Stagflation Davidson said something about global warming.
I now hand over to this column.
John Quiggin says this
"his is just a demonstration that Davidson’s idiot readers don’t understand the concept of statistical significance, and that Davidson (who knows the score) doesn’t care to enlighten them. The question at issue is “how much data do you need to be 95 per cent confident that the observed warming couldn’t have arisen by chance”. With annual data and a simple linear regression, the answer (given by Phil Jones) is that the data from 1995 to 2009 (when he was first asked) wasn’t enough, but the data from 1995 to 2012 is sufficient.
Davidson’s trick is that, if you allow for autocorrelation in the monthly data, you need a longer time period to get 95 per cent confidence. With the data since 1995, there’s a 20 per cent chance that the observed warming could have arisen by chance. Since we have satellite data going back to 1979, and surface data going back 100 years, that fact is irrelevant except to the morons who comment at Catallaxy, but Davidson is happy to play on the ignorance of his readers.
Of course, the real shame here goes to Richard Lindzen who originated this utterly dishonest line of argument. Lindzen has trashed a record of achievement that would have had him remembered as a significant contributor to science. Davidson has nothing to lose, and the applause of his idiot followers to gain."
then NickR says this
"I’d argue that Davidson’s work is far worse than that. He effectively imposes a structural break in 1995 but does so with reference to the data. Endogeneously placing a structural break makes the inference he performs with regular t statistics incorrect.
His argument here is that Phil Jones did it, so if you disagree explain why he is wrong. Incredible chutzpah really as Jones did it and drew the *correct* conclusion while Davidson is doing it to support the *incorrect* conclusion.
He also corrects for autocorrelation in two ways. Firstly with his AR(1) error (which is not necessary for the hypothesis he wishes to test and therefore a violation of Occam’s Razor) and secondly with Newy West standard errors. He is right to be concerned about the effect of autocorrelation on his inference, but to correct for it twice is to deliberately waste information. He then builds his case on their being insufficient information to establish a trend."
and then john Quiggin says this
"Obviously, there’s no reason to think anything changed in 1995, and a test of the full data set would, I’m sure, reject the hypothesis of a structural break. This is just a piece of trickery, playing on the gullibility and wishful thinking of the delusionist audience.
On yoour other point, I haven’t kept up with the latest in time-series econometrics, but I think you’re right. If you are going to use Newey-West standard errors to correct for autocorrelation, you shouldn’t also impose an AR(1) error structure."
The only thing I will say is that it is strange ( highly euphemistic) to state bit it is a Phil Jones test when it was Lindzen
say no more.
Sinclair, Goebbels started out as a super smart propagandist but ended up by 1938 believing his own bulldust!
There must be a reason why Katesy is as mad as a hatter. Be afraid be very afraid.
Beware of the dark side!!!