Mann: "We actually eliminate records with negative correlations"; Briffa: "I too have expressed my concern to Phil (and Ray) over the logic that you leave all series you want in but just weight them according to some (sometimes low) correlation"

Email 3555

We actually eliminate records with negative correlations (this is mentioned breifly in the GRL article,), and we investigated a variety of weighting schemes to assure the basic robustness of the composite--but I certainly endorse your broader point here. Many of these records have some significant uncertainties or possible sources of bias, and this isn't the place to get into that.  [Michael Mann]
...
It was a majority decision to leave the Mann and Jones 2000-year series in the Figure 1 (as it was to remove the Briffa and Osborn tree-ring based one) , and the details of the logic used to derive the Mann and Jones series is to be found in the (cited) text of their paper. Signing on to this letter , in my mind. implies agreement with the text and not individual endorsement of all curves by each author. I too have expressed my concern to Phil (and Ray) over the logic that you leave all series you want in but just weight them according to some (sometimes low) correlation (in this case based on decadal values). I also believe some of the series that make up the Chinese record are dubious or obscure , but the same is true of other records Mann and Jones have used (e.g. how do you handle a series in New Zealand that has a -0.25 correlation?) . Further serious problems are still (see my and Tim's Science comment on the Mann 1999 paper) lurking with the correction applied to the Western US tree-ring PC amplitude series used (and shown in Figure 2). There are problems (and limitations ) with ALL series used. At this stage , singling out individual records for added (and unavoidably cursory added description) is not practical. [Keith Briffa]