date: Mon, 23 Jun 2008 18:03:50 +0100
from: Steve Jones
subject: Significance testing results
to: Tim Osborn
Hi Tim,
As we discussed in our meeting last week, I've recreated my contour
comparison figures with significance testing, to show where differences
are significant at the 95% level. I've put the figures here:
http://www.squaregoldfish.co.uk/sekrett/figures.html
To calculate the significance, I used a simplified t-test algorithm that
can be used for scenarios where both data sets have the same number of
values. I found it on (dare I say it) Wikipedia
(http://en.wikipedia.org/wiki/Student%27s_t-test), but I did some quick
comparisons between that and the one in the book you gave me, and they
came out with the same results.
The slightly disappointing result of from the comparisons is that the
changes in the subtropical jet stream between 20cm3 and sresa1b (shown
in sresa1b_winter_comparison.eps) are not deemed significant, even
though the average wind is around 6m/s stronger in all the models. Since
this is the largest change seen between the different data sets, it's a
shame that it's not significant, and we therefore can't legitimately
infer anything from it in terms of changes to the NAM.
Is there anything I should do about this at this stage, or should I
simply state what's there, and not analyse it further due to the lack of
significance?
Steve.