Comments on: Attenuation in Mann et al 2005

By: Steve McIntyre

Steve McIntyre — Tue, 11 Jul 2006 02:59:00 +0000

The words “come clean” were not used by me, but were used by Mann.

I am making comments here about published articles, not about participants in the debate here. Steve B, your views are nearly always diametrically opposed to mine, but surely you can vouch that I have been punctiliously polite to you here.

I’ll continue to post up notes from time to time on the NAS Panel.

TCO, the rules apply to you.

By: TCO

TCO — Tue, 11 Jul 2006 02:35:01 +0000

Steve B. those are the rules. And if you bug Steve enough he can generally be shamed into removing some teases. However, it is an accepted practice on this blog that these rules apply to all except for me. And I think you’re a wimp for spending all your time on how things are said and how much snark there is versus argying the science.

By: Steve B

Steve B — Tue, 11 Jul 2006 02:22:42 +0000

Your rules say that “Refrain from personal abuse and swearing” and “Never attribute ulterior motives to another participant”, then I see you use words in this comment like “hyper-ventilated” and “come clean” and “descended to” and “bullying”. It seems that you are violating your own rules.

I have not made up my mind yet on the NAS report, plan to read it, and am seeking other comments. Your web site has not yet helped. I would appreciate a not-too-technical summary in one place of what you think is wrong with that NAS report.

By: TCO

TCO — Mon, 10 Jul 2006 04:49:37 +0000

Thanks for the info on when it makes a difference. I’m interested in method choices and the bases for them and the plusses/minuses. Sorta like covariance/correlation…

By: TCO

TCO — Mon, 10 Jul 2006 04:46:36 +0000

Are these the proxies that were used in MBH98 calculation of RE and can you duplicate that calculation of Mann’s.

By: Steve McIntyre

Steve McIntyre — Mon, 10 Jul 2006 04:39:50 +0000

The RE values are lower with these very tame proxies than with actual MBH proxies. It’s not because they can actually out-perform this type of system, but because the actual proxies are not signal plus white noise and there’ enough play in the method to enable amplitude tuning.

By: Steve McIntyre

Steve McIntyre — Mon, 10 Jul 2006 04:32:33 +0000

Any multivariate method just gives weighting factors for the proxies. For tame networks with equal amounts of white noise (which is what is done here), trending/detrending doesn’t make much difference.

It does make a difference in what weight is assigned to bristlecones, but the tame networks don’t have proxies like that. The whole fuss seems like a total crock to me.

By: TCO

TCO — Mon, 10 Jul 2006 03:55:36 +0000

Were you able to duplicate the RE calcs? Any ideas on when detrending would/would not make a significant difference? IOW, for what types of data sets it would drive a difference and for what not?