Thursday, March 21, 2013

Rose gets stuck on statistical significance

Too many takedowns to count for David Rose who doesn't realize that short-term fluctuations in temperature tell you little about long term trends.  The latest case is that according to one computer model, the temperature sequence ending in 2012 is close to the bottom edge of the statistical uncertainty range, a point where there's only supposed to be a 5% chance that random variation produces a temperature below the modeled range. Rose thinks this means an end to warming.

The above link shows a broader set of models gives a wider uncertainty range.  And anyway, it's within the uncertainty range for the more sensitive model albeit near the lower edge.

The being near the edge is where our Rose gets stuck. A little over two years ago, Rose declared that global warming had halted since 1995. His proof - while measured temperature had risen since 1995, the amount of rise was only near the edge of being statistically significant:
Phil Jones replies: "The key statement here is 'not statistically significant'. It wasn't for these years at the 95% level, but it would have been at the 90% level. If you add the value of 0.52 in for 2010 and look at 1995 to 2010 then the warming is statistically significant at the 95% level." [What this means is that the warming trend for the past few years previously met a lower test of statistical significance. With addition of the results so far for 2010, it now means the higher test.]
So according to Rose, being near the edge may as well be proof if it's on the cold side, but means nothing at all if it's on the warm side.

Might also be worth noting that given decades of data, random variation will actually push the actual result outside the 5%-95% band at some points.