Good work, Hu — thanks!

Thompson seems to be in the “ignore critics & hope they go away” camp. Which reminds me, I should bump my letter re the Gore video again.

These things take time & patience, but science is (in the long term) self-correcting. Then again, in the long term, we’ll all be dead

Cheers — Pete Tillman

Figure 5a of the PNAS article shows 5-year averages of dO18 ratios for 7 ice cores, including 4 Himalayan and 3 Andean sites, for 1600-1985 or later. Figure 6 shows composite dO18 z-scores for the cores in the two regions, along with a “tropical” composite series, decadally from year 0 to 2000. Data for the two figures are in Data Sets 2 and 3 on the PNAS website. (The archived data are inconveniently in non-machine-readable PDF format, but Steve has punched in Data Set 2 and I punched in the requisite part of Data set 3.)

If the Himalayan composite were any fixed linear combination of the underlying four series, with equal or unequal weights before or after taking z-scores, a regression of the composite series on a constant plus decadal averages of its four component series should be an exact fit to within the .005 rounding error in the two-digit composite series. This should be true for the 37 decades for which complete data are available, as well as for any subperiod of 5 or more periods. (5 are required in order to identify the constant and 4 slope coefficients.)

However, when the full sample is used, the standard error of the regression is 0.29, i.e. 58 times higher than the permissible value of .005. Guliya and Dunde are insignificantly different from 0 (t = 0.41, 1.14), so that only Puruogangri and Dasuopu are making any significant contribution. When the full sample is divided into 4 subperiods of 10, 9, 9, and 9 decades, even Puruogangri is only significant in the last subperiod. In the third subperiod (1800-1889), none of the coefficients is significant, and an F test cannot reject the hypothesis that all are zero (F(4,4) = 1.313; p = .399).

It is therefore safe to say that the Himalayan “composite” series is inconsistent with the any linear composition of the four supposed component series. Some kind of non-linear aggregation might give an inexact fit, but nothing this bad. Thompson et al give no indication of how the aggregation was performed, but given the simplicity (and naivete) of the composition of the two regional indices into the summary “tropical” series, one would assume that an arithmetic average was taken.

The Andean composite fares somewhat better, with a full sample standard error of 0.10, which is only 20 times the permissible rounding error. Despite the impermissibly large residuals, the coefficients are well defined and highly significant. (Quelccaya and Huascaran get approximately equal weights, while Sajama gets about half that value.) The four subperiods have higher coefficient standard errors, as is to be expected, but exhibit the same basic pattern for the three coefficients. There is some inconsistency here that ought not to be present, but the problem is not as egregious as with the Himalayan numbers.

I’ve e-mailed Thompson and such of his PNAS co-authors as I have been able to locate about this, but with no response so far, either confirming or refuting my calculations. I’ll write PNAS with the specifics if I don’t hear from them soon. In my view, PNAS should demand a correction of one or both of the figures and corresponding data set or, if no such correction is forthcoming, simply retract the article and refer it to the JIR (the Journal of Irreproducible Results, a venerable science humor magazine).

MikeinAppalachia said (February 4th, 2008 at 2:01 pm)

I’m an OSU alum and have attended a few faculty (SO is Prof in tOSU’s MEdu program) receptions where Dr. Thompson (and wife) were present. He receives “rock star” attention from the administration, who are not going to do anything that would risk funding or reflect badly on the Polar Studies program. Have to admit, for a fellow hillbilly with Marshall undergrad degree, Lonnie’s “done good”.

If I was the Admin, I’d be afraid that the attention Thopmson recieves for lack of archiving will start to attach to all professors at OSU. As far as I know, ALL OSU profs have problems archiving their data. It’s up to the Admin to prove they don’t.

And if they can prove that other profs archive, they’d have to explain why Thompson doesn’t (Catch-22 is alive at OSU).

An item from Wikipedia (that I contributed): During a storm atop
Huascaran, Peru’s highest peak (>6,000 m.), gale-force winds pushed
Thompson’s tent (with him inside) towards a precipice. Thompson saved
himself with a self-arrest, jamming his ice ax through the floor of
his tent. (Source: Bowen, Thin Ice)

Why does this make me think of “The Grinch Who Stole Christmas”?