate, imagine how many options we have for multivariate calibration!

Maybe they followed the Climate of the Past and CA postings and didn’t want to deal with the critiques.

I fear for the success of the experiment unless all or even most of the other journals follow suit.

And Nature has just announced that it will be ending its experiment with (voluntary) “open” reviewing. Apparently, authors didn’t want to risk the process, and public comment was lacking.

Curiously, the Team has mostly relied on non-peer-reviewed literature (realclimate) to respond to us and climate scientists have been prepared to rely on that.

Is this necessarily a bad thing? The blogs are fast and accessible. Where the HT — or anyone else — has a point to make, they can make it quickly on a blog and receive a quick reply. Discussion in the peer-reviewed literature, on the other hand, is painfully slow, seldom finds a large audience, and is often incomplete.

One final thought: Among blogs, the ones that seem to work best (i.e. are most enlightening and entertaining) are those with an open environment and little or no censorship, where the reader has responsibility for filtering.

However, Wahl and Ammann (Climatic Change 2006 – still not published) looks to me like the journal was trying to meet IPCC deadlines as its “provisional” acceptance and “final” acceptance both coincide with IPCC cutoff dates (as they then stood). We had a chance to review the first submission and were at least able to embarrass the journal into requiring the disclosure of the failed verification statistics (not included in the first submission).