Okay, where were we. MM03, the McIntyre and McKitrick critique of the MBH98 "hockey stick" paleoclimate reconstruction, gets dramatically different results with what I called a "reasonable approach" to what's supposed to be the same data. That's not quite accurate: MM03 is not presenting an alternative calculation to MBH98, and arguing that it is more appropriate than MBH98's, it is instead supposed to be the same calculation, albeit with calculation errors corrected. So it repeats the same calculation, on the same data, fixes some errors in the calculation and data found along the way, and gets a dramatically different result. What do the original authors have to say about that?
They don't think much of it, not surprisingly. They don't bother to submit the response anywhere, just drop it on their FTP site. Apparently they don't think anyone important reads Energy and Environment. The main observation in the note is that MM03 unwittingly omits most of the early data which allows 16th century and earlier reconstruction, and doesn't catch the error because MM03 doesn't do any cross-validation. Ouch! It is quite embarassing to have the principal feature of your reanalysis be "leaving data out", and worse yet to have this pointed out by the people you were trying to critique in the first place. But embarassment is not determinative, and MBH98 can also be seen as embarassed due to the data not quite matching up with what was reported in the original Supplementary Information. So let's set embarassment aside and continue.
Perhaps McIntyre and McKitrick do not agree that they left data out, and stick by their original claim that the differences are due to corrections of data and calculations? No, peeking ahead at their response to the unpublished note (which they submitted to Nature!), they accept that the difference is in the deleted data, and (armed now with the knowledge, provided by the unpublished note of MBH, of what data they deleted) begin to critique the quality of the deleted data series. So everyone agrees it's the data.
What about the calculations? The story about the calculations is very interesting. MM03 write: "the MBH98 principal components fail to maximize explained variances". Since a principal component maximizes explained variance by definition, it's difficult to understand what could be going on here. Perhaps the authors of MBH98 have no idea what a principal component is? Could be, but unlikely they'd base their paper on it then. Did they roll their own PC calculations, screwed up somewhere, and didn't bother testing? Perhaps, but given the number of PC implementations available, it seems like an odd choice. Did they use a well-tested PC code, which turned out to have a bug? Unlikely, but it's happened before: the singular value decomposition routine (which actually forms the core of a PC calculation) in the generally high quality, heavily used, and even open code package Numerical Recipes had a subtle bug for a number of years. Could a communication failure have occured, such that what MM03 is running principal components on is not the same as what MBH98 is running principal components on? It seems within the realm of possibility. One can have many interpretations of a communication failure ("you didn't adequately disclose", says he; "you didn't bother to read the paper", says she), but the failure to communicate is objective.
Among these many possibilities, how does MM03 determine what happened and resolve the discrepancy? It doesn't: it just "corrects" the problem, moves on, and when the final result differs from the original paper, concludes that the original result is erroneous. An interesting attitude, to be sure. I think I haven't heard the last about the calculations.