He acknowledges that the Chow test is appropriate, but then wrongly says that I believe the break model, when it is quite clear that the break analysis is used in hypothesis testing mode, not modelling mode.  

The specific claims we made were: Quirk (that there was a break in 1976 – affirmative), and that their was a break in the continuous upward trend in temperatures (contradicting Easterling’s assertion).  The fact that T had to factor out ocean indicators in his recent paper, in order to recover an underlying continuous upward trend, seems to demonstrate that Easterlings’s assertion was statistically incorrect, and so T has had to try to fix it.  

http://tamino.wordpress.com/2012/01/22/step-3/

Thoughts?

What is relevant is whether the method used is sufficiently robust to represent actual regime shifts.  As the methods in the links find shifts in random data, they show the methods they have chosen are inadequate.  Largely this is because the have chosen a model that forces step changes.  We don’t do that.  The model is free to fit a slope, or two slopes.

Showing lousy, inappropriate methods find spurious breaks, as these guys do, does not refute a more appropriate method that is less prone to errors.  

The purpose of our paper is to examine specific claims, such as Easterling’s claim there is no possibility of a regime shift, and that temperatures are continuing to increase at the same rate.  Our analysis shows there is evidence of a break, a regime shift, for whatever reason, and so the claim that temperature increases have paused does have statistical basis.  

But you need a Chow test broken regressions at least and even more rigor to really address the question of statistically significant breaks.  Not just fit steps to the data.

http://tamino.wordpress.com/2012/01/09/steps/

http://tamino.wordpress.com/2012/01/10/step-2/

http://www.skepticalscience.com/its-a-climate-shift-step-function-caused-by-natural-cycles.htm

У вас в RSS картинки не показывает……