date: Wed Jun  8 08:57:43 2005
from: Tim Osborn <t.osborn@uea.ac.uk>
subject: Fwd: Von Storch et al critique of MBH (fwd)
to: "Keith Briffa" <k.briffa@uea.ac.uk>

   Apparently we owe Mike Mann an apology!
   B*ll*cks to that!
   Tim

     Date: Tue, 7 Jun 2005 18:08:14 -0700 (PDT)
     From: "David M. Ritson" <dmr@slac.stanford.edu>
     To: t.osborn@uea.ac.uk
     Subject: Von Storch et al critique of MBH (fwd)
     Dear Tim,
     I pursued the question as to whether MBH does or does not constrain/force
     their proxy results into agreement with the observational data  for 1902-1982
     with Mike Mann.  The answer is indeed that they calibrate by forcing
     this agreeement. In more detail
     Their basic assumption was that the `growth', X_ij for proxy i and year j was
     given by
        X_ij=SUM_m(S_mi*A_mj)+e_ij
     where, A_mj is the amplitude of mth EOF for year j, and S_mi is
     the `sensitivity' of proxy i to EOF m, and the e_ij are the
     associated random noise.
     The m EOFs are determined from the observed temperature anomaly data for
     1902-1981. The sensitivity factors, S_mi are determined to best fit the
     1902-1981 proxy data to the observed data. Given the S_mi, they can
     reconstruct the best-fit A_mj from their X_ij.
     Therefore, by `definition' their reconstructed
     signal for 1902-1981 must, within errors, agree with the original
     experimental temperature anomaly signal.
     The MBH results do indeed agree within errors with the 1902-1981
     observational temperatures.
     Von Storch et al in their Science letter (supported by your perspective) in
     their Figure 2A show simulation results for 1902-1981
     that qualitively very `strongly diverge from the `observed'
     data for 1902-1981 as a function of added white noise.
     One can only conclude that VS04 did not follow the MBH calbration procedures.
     Enclosed is some correspondance with Hans relative to this. Frankly I think
     Hans, and you guys, owe Mike an apology on this point?
     Below, unfortunately (apologies) in order of present backwards in time, is
     some relevant correspondance with Hans.
     Cheers
     Dave
     ================================Correspondance=======================
     ---------- Forwarded message ----------
     Date: Mon, 06 Jun 2005 22:56:01 +0200
     From: Hans.von.Storch@gkss.de
     To: David M. Ritson <dmr@slac.stanford.edu>
     Cc: Eduardo.Zorita@gkss.de, fidelgr@fis.ucm.es, Hans.von.Storch@gkss.de,
          Julie.Jones@gkss.de
     Subject: Re: VS04 (fwd)
     Dave,
     I am impressed by your certainty, but I suggest that you just wait for our
     paper.  I am actually not that stupid as you indicate.  At least, I
     believe so.
     The statement "Fundamentally whenever you scale units, cm to km, ozs to
     kg, deg F to deg C, etc, you are "regressing a trend on a trend"." may be
     a bit too simple. In the meantime, I propose that you write a comment on
     our paper so that we can discuss that publicly - since you seem to know so
     well what this is about.
     By the way, it was MY answer not our groups' answer.
     All the best,
     Hans
     From:"David M. Ritson" <dmr@slac.stanford.edu>
     Sent:06.06.2005 22:27
     To: Hans.von.Storch@gkss.de
     Dear Hans et al,
     Thanks for your reply which clarifies your group's position. Fundamentally
     whenever you scale units, cm to km, ozs to kg, deg F to deg C, etc, you
     are "regressing a trend on a trend".
     MBH, in terms of your simplified model, use a thermometric scale based
     on the tree-ring sizes, and then, entirely correctly, they calibrate the
     proxy
     temperature scale relative to the observational absolute temperatures from
     1902 to 1981. This is exactly what students do in elementary lab classes
     when
     they fabricate thermometers, do an experiment and finally calibrate their
     fabricated thermometers against a calibrated thermometer. Your group
     according to your e-mail, omitted the calibration step. Under such
     circumstances
     results are necessarily meaningless.
     I feel embarassed to be going into such elementary considerations with you
     of
     all people.  From a formal statistical point of view when you regress a
     gradient
     against a gradient intrinsic errors from a single measurement are
     indeterminate,
     ie you are dividing zero chisquared by zero degrees of freedom. Of course
     extrinsic errors exist and are readily found from an ensemble of
     measurements,
     and/or from the errors associated with the procedure used.
     In actuality MBH98 used a more sophisticated procedure. They had eighty
     annual measurements they regressed against, and as outlined in my previous
     e-mails they calibrated sensitivity factors to a few leading EOFs derived
     from the observational data. This of course does not change the gut
     argument. MBH98 calibrated their proxy temperature scale, and according to
     your
     communication your group omitted this step in your simulation of MBH
     procedures.
     I certainly feel, in view of the scientific and media attention to your
     letter, that it is essential that this omission be fully clarified to the
     scientific community and that such a clarification would be most
     meaningful if
     it came directly from you people.
     Sincerely
     Dave
     On Sun, 5 Jun 2005 Hans.von.Storch@gkss.de wrote:
     > David,
     >
     > we had something prepared for this question for another comment we got
     on
     > our science paper. This comment was eventually no accepted for pu
     > blication, and we will have now all this in a separate publication of
     our
     > own. This is work in progress and we will make this available as soon as
     > we are done. You know that Mann had withdrawn his comment on our article
     > recently - after he had told journalists long ago that it was already in
     > press.
     >
     > The answer in a nutshell is: Thou shalt not regress a trend on a trend.
     > Or, in other words - a stats model based on 1 dgf can not be
     statistically
     > be justified (but maybe physically). A fundamental and most meaningful
     > principle of statistics.
     > Cheers
     > Hans
     >
