January 14 2004
Re:
Attached submission
Please find attached a comment on a paper by Mann, Bradley and Hughes (MBH98), published in Nature in 1998. This follows on an earlier critique we published in October 2003 (MM03), and is separate from a materials complaint currently being investigated by Nature (our contact on that file is Karl Ziemelis). MM03 explained in detail some problems we found in the data used by MB&H, and showed that our re-estimation of their results did not support their original conclusions. The materials complaint pertains to a persistent refusal to make available materials and methods used in MBH98, as well as inconsistencies between the data listings at the Nature SI site and the contents of Professor Mann’s own FTP site. For the purpose of this paper we have taken his FTP site to be the source of data actually used in MBH98. The materials complaint process is (we understand) still underway. However the comment we have written pertains to issues that are not contingent on its resolution.
This paper argues that three proxy series identified by MB&H as crucial to their results are all problematic. This is established for two of the three series just by explaining the nature of the data series themselves. For the third series, it is necessary to elucidate the particular way MBH98 computed principal components, since the procedure was not outlined in the original paper or Supplementary Information thereto. A critical step involves pre-scaling the proxy data in a way that de-centers it, so that the singular value decomposition then selects out series with a particular “hockey stick” shape for over-weighting. We demonstrate this effect three ways: by showing an example of a heavily-weighted proxy compared to a low-weighted one, by comparing results from the MBH98 algorithm versus a standard PC software on a matrix of random numbers, and by comparing the results on a key North American proxy roster. We then show that the central conclusion of MBH98—that the late 20th century values of their northern hemisphere temperature index are unusual compared to the millennium—cannot be maintained when the data and PC problems are rectified.
The paper makes reference to a document released by Professor Mann on the internet (which has been adapted into a paper currently under review at Climatic Change) responding to our MM03 paper. We have attached it, as well as MM03, for the benefit of reviewers.
The MBH98 results have been prominent in the global warming debate, so a critique of a potentially obscure aspect of their method (in this case, principal component analysis) has to be spelled out in detail if readers are to understand why it is material. While many people are familiar with the results in MBH98, it can be difficult to communicate briefly a comment on some aspect of the paper’s structure. There is a lot of terminology related to tree ring proxies and rosters, including codes for regions and time intervals. Therefore the attached is longer than an ordinary comment, though shorter than (and different in intent than) a letter. For this reason, we are concerned that the paper may “fall through the cracks” of the editorial process. We hope that our submission will be assessed on the strength of its content, and if the format must change to facilitate publication, we are open to advice on this.
The attached is not a paper about paleoclimatology, it is about numerical calculation and data quality. However, the results here may prove very contentious and controversial in the climate debate. So in finding third-party reviewers, we would request that at least one reader be an expert in statistics, especially principal component analysis, with no formal connections to the climate field. Our calculations for each figure have been archived and are promptly available for an interested reviewer.
Suggested reviewers include: [snip]
Sincerely,
Stephen McIntyre
Ross McKitrick