To be more convincing I'd like to see what happens to their model if they used 5, 10, or 15 terms instead of 17. Plotting the quality of fit against the number of terms used would be revealing.
Also I'd like to see what happens to their projections if they used a shorter or longer range of dates. It must be possible to fit any data set given enough terms so the question is how sensitive the parameters of the fit are. Future data isn't available so they'd have to work with smaller segments from the past to pull this off. For example, they used about 44 years of data (1962 to 2006); how well would their model work if they used only 10 or 20 or 30 years instead of 44? Plotting quality of fit against the number of years used would be interesting. Or fairer yet, against the ratio of years of data in the past vs. years of prediction in the future. /tvb _______________________________________________ LEAPSECS mailing list [email protected] http://six.pairlist.net/mailman/listinfo/leapsecs
