Future prospects - Ice Ages and Interglacials: Measurements, Interpretation, and Models - page 345

the number of truly
independent observations (i.e., the effective sample size) may be just too small
for accurate reconstruction. Climate scientists have greatly underestimated the
uncertainty of proxy-based reconstructions and hence have been overconfident in
their models

The final point is particularly troublesome: ...

. Proxy based models with approximately the same amount of
reconstructive skill produce strikingly dissimilar historical back-casts: some of
these look like hockey sticks but most do not. Natural climate variability is not
well understood and is probably quite large. It is not clear that the proxies
currently used to predict temperature are even predictive of it at the scale of
several decades let alone over many centuries.''

...

Thirteen independent groups or individuals wrote commentaries on the
M&W paper. These commentaries demonstrated there is very little objectivity in
paleoclimatology as evidenced by the facts that the establishment alarmist
climatologists vigorously defended Mann et al., statisticians made abstruse mathe-
matical comments, and several climatologists exterior to the paleoclimatological
cabal indicated support for M&W.
One aspect of this controversy is the use of principal component analysis
(PCA). Before applying PCA one starts with a set of data from various proxies at
various locations over various time periods. If one adds these up and apportions
them equal weight, one obtains mainly mush—a smear of sparse data with no
apparent direction or structure. Then, PCA is applied. While one might naively
treat all proxies equally, PCA assigns weights to the various proxies on the basis
that those proxies with the least tendency toward a trend are given low weight and
those proxies with the greatest tendency toward a trend are given greater weight.
As M&M and Wegman showed, MBH gave some proxies hundreds of times the
weight of other proxies in the extreme case. The dataset was very sparse to begin
with, and PCA further reduces the dimensionality of the dataset by placing a
microscopic focus on those few proxies that demonstrate a strong trend, some of
which were suspect tree ring proxies. How can a weak sparse dataset be improved
by throwing out most of the data? Statisticians might respond by saying they have
identified the proxies that generate the trend for the whole set but, considering the
uncertainty and unreliability of all proxies, this seems like a very biased counter-
productive approach. PCA gives climatologists and statisticians fodder to play
with, but the whole process seems to add up to GIGO. In short, this writer thinks
the use of PCA as a method in this application is highly suspect.
M&W's rebutted the various commentaries by the 13 authors on their paper
in some length. M&W concluded by characterizing the assumptions made by the
hockey stickers (linearity, stationarity, data quality, etc.) as ''questionable, perhaps
even indefensible''. They also said: ''we reiterate our conclusion that 'climate
scientists have greatly underestimated the uncertainty of proxy-based reconstruc-
tions and hence have been overconfident in their models'.'' They closed with:
''Finally, and perhaps most importantly, the NRC assumptions of linearity and
stationarity outlined in our paper are likely untenable and we agree with Berliner
in calling them into question.''