I have just completed reading your most recent GRL paper (Schmutz et al., 2000) on NAO reconstructions in which you show that proxy-based NAO reconstructions are probably wanting. It is not possible to strongly defend my reconstruction at this time (indeed I was extremely cautious in my description of it with regards to over-fitting problems, etc.).
...Indeed, I did make some effort to "verify" my reconstruction against early instrumental records, with somewhat contradictory and potentially interesting results. Over the 1841-1873 period, my record correlates significantly with Stykkisholmer SLP (-0.456) and Oslo temperatures (0.323), but not Bermuda SLP (0.156) and Central England temperatures (0.211). The "appearance" of significant verification with only the more northerly instrumental records may be telling us something about differences in circulation and SSTs over the North Atlantic from what is now the case. This could affect the way in which the NAO affects climate jointly over North America and Europe. Of course, when I added some earlier observations (same stations) to the verification tests (Table 4 of my paper), the results weakened considerably. So, maybe this means that my NAO reconstruction is indeed poor. However, I must admit to having doubts about the quality of the early instrumental records despite the great efforts made to homogenize and correct them. This is especially the case with regards to low-frequency variability, but can also extend to individual values as well. I talked with Phil Jones about one suspect datum in the early portion of his extended NAO record that largely destroys any correlation with proxy-based NAO estimates (the sign of the instrumental index appears to be wrong to me). Yet, Phil is convinced that that datum is good and he may very well be right. Either way, more robust methods of association between series may be jusitified to guard anomalous values. [Ed Cook]