I had previously recommended the blog, Science Of Doom, in my links. There is a lot of careful science there, especially in its exposition of atmospheric radiation and some of the subtleties of interpreting common data, such as those pertaining to Milankovitch cycles. However, despite the care, there is also a tolerance for what I professional consider to be unsubstantiated claims in the comments, and excessive appeals to authority. In particular, if extraordinary claims are made, I would prefer to see links to concrete and reproducible demonstrations of the claims, perferably with the code and data used, rather than links to papers which are devoid of that, and are accepted because some famous person is saying so.

Recently, the author has been pursuing an “it’s all about the chaos” theme in his investigations, something to which I greatly object, because I find it fails the most rudimentary of scientific criteria, falsifiability. When I presented that, I was essentially told it’s up to me to prove that the claim is not true, and that’s not at all how science works. And, when I asked for specific codes and demonstrations of some of the claims there, I was told that if I wanted to see those, I should reproduce them myself from the publications. Naturally, that’s a fools errand, for if I were to do so and find a discrepancy, the claimants could argue I misinterpreted how their results were to be implemented. This is why code and data need to be made public.

Moreover, there is no notion there at all of including a kind of Akaike information criterion or an information criterion derived from Bayesian perspectives in assessing their various models. Further, despite continued prodding, the assembled community completely ignores the critiques of Bayesians of frequentist hypothesis testing methods. If such a technical community does, there is, in my opinion, no good reason to continue discourse with them, since they are being inconsistent. Worse, this community systematically ignores complaints about matters like ignoring false discovery rate.

Thus, with regret, I can no longer in good conscience recommend them as a technical resource.

Instead, welcome to two new resources: QuantPalaeo, and Rabett Run.