Climate Audit

In my post on Tornetrask showing the impact of the Briffa-Jones "adjustment", I didn’t comment on confidence intervals.

Most of my thoughts on this can be deduced merely by framing the question. If you look at the "adjusted" diagram, in the bottom panel, you will see that the differences between the adjusted and unadjusted versions after fitting to modern instrumental record will be negligible. (You can’t tell this from the published Figure 7 version which makes it look like the impact of the adjustment is on the modern period, rather than the prior period.) But once you re-normalize or do regression fits, the impact is the difference between the bottom panel versions.

The way that Mann et al. (and other members of the Hockey Team) do confidence intercal estimations is to calculate the residuals in the calibration period, either assume them to be normal or, if you really want to look statistical e.g. Mann, do a couple of hokey tests and declare them to be normal, then calculate 2 standard deviations and presto, you have a confidence interval.

Ponder the impact of this method on the Briffa-Jones Tornetrask reconstruction. You take the "adjusted" reconstruction, do all of the above and you get "confidence intervals". It looks very impressive to the laity. Of course, the unadjusted version would have almost identical confidence intervals. I suspect that the unadjusted version would be outside the "confidence intervals" of the adjusted version and vice versa. I’ve been thinking out loud here. I’ll try to remember to do the calculations, since the results will probably be pretty interesting and put some of these "confidence interval" calculations in a nice perspective.

There is a knock-on effect into the multiproxy studies. Jones et al. [1998] has only 3 proxies in the critical 11th century (the Viking period of the MWP) – Tornetrask is one of them. (Polar Urals, which I’ll get to, another Briffa creation, is just as bad.)