On Wed, 05 Aug 2009 04:16:30 -0700, Roger Coppock wrote:
RSS Satellite Data Clearly Show A Warming Global Climate
[...]
Coefficients:
Estimate Std. Error t value Pr(|t|)
(Intercept) -30.38442 2.02506 -15.0 2e-16
YEARMON 0.01528 0.00102 15.1 3.8e-40
Residual standard error: 0.172 on 365 degrees of freedom
Multiple R-Squared: 0.38
F-statistic: 226 on 1 and 365 DF, p-value: 3.8e-40
Why those much discussed trendlines are probably insignificant
--------------------------------------------------------------
Data sources:
-------------
UAH:
http://vortex.nsstc.uah.edu/public/m.../tltglhmam_5.2
RSS:
ftp://ftp.ssmi.com/msu/monthly_time_...cean_v03_2.txt
Regression:
-----------
UAH y = 0.01268 * x + -25.21167 ; rē = 0.26293 ; sd = 0.18073 ;
n1 = 68.08 ; n2 = 95.20
RSS y = 0.01528 * x + -30.38425 ; rē = 0.38279 ; sd = 0.17127 ;
n1 = 71.66 ; n2 = 94.82
Variables:
----------
x: the month (as a number, 2009.0 is January 2009, 2009.5 is July 2009,
half a year later)
y: expected value for month x (i.e. what the trendline says)
rē: correlation between time line and observed values (1 = perfect, 0 =
no correlation at all)
sd: standard deviation (calculated from the difference between observed
and expected values)
n1: percentage of values within plus/minus one standard deviation
n2: same for two standard deviations
The point is:
-------------
In theory, n2 is 95 percent, if the observed values are completely random
(no correlation, no causation, no real connection).
As the n2 values are in fact still near 95 percent, we still are on the
edge of pure noise. Or, maybe, the whole idea of a linear trend is just
not what's going on in the real world.
Joern