Statistical analysis of IPCC AR5

[TheRobin]

So, apparently even the IPCCs own statisticians reject the model the other devisions use to claim that "significant" temperature increase.

All in all a great read, as it also explains the core problems of statistical models in general (which is a problem in a lot of fields of science today). Very enlightening. Enjoy

http://www.informath.org/AR5stat.pdf

[percentient]

Indeed, very fun to read, thanks. I love this kind of stuff. But the more I read it, the more it seems to me that he is a polemic and a nit-picker rather than a truth-seeker. At some point, after insulting enough professionals, it's more likely that he's the clueless one. Just you know statistically speaking.

 

His essential gripe is that IPCC et al. use the wrong words. In his mind "significant trend" should be reserved for cases where no reasonable stochastic process whatsoever could produce a time series as the one observed. In fact, not even unreasonable processes, like the difference-based one he used to draw the "stock price" random walk curves in the beginning (his ARIMA(3,1,0) model is just silly, temperatures don't work that way -- I'm pretty sure it allows them to approach literally any value given enough time). Instead, IPCC just say "look, the error bars clearly show that the temperatures have risen, and that the changes are down to external forcings or unkown multidecadal internal processes, instead of being produced by simple short-term processes that would be visible in the time series itself". They do mix up some words and concepts in the AR5 in my opinion, but it's not dramatic. The Breusch & Vahid article, also a subject of his criticism, make the opposite sin of being super rigorous to the point of allowing the ARIMA(3,1,0) type unphysical models.