I made this point to William Connolley. He told me that

“….it wasn’t all green! just like now, it was mostly covered in ice!”

It seems that the Hockey Team believes that the Vikings colonized a glacier.

For the Vikings to colonize a glacier which was located several days’ sail from their nearest permanent settlement would be somewhat analogous to 21st century man establishing a colony on the moon. You would need to bring in most building materials and at least some of the food to keep the colony going. The logistics would have been daunting (without the MWP).

Regarding next broadcast of LIA on the History Channel, is at 17:00 on 26-NOV (Saturday).
http://www.historychannel.com/global/listings/series_showcase.jsp?EGrpType=Series&Id=16021920&NetwCode=THC

NOTE: As part of the advertising teaser, the claim is it (the LIA) “decimated the Spanish Armada!” Oh well, looks like hyping the local WEATHER event impact of a changing global CLIMATE can happen on both ends of the temp scale.

Will we soon see a thread on how the show actually supports MBH’98 and ”
98 on RC.org?

Haven’t seen the show, but selective sea sediment core choices could easily been made. Dr. Mann has already showed a preference for Dr. Keigwin’s Pickart’99 Newfoundland core (which statistically cools the MWP, and warms the LIA in multi-proxy), to Keigwin’s Sargasso Sea cores which show a pronounced MWP and to a lesser extent LIA. This was discussed in http://www.climateaudit.org/?p=145 — as was pointed out there, Monnin included Keigwin’s Sargassos ’96, and of course showed a MWP and LIA more in line which pre-HS thinking.

Is there a simple metric that indicates how much more likely ‘model A’ (eg an ARMA) is to be true than say ‘model B’ ( an AR1)?

to my surprise, btw, the article also referred to them as “bandpass” filters, which seems odd to me. i commented in another thread about this not being the case, but apparently i was mistaken (at least, mistaken that they are viewed in such a manner). i suppose the end-stage coefficients are indeed from a selected band, but individually they behave as lowpass and highpass pairs.

IMO, the primary reason for using any sort of wavelet analysis is that there are less people in the world that have even heard of it. the time-frequency characteristic is nifty, but not that much more nifty than a spectrogram. also, you’d still have the problem of weighting depending upon how you stitched the various proxies together.

mark

PS: i watched part of the LIA show, too. interesting things happened in an era that apparently didn’t exist according to bristlecone pine trees.

The long runs are going to have a lot more squared variation than the noise type behavior within the run, and thus the first principal component should be expected to pick up the period of those runs as opposed to the shorter period sawtooth behavior within the long runs. The greater the AR(1) coefficient the more likely the first PC is to be dominated by the long periods. Given a set of K series, there will be one with considerably longer apparent period than the average for the particular AR coefficient. This presumably is the reason for the greater period in the PC1 of the 70 series versus the distribution of variance in the typical series.

There is an analogy in regression of PCs on PCs. Take two sets of NxK series of white noise, X’s and Y’s. Run regressions of the individual series on individual series and you get the expected rejection rates on the coefficients for the X’s. Same holds for the PC1 of the Y’s on PC1 of the Y’s. But when you throw in serial correlation, say AR(1) = 0.8, the individual series will reject at roughly a 35% rate instead of the 5% rate of the classical model. That is an example of the standard “spurious regression” result. But the PC1 of the Y’s on the PC1 of the X’s will reject at 50%. Further, the PC(K), the last PC, on PC(K) will reject at a 15%, much closer to the classical mode. (These results are for a 400×40 set of Y’s and X’s. The effect will increase/decrease with the increase/decrease of the number of columns.) This exacerbation of the spurious regression effect could be interpreted as saying the PC1 removes the big changes, which are the long runs and hence long periods, and the sequential PC’s continue to “filter” out the remaining longest runs until the last PC is left with something that is most like white noise. Of course the “filtering” should not be taken literally since all the principal components can do is pick axes of the transformation in the data space.

These characteristics are what make principal components (and factor analysis) useful, but it is really not a technique that should be used to make regressors and regressands or any thing else to be used in subsequent squared distance minimization or maximization routine.