http://notrickszone.com/2013/04/15/back-to-the-penalty-box-chinese-paleo-climatolgists-slap-down-high-sticking-michael-mann/

It just shows that these threads repay a revisit; I missed this the first time round.

A very compelling graphic demonstration, thanks.

I understand validation: you take only half of the information for fitting, and judge the results (I guess some sort of weighting factors) with the other half.

Now temperatures are well measured only in the latest decades, I guess. Fitting on 1850-1949 is actually using comparatively bad temperature data, compared to sattelites that we have now.

Would it not be more appropriate to fit the proxies on modern data and validate them on older ones? Moreover, one could compare the differences in the resulting weights by fitting with modern (e.g. 1940+) vs. old (1850-1949) data and discuss the robustness of the fitting.

One step further – make the fitting on stepwise more and more data (e.g. first 1850-1900, then 1875-1925, then 1900-1950 etc…), and see how the weighting factors change. If it is robust, they should converge and stay pretty constant. Otherwise they will move “randomly”.

Of course, that is less Auditing, and actually more something that the researcher may do…

OK, but that is part of 3, not 2.5, just to be precise.

And thank you for the interesting additive note and that you are going to look at it. snip

Steve – different orientations for the same series is exotic, but is in all steps because the Yemen speleothem is affected. See the change from red-to-blue. My surmise is that the switch of this one series has a noticeable contribution – perhaps the major contribution – to the latem and earlym reconstruction difference and thus despite being an exotic circumstance impacts results. it’s on my radar but not in the near term.

1. I’m sorta uneasy with EIV in general because of the global climate field. Mann talks about doing EIV with local proxy matches and hemispheric ones, but I kinda didn’t read to see that he tested everything…and I don’t think he really reports in detail. In general, my concern with EIV is similar to the concern with MBH98, that the regressions may make us prone to fishing and data mining. Intuitively, I think that some sort of direct area method or select proxy method is more likely to give us a higher quality Bayesian guess. However, it’s probably good that SOMEONE is trying these very complicated and more tenuous approaches. So if we didn’t have Mike, we would have to invent him. Just that we should take EIV* with a grain of salt.

2. The issue of flipping recons is just another way in which EIV is more fishing for signal, versus having a strong physical argument. So when we evaluate EIV, we need to keep that in mind. Doesn’t mean no one should try very complicated methods. Just we should keep the results in mind with that concern. And to the extent that we can somehow (I don’t know how) figure out if the aphysical math-heavy approach is better or the more conservative approach is, we should do so.

- snip – [SM - meandering complaining ]

3. The different periods, different signs is even more of a concern physically, but is basically similar in concept to the issue discussed above.

4. I guess if you feed enough grist into the mill, you should not be surprised that some of it gets turned upside down or even upside down in different periods. The question comes if you can just live with that and it evens out, or if you should set up additional screens for it. I don’t know.

*I’m not even a statistician, so when people use terms like CPS or EIV, I’m a little in the dark. Might as well call them red and green. I think perhaps the bigger issue is not the kind of regression, but that one uses the global climate field and one uses local. (Yes, he does talk about running local climate WITH EIV, but I don’t get the sense this is what he did for most of the actual repeorted results.)

Steve: This is a blog. I’m more interested in the process of how things work and analysis of this methodology is a work-in-progress. There are many other sites where you can find short summaries, if that’s what interests you.