Yes of course..Craig Loehle et al 2008. http://www.ncasi.org/publications/Detail.aspx?id=3025,
this is the first version the new version is unfortunately unavailable anymore.

love your name by the way.

By the way, how did they calculate anomalies for the Briffa data relative to 1961-1990 when the data were truncated in 1960? Did the do the anomalies with the post-1960 data (as if it was valid) and then chop it? Because they had to somehow make it come out right in those decades in order to stitch the instrumental data on it.

I don’t think that’s a problem either, Craig — It would be quite possible to truncate the Briffa data at 1960, and then calibrate what’s left to instrumental temperature that has a reference period of 1961-90. The resulting reconstruction would still be referenced to 1961-90, and would give plausible results if the temperature data were spliced onto it before smoothing.

The curves wouldn’t match up well, however (I think), if the full Briffa series were calibrated to instrumental temperature and then truncated back to 1960 before splicing. But of course, the correlation with temperature probably wouldn’t be significant if the post-1960 portion of the series were used for the calibration. It might even be negative! It was probably surely only by cherry-picking the early part of the series that it appeared to be a temperature proxy.

Another point about the 1961-1990 anomaly calculation: by picking a recent period to set all the series equal to 0 (as anomalies) you force them all to come together and match the instrumental record artificially.

I don’t think that’s right, Craig.

If there’s a constant in a regression, the regression residuals must sum to zero over the calibration period. This implies that if you regress a proxy on instrumental temperatures, the predicted values of the proxy must sum to the actual values of the proxy over the same period.

Then if you invert the regression to get predicted temperatures from the proxy, dividing by the same slope coefficient you multiplied the temperatures by to get predicted proxy values, the temperature errors will just be scaled versions of the proxy errors, and so the predicted temperatures must sum to the same value as the observed temperatures, again over the full calibration period.

However, there is no particular reason that either the regression residuals or the temperature errors would sum to zero over the reference period. The choice of reference period may have some subtle implications, but I don’t think this is one of them.

The same would be true if someone inappropriately directly regressed temperatures on the proxy.

Of course, splicing a common temperature series onto the end of a proxy reconstruction before smoothing will force the smoothed “reconstruction” to look like smoothed temperatures at the end.