Jean S observed in comments to another thread that he was unable to replicate the claimed “significant” correlation for many, if not, most of the 27 Gergis “significant” proxies. See his comments here
Jean S had observed:
Steve, Roman, or somebody , what am I doing wrong here? I tried to check the screening correlations of Gergis et al, and I’m getting such low values for a few proxies that there is no way that those can pass any test. I understood from the text that they used correlation on period 1921-1990 after detrending (both the instrumental and proxies), and that the instrumental was the actual target series (and not the against individual grid series). Simple R-code and data here.
I’ve re-checked his results from scratch and can confirm them.
The following graphic shows t-values for a regression of each detrended “passing” proxy against the target instrumental series, all downloaded from original data. ( I got exactly the same correlations as Jean S.) These calculations calculate significance in the style of Santer et al 2008. (The AR1 coefficient of residuals of proxy~instrumental is calculated and the number of degrees of freedom adjusted. A two-sided 5% significance test (again in Santer style) has a t-value benchmark of about +-2 for series of the length of the calibration period. (It varies a little, but this is not material for the point here and t-values of +-2 are shown in the graphic.) Only six of the 27 proxies have t-values exceeding +-2.
The next figure plots the locations of the screened-out proxies on the Gergis location map. I’ve shown the “not used” proxies by a red plus sign. I’ve also placed a circle around the proxies passing a t-test as above. (In two cases, the relevant t-value was lower than 2 and thus there are eight circles.)
One of the underlying mysteries of Gergis-style analysis is one seemingly equivalent proxies can be “significant” while another isn’t. Unfortunately, these fundamental issues are never addressed in the “peer reviewed literature”.
Gergis et al 2012 had stated:
For predictor selection, both proxy climate and instrumental data were linearly detrended over the 1921–1990 period to avoid inflating the correlation coefficient due to the presence of the global warming signal present in the observed temperature record. Only records that were significantly ([pre]p<0.05[/pre]) correlated with the detrended instrumental target over the 1921–1990 period were selected for analysis.
Bolded values are significant as determined by a normal distribution white noise p-value, p<0.05.