A lesson in statistics
Ken Parish supply the expected quibbles for my post on global warming. He illustrates some important statistical fallacies in the process. To support my case that the weather has got no cooler relative to the “the climatic extremes of 1998-2002″, I linked to the NOAA/National Climatic Data Center which showed that during the period since Daly’s post “January-July 2003, the global average land and ocean surface temperature was 0.54¡C (0.97¡F) above the long term mean, third warmest “.
Ken’s response is to go back to the underlying data and perform comparisons between Jan-July 2003 and the corresponding period in 2002, as well as month by month comparisons between 2002 and 2003 . Since the data set consists of deviations from long-term averages, it has no seasonal pattern, and there is, therefore, no justification for this procedure, or for focusing on 2002 when my post referred to the period 1998-2002 (the warmest five-year period on record,). It does, however, mean that the base period for his comparisons includes three of the four warmest months (in deviation terms) in the entire data set, all of which were in early 2002 (you can see them in the graph below). This is a prime example of ‘cherrypicking’.
Ken’s next move is to compare the individual months of June and July 2003 with the corresponding months in 2002, finding them to be cooler by 0.09 and 0.05 degrees respectively, and says “Thus the weather has got cooler since last year, albeit only slightly so far. ” In drawing this conclusion, he ignores the problem of statistical signifiance, which is acute when making pairwise comparisons for periods as a short as a single month. It’s easy to check that, even if you suppose that the data is an average of 1000 independent observations (far too many, given that short-term weather patterns at adjacent points are highly correlated) and that the standard deviation of monthly average temperatures at a given point is 1 degree, the difference found by Ken is statistically, as well as climatically insignificant.
All this might lead you to conclude that you can prove anything with statistics. But in fact, this is one case where you don’t need elaborate statistical analysis. The graph below (taken from NOAA) makes it clear that there is a warming trend that swamps the short-run cycles associated with sunspots and El Nino events.
To end on a positive note, I broadly agree with Ken’s assessment that
If the global average temperature falls back to well below the long-term average by 2006 that would suggest Daly may be correct. On the other hand, if it continues to fall but remains above the long-term average, that will clearly demonstrate the imprint of man-made global warming.
To be more specific, I predict that the average global temperature for 2006, as measured by NOAA, will be above the average for 1971-2000 (the baseline in the chart above) and I promise a retraction if this prediction is not correct.