Climategate 2.0: Mann’s ‘unpleasantness’ and ‘suspect’ statistics

Don’t bother Michael Mann about his suspect statistics unless you want to get your head chopped off.

From the Climategate 2.0 collection: The Max Planck Institute’s Klaus Hasselmann is concerned about Hans von Storch’s criticism of the Michael Mann’s hokey stick. Hasselmann suggests that someone try to reconcile the two.

The University of California-San Diego’s Tim Barnett responds,

not to be a trouble maker but……if we are going to really get into the paleo stuff, maybe someone(s) ought to have another look at Mann’s paper. His statistics were suspect as i remember. for instance, i seem to remember he used, say, 4 EOFs as predictors. But he prescreened them and threw one away because it was not useful. then made a model with the remaining three, ignoring the fact he had originally considered 4 predictors. He never added an artifical skill measure to account for this but based significance on 3 predictors. Might not make any difference. My memory is probably faulty on these issues, but to be completely even handed we ought to be sure we agree with his procedures. [Emphasis added]

Oxford’s Myles Allen responds to Barnett:

I completely agree with Tim, but the question is whether we have either the energy or thick enough hides. My recollection of the experience of asking (I thought quite politely) Mike about this kind of thing is rather unpleasant [Emphasis added]

The e-mail exchange is below.
cc: “Bamzai, Anjuli” , christopher.d.miller@noaa.gov,
David Karoly , francis , Nathan
Gillett , “Prof.Dr. Hans von Storch”
, Gabi Hegerl , Jesse Kenyon
, Klaus Hasselmann , “Stott, Peter”
, Ben Santer , Reiner Schnur
, “Tett, Simon” , Karl Taylor
, Tim Barnett , Tom Crowley
, “Pennell, William T”
date: Thu Oct 14 13:03:44 2004
from: Phil Jones

subject: RE: spring meeting
to: Hans.von.Storch@gkss.de, Myles Allen
Dear All,
I’ve been away and am only just picking up the various emails. Hans’ study is just a
critique of the MBH method as he says. It is not an alternative, nor should anyone claim
that a modelling result is a substitute for reality. The only way to improve our knowledge
of the past is to get more evidence and put it together in a variety of different ways.
There is no way I can accept Hans that your reconstruction (because it uses no proxy
data)
can be a reasonable reconstruction. The only reconstructions we should be considering
of
the past are those based on proxy data. Your paper is what you said a methodological
critique –
not a replacement/alternative. Models should never replace reality. Models can help our
understanding, telling us how to use our data better.
I too suspect that Mike will write a rebuttal – knowing Mike he probably has already.
The point
that I think Myles alluded to, which was in the Perspectives piece (by Tim Osborn and
Keith Briffa) that maybe not all have seen (as this was Science Express) is that if the
amplitude of the Mann curve is larger then the climate sensitivity is larger. The
skeptics,
by their own logic, should be accepting the Mann series as it implies less change in the
21st
century. Instead in a number of emails that I’ve glanced at whilst away they seem to
accept
the Hans et al paper with gusto – while in September they were saying all models were
wrong
because they couldn’t replicate features of the 20th century. To them, models are right
only
if they in line with their preconceptions.
I still think we should have a paleo focus during the meeting in April. The issue is not
going
to go away.
Cheers
Phil
At 08:24 12/10/2004, Hans.von.Storch@gkss.de wrote:
Liebe Leute,
it seems that you have included me in your correspondence. I guess this was not
intended. NOne of usn is part of IDAG (whatever that may stand for).
After having conferred with me co-authors I will respond to your inquiries in a
constructive manner. But let me emphasize that we do not claim that “my” curve is
better than Mann’s. First it is not “Hans'” curve but the ERIK-DEN-RODE curve, derived
from an ECH-G simulation run by Fidel Gonzales Rouco; the analysis was done by the
whole consortium, in particular by Eduardo Zorita; I had the pleasure to come up with
the general strategy of THIS specific study. Thus, please not: “Hans’ study” or “Hans’
curve”.
The study is merely a methodical critique of the MBH method. Our curve is consistent
with the hockey stick, but it may very well deviate significantly from the “truth”. We
are claiming, however, that our curve is as reasonable a reconstruction of the real
conditions as the MBH curve. We may sharpen this conclusion in some future.
Eduardo has examined many aspects of the ERIK-DEN-RODE simulations and
compared with the
performance of simpler models and with claims derived from paleo-studies. We expect
a
rebuttal from Mike Mann, and we intend to use the material for dealing with this
rebuttal. We are very confident that Eduardo’s arguments are more than sufficient for
this prupose, but we prefer not to publish these findings before we have seen officially
Mann’s rebuttal.
Regards
Hans
Hans von Storch
Institute for Coastal Research, GKSS Research Center
Max-Planck-Strasse 1, 21502 GEESTHACHT, Germany
ph: +49 4152 87 16830, fx: +49 4152 87 2832
mobile: + 49 171 212 2046
[1]http://w3g.gkss.de/staff/storch; storch@gkss.de
Myles Allen
10/12/2004 11:44 AM
To
Tim Barnett , Gabi Hegerl , Klaus
Hasselmann

cc
“Prof.Dr. Hans von Storch” , francis
,
Reiner Schnur , Phil Jones

, Tom Crowley
, Nathan Gillett , David Karoly
, Jesse Kenyon ,
christopher.d.miller@noaa.gov, “Pennell, William T” , “Tett,
Simon” , Ben Santer , Karl Taylor
, “Stott, Peter” , “Bamzai, Anjuli”

Subject
RE: spring meeting
I completely agree with Tim, but the question is whether we have either
the energy or thick enough hides. My recollection of the experience of
asking (I thought quite politely) Mike about this kind of thing is
rather unpleasant.
Myles
Climate Dynamics Group
Atmospheric, Oceanic and Planetary Physics
Department of Physics, University of Oxford
Tel: 44-1865-272085/095
Fax: 44-1865-272923
E-mail: myles.allen@physics.oxford.ac.uk
—–Original Message—–
From: Tim Barnett [[2]mailto:tbarnett-ul@ucsd.edu]
Sent: 11 October 2004 16:42
To: Gabi Hegerl; Klaus Hasselmann
Cc: Prof.Dr. Hans von Storch; Myles Allen; francis; Reiner Schnur; Phil
Jones; Tom Crowley; Nathan Gillett; David Karoly; Jesse Kenyon;
christopher.d.miller@noaa.gov; Pennell, William T; Tett, Simon; Ben
Santer; Karl Taylor; Stott, Peter; Bamzai, Anjuli
Subject: Re: spring meeting
not to be a trouble maker but……if we are going to really get into
the
paleo stuff, maybe someone(s) ought to have another look at Mann’s
paper. His statistics were suspect as i remember. for instance, i
seem
to remember he used, say, 4 EOFs as predictors. But he prescreened them
and threw one away because it was not useful. then made a model with
the
remaining three, ignoring the fact he had originally considered 4
predictors. He never added an artifical skill measure to account for
this
but based significance on 3 predictors. Might not make any difference.
My
memory is probably faulty on these issues, but to be completely even
handed
we ought to be sure we agree with his procedures. best, tim
At 07:30 AM 10/11/2004, Gabi Hegerl wrote:
>Hi Klaus,
>
>Several responses:
>- the IDAG wants to raise the issue at the spring meeting from the
paleo side
>
>- the signal-to-noise analysis may be so straightforward it hurts, but
a >good idea.
>What I have already published (GRL, 2003) and would be super-simple to
>update is a simple linefitting
>detection analysis (time-only multiple regression for decadal means)
>between ghg, solar and
>volcanic forcing. The residual (so 50-100 decades minus 3 fitted
degrees
>of freedom) is used
>to estimate the noise, so more noisy paleo time series => smaller
>signal-to-noise ratio.
>It didn’t work too well with Mike’s timeseries (which is not discussed
in
>the paper
>since an anonymous reviewer did not like that, but I suspected it was
>because of inhomogeneity in time).
>So in Mike M’s original hockeystick, greenhouse warming is only
detectable
>by 1980, with
>relatively small scaling, and there is a very substantial residual
>variability not explained.
>With lots of other records, the fit was much better, with the residual
>looking much safer etc.
>Of course Hans’ warped hockeystick is just fake, so it would be an
>academic exercise.
>Myles, to answer your question – I still would be a bit worried to do
this
>with a Mike -type timeseries,
>but Hans’ real one isn;’t. Hans, could I have the pseudo hockeystick? I
>can make the warped one myself.
>
>- Myles and I have a paper in the works trying to constrain little ice
age
>to today temperature change.
>Its not very tighlty constrained by records, but with a total least
square
>fit one can give lower limits.
>So its not like we have no idea at all. Also, the simple records that
>average may be a bit safer to use,
>and loose less variance even with ordinary least square fit.
>
>Gabi
>
>
>Klaus Hasselmann wrote:
>
>>Hi IDAG lot and Hans,
>>Hans von Storch’s recent article in Science suggesting the noise level
in
>>Manne’s hockey stick had been strongly underestimated has created
quite a
>>stir in the media (comments in Nature, New York Times, Spiegel, etc).
>>Although Hans clearly stated that this did not affect the conclusion
that
>>the anthropogenic global warming signal can be detected today, this
was,
>>of course, downplayed or ignored in some of the media, and even more
so
>>by the inevitable global warming critics.
>>
>>Even assuming that Hans is correct and that Manne’s criticism that
Hans
>>overestimated the noise in the time series does not hold, I suggest
that
>>someone in IDAG (Gabi?) picks up where Hans stops and actually
computes
>>and quickly publishes in Science or Nature what the signal-to-noise
>>level really is, both in Manne’s original hockey stick and Hans’s
warped
>>hockey stick. I suspect that there’s not that much difference, as
Hans’s
>>warp seems to lie mainly in the super-century time scales.
>>
>>The analysis is very simple. On the most elementary level, one merely
>>computes the mean global warming over some period T, say 100 years, 50
>>years or 30 years, by averaging the growth rate over that period. One
>>then compares the ensemble of warming values computed over the set of
>>T-time segments of the observed time series, without the last 120
years,
>>with the signal found for the last T years. I suspect that the highest
>>signal-to-noise level would be found, as in our earlier analysis, in
the
>>last 30-40 year global warming perios.
>>
>>In a slightly more sophisticated analysis one could apply the optimal
>>signal detection filter by taking the Fourier transform of the
predicted
>>global warming signal, divide this by the noise in the 1000 year time
>>series spectrum, and then apply this optimized filter to the data.
>>
>>The whole thing could be done very quickly, and I am sure that Hans
would
>>cooperate.
>>Cheers
>>Klaus
>>Klaus Hasselmann
>>Max Planck Institute for Meteorology
>>Bundestrasse 55, 20146 Hamburg, Germany
>>Voice: +49-(0)40-41173-236
>>Fax: +49-(0)40-41173-250
>>Email: hasselmann@dkrz.de
>>URL: [3]http://www.mpimet.mpg.de/~hasselmann.klaus/
>
>–
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~