or…Some thoughts on Personal Responsibility and the Peer Review Process

Eric Steig

Ryan O’Donnell made a series of serious of allegations against me at ClimateAudit, in the context of our friendly dispute about whether his new paper in the Journal of Climate supports or ‘refutes’ my own results, published in Nature.

To his credit, Ryan has offered to retract these allegations, now that he is a little better acquainted with the facts. However, it is still important, I think, to set the record straight from my point of view. There were such a great number of claims about my “dishonesty,” “duplicity” and [implied] stupidity, all of which are untrue, that it really isn’t worth trying to respond in any detail. Just responding to the main two ought to suffice to make the point.

“Eric recommends that we replace our TTLS results with the ridge regression ones (which required a major rewrite of both the paper and the SI) and then agrees with us that the iRidge results are likely to be better . . . and promptly attempts to turn his own recommendation against us.”

“[in his RealClimate post]…he tries to … misrepresent the Mann article to support his claim [about the iridge routine] when he already knew otherwise. How do I know he knew otherwise? Because I told him so in the review response.”

While it is quite possible that O’Donnell believes both of these claims, they are both false, as it is rather easy to demonstrate.

First, I never suggested to the authors that they use ‘iridge’. This was an innovation of O’Donnell and his co-authors, and I merely stated that it ‘seems’ reasonable. As O’Donnell’s co-authors are fond of pointing out, I am not a statistician, and I did not try to argue with them on this point. I did, however, note that previously published work had shown this method to be problematic:

“The use of the ‘iridge’ procedure makes sense to me, and I suspect it really does give the best results. But O’Donnell et al. do not address the issue with this procedure raised by Mann et al., 2008, which Steig et al. cite as being the reason for using ttls in the regem algorithm. The reason given in Mann et al., is not computational efficiency — as O’Donnell et al state — but rather a bias that results when extrapolating (‘reconstruction’) rather than infilling is done.

Second, I was the reviewer of the first three drafts of O’Donnell et al submission. However, I did not the review draft four, which was the published one. , and which is markedly different from draft 3 [note correction: it has been pointed out that it’s not really very different; in other words, my criticisms of draft 3 were ignored]. Nor was I ever shown their response to my comments on draft 3, so I did not in fact ‘already know’ what O’Donnell claims I did. It appears that the editor was swayed by the arguments that I was not a helpful reviewer. In other words, even if one believes that I was “bullying” them into showing particular results, they still had the last word (as any author should).

The fact of the matter is that my reviews of O’Donnell’s paper were on balance quite positive. I wrote in the confidential comments to the editor in my very first review that

I emphasize that I think that a fundamentally reworked version of this manuscript could potentially provide a useful scientific contribution, and many of the points made do indeed have scientific merit. Indeed, the authors have done a very thorough analysis, and are to be congratulated on this.

In my second review, I wrote that “O’Donnell et al. have substantially improved their manuscript … and clarified a series of items that led to some confusion on my part.”

Page 1 of 3 | Next page